The advent of AI marks a new era in terms of how digital infrastructure is built, connected, and scaled. Each new generation of AI model demands exponentially greater GPU power, storage capacity, and interconnectivity.

As these workloads rise in complexity, the network responsible for moving data between GPU clusters and across data centers, also referred to as “scale-across”, faces mounting pressure.

Traditionally seen as a straightforward background utility, the transport layer is fast emerging as a strategic foundation for AI-driven infrastructures.

It is evolving from simply moving packets to intelligently orchestrating massive data flows with deterministic performance, low-latency, and seamless scalability.

AI-optimized data centers, built for training and deploying large-scale models, require dense GPU fabrics, vast storage, and, most importantly, high-bandwidth, ultra-low latency connectivity. As workloads increase, so too does the critical importance of Data Centre Interconnect (DCI) solutions.

The transport layer is no longer just a dumb pipe, but the backbone of the AI era, enabling real-time updates of AI LLMs (Large Language Models), continuous learning of those models, and the efficient movement of intelligence across a global digital landscape.

The demands placed on traditional networks by AI

AI training and inference pipelines generate overwhelming volumes of data. One large language model alone requires thousands of GPUs operating simultaneously, continually exchanging parameters, gradients, and checkpoints.

The outcome is a relentless demand for ultra-high-bandwidth, low-latency connectivity between clusters, whether across campus environments or regions, or out on the edge.

However, the traditional transport architectures developed to facilitate predictable enterprise or content delivery traffic buckle under the weight of emerging patterns.

East-west traffic has surged, bandwidth per node has soared, and latency requirements have tightened to microsecond precision. Simultaneously, operators face mounting pressure to control power and space consumption at the metro and edge layers, primarily because AI workloads are currently most ubiquitous here.

Scaling optical systems is clearly inadequate. Therefore, we must look at re-engineering the entire transport layer to support performance, agility, and efficiency.

What does it take to design an AI-first transport layer?

Next-generation transport layers need to evolve beyond moving bits quickly and adapt to the fluctuating demands of AI workloads. In practice, this means delivering ultra-high capacity and consistently low latency, while enabling the network to reconfigure itself as AI tasks switch between training and inference. This ensures that the transport network is transformed into an ecosystem as flexible and intelligent as the workloads it supports.

In addition to bandwidth scaling to meet demands, latency determinism is equally important. Distributed AI training depends on precise synchronization across thousands of GPUs, given that even marginal timing variations can stymie performance. Transport systems must therefore ensure reliable and constant optical path behavior, beyond simply average low delay.

Automation and energy efficiency are also critical to creating an AI-first transport layer. When power becomes a key enabler of data center growth, an AI-ready transport layer needs to sense, adapt, and optimize in real time, allocating capacity where it’s most needed, and shutting down redundant channels, seamlessly integrating with orchestration layers above.

Engineering architectures of the future

Today, overcoming the challenge of developing an AI-first architecture requires a new approach. The next generation of transport systems must be built on coherent optical technology capable of scaling above 400G and 800G. These advances allow operators to extract maximum capacity from existing fiber infrastructure while maintaining the low-latency performance which is essential to distributed AI training and inference workloads.

The evolution of software-defined optical control is equally important. Embedding intelligence into the transport layer facilitates real-time telemetry, closed-loop automation, and predictive optimization. Networks can monitor performance, anticipate congestion before it occurs, and dynamically re-route traffic to preserve stability and efficiency. Effectively, the optical layer evolves to be self-aware, adaptive, and capable of responding to changing workload demands.

IP-optical convergence is another area of advancement, unifying packet and optical transport under a unified control and management framework. This convergence reduces latency, rationalizes operations, and accelerates service provisioning. The advantages for hyperscale DCI environments include fewer network elements, reduced complexity and a more deterministic performance profile.

At the metro and edge levels, where space and power constraints are most stringent, compact modular DCI systems are emerging as a critical component of the AI ecosystem. Such platforms introduce high-capacity optical connectivity closer to compute resources, enabling real-time inference, analytics, and automation at the edge. Augmenting optical performance beyond the hyperscale core helps operators support increasingly distributed AI workloads and shorten the distance between data-generation and decision-making.

Such transformations are also supported by greater automation and more open orchestration frameworks. Modern transport networks are evolving into API-rich, software-driven environments that allow telemetry data from the optical layer to feed directly into higher-level AI and cloud management systems. This creates a feedback loop between application and infrastructure that, in practice, enables an adaptive network: one that learns, optimizes, and responds dynamically to workload patterns in real time.

5G + AI: the formula for success

The convergence of AI and 5G represents a new milestone in network evolution. As operators accelerate the deployment of 5G Standalone cores and distributed edge computing, the demand for high-bandwidth, deterministic transport now extends beyond data centers to cell sites, aggregation hubs, and metro edges.

AI is increasingly used to automate and optimize 5G operations, from traffic prediction and spectrum management to self-healing and real-time orchestration. These capabilities rely on the same transport tenets that power AI training: massive data movement, ultra-low latency, and intelligent routing.

As AI and 5G ecosystems grow in sophistication, they’re forming a unified, intelligent network infrastructure that seamlessly connects core clouds, AI clusters, and edge nodes. And today, optical transport systems that move terabits of training data between data centers will be able to support time-sensitive control traffic and AI-driven services at the edge in the near future.

Reimagining transport for the era of AI

AI is having a profound effect on all layers of digital infrastructure, but nowhere is this more evident than in transport. The optical layer is evolving from a passive conduit into an intelligent, agile system that anticipates demand, reconfigures dynamically, and continuously optimizes for performance.

This transformation is being propelled by advances in optics, automation, and converged architectures. These elements combined enable self-optimizing networks capable of sustaining AI and data-intensive services at a global scale.

Fundamentally, to meet the expectations of an AI-first world, transport networks must evolve into intelligent systems that are as dynamic and responsive as the workloads they carry. By rearchitecting the optical layer with agility, automation, and convergence at its core, the industry can lay the foundation for scalable, distributed intelligence across the global digital landscape.

Use the best business cloud storage to manage your data.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Amazon's gaming boss thinks ballooning hardware prices could push more hardcore gamers towards cloud streaming services
• He argued that the price of consoles increasing by 'six, seven times' would mean some gamers become 'priced out of the market'
• He thinks the RAM crisis, and rising prices, will not be 'going away anytime soon'

Amazon GM of gaming Jeff Gattis thinks the rising cost of consoles and PC components could push more hardcore gamers towards cloud streaming, via services such as Amazon's own Luna.

In a recent interview about the service, I asked whether Gattis anticipates cloud gaming becoming a more popular option for core players as traditional systems become much more expensive.

"Yeah. I wouldn't have said that 18 months ago, and I will say our business model is not really predicated on this this massive shift where everybody turns in their consoles and PCs and comes over to cloud" he said.

"But gosh, when you start to see the prices of the Steam Machine and the rumors of PlayStation and Xbox. As you get to $1,000, $1,500 hardware [...] you're going to see more people that are priced out of the market. They're going to consider other technologies".

Gattis, who previously led platform and subscriptions at Xbox, went on to share an anecdote to illustrate just how much prices have risen.

"I was telling somebody the other day that when I was back at Xbox a couple of years ago, we had our Series S console on discount [for a] holiday. I think it was about $199. To think in a span of two to three years we've gone from $200 to what's likely going to be six, seven times that price.

"I understand it's a it's a RAM crisis, it's worldwide, but I don't think that's going away anytime soon."

He was, however, careful to clarify that hardcore gamers moving to the cloud isn't a core part of Amazon's strategy: "It's not our strategy to accelerate that, but it's a nice tailwind if it does happen, and it's realistic to think that it might."

The comments come as Amazon rolls out a Fire TV update that gives Amazon Prime subscribers the ability to stream games like Fallout 4 or Fortnite from directly within Prime Video via Luna in a bid to get more eyes on the service.

Within a short space of time, AI has made the leap from experimental technology to everyday business tool. Enterprises across sectors are today deploying AI to take on routine, time-intensive tasks to allow their teams to focus on more strategic tasks or work that requires human judgement. According to McKinsey, most organizations are using AI in at least one business function.

Network management and monitoring is one of the fastest-growing areas of interest. The timing is no coincidence. Technologies like cloud computing are driving significant increases in both network traffic and complexity, making modern infrastructure far harder to manage than it was even a few years ago.

To stay ahead, IT teams are embracing AI and automation - but how widely is AI-based network monitoring used today and where is it headed next?

Relieving the pressure on IT staff

The task of monitoring network activity is complex and requires continual attention. It involves keeping infrastructure healthy, identifying faults, and reacting swiftly to unusual activity. These tasks were once handled manually but growing network scale and an increasingly hostile cyber threat environment have made traditional approaches hard to sustain.

IT professionals now find themselves spending a disproportionate amount of time on repetitive work such as firewall management, network provisioning, and routine monitoring.

AI addresses this directly by automating large portions of network supervision. Machine learning models can continuously process enormous volumes of network data, identifying anomalies such as traffic spikes, suspicious access patterns, or behaviors associated with known threats - and doing so in real time. This means teams can intervene before a problem becomes an outage or a security breach.

AI also sharpens focus. Rather than requiring specialists to wade through endless logs and alerts, AI-powered systems sift swiftly through these, filtering out the noise and drawing attention to only the issues that pose genuine concern. False positives are reduced, and teams can direct their energy toward real risks - including fast-moving threats that rule-based tools simply cannot keep pace with.

Scalability is another advantage. As demand fluctuates, AI monitoring systems can expand their coverage automatically, without the need for additional headcount. In environments where networks are growing larger and more dynamic by the day, this kind of elasticity is no longer a luxury.

Adoption today

AI-enabled network monitoring is already embedded across a wide range of industries. For many networking professionals, automation and AI are now considered core operational capabilities rather than nice-to-haves. A meaningful and growing share of network management activity - covering design, deployment, maintenance, and troubleshooting - is already handled through automated processes.

However, adoption does not automatically guarantee success. Many organizations are actively deploying AI features within their network tools, and some are even training models on their own IT and security data. Yet far fewer report achieving fully successful outcomes. The gap between using AI and genuinely benefiting from it reflects the real-world difficulty of moving beyond pilots to reliable, production-grade operations.

Two challenges consistently hold organizations back. The first is data quality - incomplete records, inconsistent formats, and poor documentation undermine AI model performance before it even gets started.

The second is skills. Many IT teams simply do not have the in-house expertise required to deploy, train, and manage AI-driven networking tools effectively, which slows progress and erodes confidence in outcomes.

Agentic AI: what's coming next

Despite these hurdles, the direction is clear. AI-based monitoring is a crucial component in networks management, and the next evolution, agentic AI, is already beginning to take shape.

Where conventional AI tools focus on detection and recommendations, agentic AI goes further. These systems can identify anomalies, diagnose root causes, predict capacity issues, and take corrective action - either autonomously or with minimal human sign-off. Rather than simply flagging problems, agentic AI is built to analyze, decide, and act, moving networks toward genuinely autonomous operations.

Consider a practical example. On a hospital campus, a staff member unknowingly connects an unauthorized access point to the network. A rogue SSID appears - a classic vector for man-in-the-middle attacks.

An agentic network management system detects the anomaly instantly, classifies the threat based on policy, and presents the administrator with a targeted remediation action: block the port or quarantine the MAC address. In sensitive environments, human sign-off is preserved by design. The AI does the analysis; the human makes the call. This is not a future concept - it is in production today.

Industry analysts expect agentic approaches to gain significant traction over the next few years, particularly in large and complex network environments. For most organizations, however, getting there will require a phased approach.

The immediate priority is deploying AI-based monitoring solutions that integrate cleanly with existing infrastructure, while ensuring teams are trained and confident in using them. As organizations build trust in their data quality and in the reliability of AI-generated insights, they can progressively introduce more autonomous capabilities.

The direction is clear: the organizations that treat AI-driven network management as a strategic investment today will operate faster, more resilient networks tomorrow - while those that wait will find the gap increasingly difficult to close.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Cybersecurity risks in water infrastructure have consequences that reach far beyond systems and networks.

Across the sector, the systems responsible for treatment and distribution are becoming more connected. Pumps, sensors and control environments that once operated in isolation are now linked to wider networks, often in the name of efficiency or modernization.

The problem is that these systems were never built with continuous exposure to cyber threats in mind and connecting them has introduced new attack surfaces that are difficult to maintain visibility of and control.

At the same time, many facilities have crept towards an increasingly blurred line between IT and operational technology (OT). Driven by incremental needs and very practical limits on systemic refresh, these systems have been added to and grown more complex over time rather than being designed securely from the ground up.

As a result, they become more tightly interconnected, with access stretching further across networks and systems than it should. Once an attacker breaches a system, it becomes far easier to move laterally across the network and get closer to critical infrastructure.

Meanwhile, cyber threats continue to encroach on the water sector, with attacks becoming increasingly frequent and their potential impact is hard to ignore. Disruptions to drinking water treatment or control systems can quickly escalate, interrupting supply or affecting water quality and, in turn, the communities that depend on them.

Structural challenges in securing water systems

Securing water infrastructure is made more difficult by the operational realities many providers face. Technology environments have expanded over time, often without dedicated cybersecurity resources growing at the same pace, making it harder to maintain consistent oversight across increasingly ageing and disparate systems.

Many organizations are also balancing modern hyper-connected digital management expectations with the ongoing operation of originally isolated, long-established systems. This places additional pressure on teams responsible for maintaining both resilience and day-to-day continuity.

Another persistent challenge is the divide between IT and operational technology (OT) teams. Because these environments have traditionally evolved separately, with different design approaches, responsibilities, priorities and expertise, they are not always closely aligned, which can slow decision-making and create gaps in visibility during an incident.

In smaller providers, cybersecurity responsibilities may sit with operational staff whose primary expertise lies in running facilities rather than managing cyber risk. Larger organizations may have more specialized cyber teams, but greater separation between functions still introduces coordination challenges and the risk of operational blind spots.

Connectivity without constraint

The growing use of cloud computing platforms and remote access tools has brought clear operational advantages to critical infrastructure like water systems. However, it has also reinforced a default position of keeping systems online at all times, often without a genuine operational imperative for continuous connectivity.

This “always-connected” approach can unnecessarily increase exposure, particularly as more assets become reachable across wider networks. Without clear control over the time windows when systems need to be accessible, organizations may be creating more risk than expected, certainly more than is required.

A stronger, resilient approach starts with recognizing that security grows by making connectivity intentional. Not every system needs to remain online continuously, and limiting unnecessary access significantly improves security outcomes.

This can be achieved by creating stronger separation between critical systems and the wider network, using controls that allow connections to be enabled only when required while maintaining essential operations. In this model, connectivity is actively managed to define resilience on demand.

Containment as a first line of defense

In the event of a vulnerability or compromise, response speed is critical, notably in environments where interconnected systems enable threats to spread rapidly across the network. Without effective connection controls in place, attackers can exploit this unconstrained accessibility to extend their reach before a full response is underway.

The ability to isolate systems in real time helps change this state. Segmenting critical parts of the network helps limit lateral movement, and deeply segmenting down to high criticality digital elements enables organizations to contain threats far more substantially and focus their efforts on speeding up the incident response.

Having this level of control helps limit the spread of disruption and supports a more structured response. It also creates clear, demonstrable evidence of how risk is being managed - something that’s becoming increasingly important as regulatory scrutiny and cyber insurance requirements become more demanding.

Moving to controlled access

The most resilient model possible with physical connection control treats access to critical systems as fully conditional. Rather than keeping them permanently online, connections can be limited to where, when and why they are required for business reasons. As risk levels change, this can be refined or tightened at will.

This lowers both risk and potential impact, minimizing loss, while preserving the flexibility required for day-to-day operations.

For water providers, deliberately managing connectivity and segmenting networks at an infrastructure level should be a priority for resilience. Clearer boundaries and reduced unnecessary access make it easier to protect infrastructure that plays a vital role in public safety.

We've reviewed, rated, and ranked the best ransomware protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

There's a pattern many organizations know well. A new technology arrives, adoption accelerates faster than governance can keep up, and a few years later, the finance team is staring at a spreadsheet, wondering how the bill got so large and who signed off on it.

The SaaS era exposed what happens when technology adoption outpaces financial oversight. And with agentic AI embedding itself into everyday workflows, organizations risk heading down a similar path.

Recent reporting on Amazon employees 'tokenmaxxing' - gaming internal AI metrics to inflate adoption figures, is an early signal of what that looks like in practice.

But the dynamic is not specific to Amazon. When AI usage isn't fully visible, and the incentives favor showing more activity rather than less, accountability tends to disappear quietly. It points towards a governance failure - and governance failures require structural solutions.

The opportunity is clear. The governance isn't.

The need for visibility

AWS's Banking on the Cloud 2026 report makes the strategic case for agentic AI in financial services clearly and compellingly. Cloud computing infrastructure and AI agents are positioned as the foundation of next-generation banking, which means faster decisions and more responsive customer experiences.

What the report focuses on is what AI can save. The other part of the equation is what AI itself costs to run at scale, and who's accountable for that.

The cost model for AI agents behaves differently from anything most enterprise finance teams have managed before. When you license a conventional software tool, there's usually a fixed price and a user count. The spending is visible even when it isn't well-controlled.

AI agents work differently as they run continuously, calling on external services and triggering actions across systems as they go. Each step consumes resources, and because agents operate autonomously, often handling tasks that would previously have required human judgment, that consumption can scale quickly and unpredictably. There's no contract line that captures it cleanly and no renewal date that forces a review.

Getting ahead of this requires visibility that most organizations are only now beginning to build.

The sprawl problem

The SaaS management challenge is a familiar one to most IT and finance leaders. Application estates grew faster than procurement could track them, governance lagged behind adoption, and many enterprises spent years rationalizing software stacks they never intended to build. It’s an ongoing problem that businesses are still managing, years down the line.

AI agent sprawl will likely develop differently, but the underlying problem is similar. The critical difference is pace.

A SaaS tool that gets deployed and forgotten sits there, quietly billing at a fixed rate. An AI agent generating outputs in real time is actively consuming resources from the moment it runs, and that financial exposure, left unmonitored, compounds in ways a forgotten software subscription simply doesn't.

Organizations that get the right comprehensive visibility in place early will be in a significantly stronger position than those treating cost governance as something to formalize later.

There's also a regulatory dimension that's coming into sharper focus, particularly in financial services. AI agents frequently depend on external model providers and third-party data sources. Each dependency introduces a potential point of failure - and in regulated industries, potential compliance exposure.

Regulators are paying attention, the EU AI Act's full obligations for financial services AI land in August 2026, and DORA audits are already underway, which means the question of who owns that chain of accountability will need a cleaner answer than most organizations currently have.

What good governance actually looks like

The encouraging part is that none of this requires building new disciplines from scratch. It requires applying familiar ones to a new context and doing it early.

The right starting point is understanding cost in relation to outcome. What does it actually cost to complete a task using an AI agent, and what is that task worth to the business? Answering it means connecting AI spending data to the broader picture of how technology is used and what it delivers, so that finance and engineering are working from the same information rather than talking past each other.

Controls also need to be built into the infrastructure rather than layered on top of it. As agent deployments grow, no team can realistically review individual workflows by hand. Policies that depend on someone remembering to check a dashboard aren't really policies; they're suggestions. When a budget review turns difficult or a regulator asks questions, suggestions don't hold up.

Most importantly, ownership needs to be established from day one. Which budget carries this deployment? Who reviews it when consumption shifts? Right now, many AI agents are being deployed by engineering teams without meaningful involvement from finance. That gap is entirely closable. Closing it before the bill arrives, rather than after, is where the real advantage gets built.

The organizations that navigate agentic AI well will be the ones that treat governance as part of the deployment decision rather than an afterthought to it. Cleaner accountability means faster decisions and AI investments that can actually be defended against the board or a regulator. Throughout the rest of this year and beyond, that is the key differentiator between organizations that scale AI confidently and those that are still untangling the bill.

We've reviewed, rated, and ranked the best personal finance software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Every enterprise I talk to right now has the same complaint dressed up in different language.

Their AI bills are climbing faster than anyone budgeted. Their model invoices look reasonable when viewed on their own.

But somewhere between boardroom approvals and the monthly cloud statements, money is disappearing in ways that nobody can fully explain.

Here’s the central issue that people struggle to understand: the most expensive part of AI isn't always the model itself. It's the infrastructure, orchestration, retries, idle GPUs, oversized context windows, and inefficient routing decisions that sit between a user's prompt and the final response.

This is something I call the hidden tax on AI adoption, and it's growing faster than most organizations realize.

Three numbers that should change how you think

Recently, at FinOps X in San Diego, a Goldman Sachs projection appeared on the main stage. Current enterprise token consumption globally sits at around six quadrillion tokens. The three-year projection: 120 quadrillion. That is not a rounding error… it is a 20x expansion, and it is arriving faster than the governance frameworks to manage it.

The same conference saw our launch of the Tokenomics Foundation (I’m fortunate to be a governing board member). It’s a vendor-neutral body inside the Linux Foundation dedicated specifically to the economics of AI token consumption.

The FinOps community, practitioners who have spent the better part of a decade building discipline around cloud computing spend, recognized that tokens represent a fundamentally different problem. Not a harder version of cloud cost optimization. A different one.

Here is why. When your organization runs a cloud workload, the cost is relatively legible. You provision compute, it runs, and you receive a bill. The relationship between action and expense is traceable. With AI tokens, that relationship fractures across three layers, and most organizations have visibility into only one.

Production → consumption → value: the three layers most teams ignore

The first layer is production. Before any AI model responds to any prompt, your infrastructure has to manufacture the tokens. GPU clusters, inference nodes, autoscaling policies, Kubernetes configurations: these are your token factories. Their efficiency, or lack of it, determines the base cost of everything that follows. A GPU node at 30% utilization is an expensive factory running at a third of capacity.

The second layer is consumption. This is where counterintuitive economics live, and it is what I spend most of my time thinking about. Two organizations can send identical prompts to different coding agents and arrive at radically different costs depending on how they manage context, caching, retries, routing, and the infrastructure supporting inference.

Prompt length, context window usage, caching strategy, and model routing decisions all compound. The common assumption that routing a task to a cheaper model always saves token cost, turns out to be wrong often enough to matter. A routing decision that invalidates a warm cache can make a "cheaper" model call more expensive than the frontier option it was meant to replace. These are second-order effects. They do not appear in standard dashboards but they appear in your monthly bill.

The third layer is value. This is the one that FinOps teams are comfortable with, and the one that matters least until you have the first two under control. Mapping token spend to business outcomes is a legitimate and important discipline. But you cannot govern at the value layer without instrumentation at the production and consumption layers. You are doing math with incomplete inputs.

Why 85% of your AI spend is probably misallocated

Here is a pattern I see consistently. Organizations treat frontier AI models, the most capable, most expensive models available, as their default infrastructure. Every task goes to the same model. Every prompt is constructed the same way. There is no routing logic, no tiering, no architectural distinction between work that genuinely requires the full capability of a frontier model and work that does not.

Based on my observations across organizations deploying AI at scale, roughly 15% of software development tasks actually require frontier model capabilities. The remaining 85% of routine coding, summarization, classification, and retrieval work can be handled by smaller, faster, and less expensive models, if you have the infrastructure to make those decisions intelligently and automatically.

The unlock is not picking better models manually. Manual model selection does not scale and degrades the developer experience by introducing friction at the moment when a developer needs to move fast. The unlock is building infrastructure that makes routing decisions for you: one that understands the task, routes it to the appropriate model tier, evaluates output quality, and escalates if needed. You specify the outcome you need. The system handles the economics of achieving it.

This is the direction the industry is moving, and the organizations that build this capability first will have a structural cost advantage that compounds over time.

The invoice arrives last. And you realize something has gone horribly wrong

There is a phrase I have started using with customers that captures the core problem: the invoice arrives last.

By the time you see the model provider bill, the cost decisions were made weeks ago in infrastructure configurations, autoscaling policies, and prompt architectures that nobody has reviewed since the initial deployment.

The retry logic runs silently when an upstream service slows down. The GPU nodes were reserved for peak traffic that never came. The agentic workflow, where a single user request fans out into dozens of model calls beneath it, each billed separately, none visible in the tool that generated the original request.

These costs do not live in the model invoice. They live in the infrastructure layer, in the consumption layer, and in the gap between how teams think their AI systems work and how they actually behave in production. You cannot govern what you cannot see. And right now, most teams are looking at one layer of a three-layer problem.

When AI goes from copilot to coworker, the stakes multiply

There is a shift underway that makes all of this more urgent. The AI deployments most enterprises built over the last two years were assistants, tools that accelerated individual work by handling the first draft, the next suggestion, and the boilerplate. A human remained in the loop at every consequential step. The economics were bound by how many people were using the tool and how often.

Autonomous agents change the economic profile entirely. When an AI system can receive a goal, build a plan, execute multi-step work, evaluate its own outputs, and iterate to completion without human intervention at each stage, you are no longer running an assistant. You are running something closer to a coworker, one that operates continuously, scales horizontally, and generates token consumption at rates that individual user interactions never approached.

The transition from copilot to coworker has already happened. And the governance implications are significantly more serious. A copilot with poor token economics costs you some efficiency. An autonomous agent with poor token economics runs that inefficiency at scale, continuously, without generating the natural friction that would cause a human user to pause or change approach. Infrastructure discipline and token optimization need to be in place before autonomous workloads scale rather than retrofitted afterward when the bill arrives.

The mandate for infrastructure teams

The right answer to this problem is not more dashboards. More visibility into a system you cannot control is just a more detailed invoice; it arrives with the same lag, and it changes nothing about the decisions that were already made upstream.

What infrastructure teams actually need is control that operates at the layer where costs are determined, not where they are reported. That means autonomous management of GPU and inference workloads, continuously rightsizing to match actual demand rather than peak assumptions, absorbing the bursty consumption patterns that agentic jobs produce, and moving compute capacity across providers when one environment becomes the bottleneck.

This is especially urgent now, because the transition from copilot to coworker does not give you a grace period to retrofit discipline. Autonomous agents do not pause. They do not get frustrated and choose a different approach. They run the inefficiency you built into them at scale, continuously, until something external stops them.

So the next phase of enterprise AI is defined by who can deploy models efficiently. As AI systems become more autonomous and token consumption accelerates, competitive advantage comes from understanding the full economics of AI, and not just the price of a model call.

Because by the time the invoice arrives, the decisions that shaped it have already been made.

Use the best business cloud storage to manage your data.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Windows 10 21H2 and earlier is losing support for all OneDrive updates
• 22H2 will still get feature and security updates until October 2028
• Affected users are being urged to upgrade to Windows 11

Microsoft has announced it will end OneDrive sync app updates on Windows 10 version 21H2 and earlier starting August 15, 2026.

Besides no longer offering new features, affected devices will also be incompatible with new bug fixes and security updates, marking the official end of OneDrive on those devices.

The app will continue to function after losing update support, but system changes and emerging bugs could render it unfixable with time.

Windows 10 is starting to lose its OneDrive sync support

From the August 2026 deadline onward, users may experience reduced functionality, compatibility problems and security risks, with the company instead pushing Windows 11 as a viable alternative to keep OneDrive functioning as expected.

Some later versions of the ongoing OS, including version 22H2, will continue to receive feature, reliability and security updates until October 10, 2028. 22H2 was the final version of Windows 10 – the next upgrade after that represents a brand new OS that brings with it some hardware limitations.

"This change aligns OneDrive support with the Windows lifecycle policy and helps Microsoft focus ongoing investments on supported operating systems," Microsoft wrote in a support article.

Users who are reluctant to upgrade OS could still access OneDrive via the browser if the app end up breaking due to incompatibility.

Rather than issuing advice to make OneDrive compatible on Windows 10, Microsoft is urging affected users to finally upgrade to Windows 11. Windows 10 still accounts for around 28% of all Windows installs (per Statcounter), compared with 70% for Windows 11.

• Amazon's AWS cloud services billing system went wild on Friday
• Customers saw bills in the trillions
• AWS apologized and is working on a fix

AWS, the cloud platform that serves millions who use Amazon's servers to run websites and businesses you use every day, just freaked out many of its customers with usage bills that literally went into the trillions.

"I just saw $1.5 trillion on my AWS bill and my soul left my body," wrote Bharath_uwu on X (formerly Twitter), and he wasn't alone.

The Guardian originally spotted the Tweet and multiple social posts relating to the seemingly widespread glitch.

AWS doesn't mail out invoices. Instead, customers logging into the Billing and Cost Management console spotted the billion- and trillion-dollar bills and expressed emotions ranging from concern to panic.

As Dan Harvey, who heads marketing at Learning Through Landscapes, told the Guardian, "I almost had a heart attack when I received an email alert from Amazon Web Services with the billing for our charity’s school grounds audit app."

Some tried to take the billing error in stride. Chinmay on X encouraged others to post their highest AWS bills. As of 10 hours ago, he was at $333B.

Others called out AWS for the error with something less than good humor. "I bet someone must have had a real heart attack. This shouldn't be legal...," wrote Mr Doob on X.

Amazon, to its credit, has been delivering multiple updates starting at around 1:30AM PDT on Friday, July 17, when they first spotted the error: "We are investigating issues with Cost Explorer reflecting inaccurate estimated billing data."

A few hours later, AWS reported it had found the culprit: "We have identified the root cause as an issue with unit pricing within the estimated billing computation subsystem, and we are working on a mitigation," they wrote on the AWS Status page.

However, hours later, AWS admitted, "Our efforts to backfill corrected estimated cost and usage data are still underway. We are progressing slower than anticipated."

The good news is that AWS is not asking anyone to pay these incorrect fees and that, while this has been a big frustration for services that run their websites and businesses on AWS, the cloud system has been working flawlessly, and there have been no reports of outages or performance issues on Friday (if you don't count the years shaved off freaked-out customers' lives).

AWS has even tried to make light of the mistake, posting this on X:

"Typo alert: Some customers saw quadrillion-dollar AWS billing estimates today. Slight miscalculation on our end (very slight ). We're fixing it now. No action needed on your end. Sorry for the confusion. Real question: what will you do with those trillions instead?"

Yes, everyone can all laugh now, though I'm not certain customers like Harvey, Bharath, and others are chuckling along with them.

Was your business affected? How big was your AWS Billing mistake? Let us know the "damage" in the comments below.

Across Europe, “tech sovereignty” is rising up the agenda. For business leaders, however, the implications are practical rather than political. At its core, sovereignty is about control, resilience and trust in the systems that underpin modern operations.

The European Commission’s recently announced technology sovereignty package reflects this shift. With proposals including the Chips Act 2.0, the Cloud and AI Development Act, an EU Open-Source Strategy and a Strategic Roadmap for Digitalisation and AI in Energy, it is intended to strengthen Europe’s digital independence and resilience.

That matters because IT infrastructure is now vital infrastructure. Cloud platforms, connectivity, AI systems and cyber security capabilities are becoming as essential to economic growth and national stability as energy and transport networks.

For years, digital transformation was driven by globalization, scale and efficiency, with organizations prioritizing rapid innovation, cost optimization and access to global technology ecosystems.

But cyber-attacks, regulatory divergence and geopolitical uncertainty have exposed a fundamental reality: efficiency without resilience creates fragility.

From efficiency to resilience

Today, organizations are focused not only on whether systems can withstand cyber-attacks, but whether they can keep operating if a provider, jurisdiction or supply chain becomes unavailable.

Protection and prevention remain essential. But resilience also depends on where systems are hosted, who controls critical infrastructure, how data moves across jurisdictions and whether essential services can be restored quickly during disruption.

Sovereignty is best understood as the ability to continue functioning with confidence when external conditions change. This is particularly relevant for organizations delivering critical services.

Why this matters for the UK

The UK faces many of the same pressures as Europe: rising cyber threats, tighter regulation and rapid AI adoption. But it also has real strengths, including a mature cyber security sector, world-leading professional services expertise, and a strong reputation for governance and innovation.

Those strengths give the UK a significant opportunity to position itself as a trusted partner for secure, resilient digital infrastructure. Realizing it, however, will require continued investment in infrastructure, skills and technology ecosystems.

The UK already has strong foundations. Within Vodafone Business, for example, security services for government and defense customers date back to 1989, underlining the long-term importance of trusted communications and secure operations.

Sovereignty must include AI

The sovereignty conversation is no longer limited to networks, cloud computing infrastructure or cyber security; it now extends to AI itself.

The UK government’s recent £400 million commitment to next-generation AI chips reinforces that ambition and signals a more deliberate push to build sovereign capability in the technologies that will underpin future competitiveness and resilience.

Government investment in sovereign computing capability and broader AI infrastructure also signals recognition that access to advanced computing resources is becoming a strategic national asset.

This matters because AI is rapidly becoming foundational infrastructure. Organizations are embedding it into business operations, cyber security programs, customer engagement and decision-making.

For businesses, sovereignty is not about limiting innovation. It is about ensuring critical capabilities can be developed, governed and accessed in ways that support long-term economic resilience and trust.

Connectivity as critical infrastructure

One of the most important and often overlooked aspects of sovereignty is connectivity. As organizations rely more on AI services, IoT devices and real-time data exchange, networks become the foundation of operational resilience.

If connectivity fails, everything built on it is affected, from customer services and supply chains to communications and core business operations.

That is why investment in secure, resilient, high-capacity networks is central to the sovereignty debate. Without trusted connectivity, digital sovereignty remains theoretical rather than practical.

The formation of VodafoneThree is a significant step in strengthening the UK’s digital backbone. With a commitment to invest £11 billion in next-generation connectivity and an ambition to deliver 99.96% population coverage by 2034, the UK is building infrastructure to support future growth and resilience.

As AI workloads, edge computing, hybrid working and data-intensive applications expand, resilient connectivity becomes a strategic national asset.

This is particularly important for organizations delivering essential services. Today, 77% of UK Blue Light services already run on Vodafone Business networks, underlining the growing importance of trusted connectivity in critical operations.

A strategic moment for the UK

Europe’s emerging sovereignty agenda should not be mistaken for digital isolation. Rather, it reflects a growing recognition that interdependence must be understood, managed and secured.

For security leaders, that means broadening the conversation beyond traditional threat protection to include critical dependencies, digital supply chain resilience and operational continuity. Cyber security is increasingly part of a wider discipline of digital resilience, where security, connectivity, infrastructure and governance converge.

By combining cyber security expertise, growing sovereign AI capabilities, regulatory strengths and continued investment in connectivity, the UK has an opportunity to establish itself as Europe’s trusted security ally.

In the next phase of digital transformation, success will not belong only to those with the most advanced technology, but to those with the most trusted, resilient and transparent foundations.

The sovereignty economy is already taking shape. The question is whether the UK chooses simply to participate in it, or to help define it.

We've reviewed and rated the best business cloud storage services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

In the summer of 1995, the future of computing briefly seemed to belong to Netscape.

Netscape went public that August, barely sixteen months after it had been founded. Its stock doubled on the first day. The company had no empire of hardware, no installed operating system, no grip on the office desktop. What it had was a window into a new world.

Open Navigator, type an address, and the Internet appeared with a little throb of electricity. The browser did not feel like an application. It felt like a passage out of Microsoft's world.

Microsoft noticed.

The battle that followed was called the browser wars, a phrase that makes it sound tidier than it was. Really, it was a fight over who had the right to stand between the user and the next era of computing.

Netscape believed the browser would make the operating system less important. Microsoft believed that anything capable of making Windows less important needed to become part of Windows, preferably yesterday.

By the end of the decade, the company that had introduced so many people to the Web was no longer the Web's gatekeeper. It had been out-distributed, out-bundled, and finally absorbed into a stranger corporate afterlife.

A familiar temptation

There is a familiar temptation now to ask which artificial intelligence company is "the Netscape of AI." The answer usually offered is OpenAI, and the comparison is not wrong. ChatGPT did for artificial intelligence what Navigator did for the Web. It turned a technical architecture into a public experience. It gave the future a text box.

But the more important question is this: Who is today's Microsoft? Who understands that the winner is often the company that owns the default?

Every platform shift begins with a miracle and ends with a map of choke points. The miracle is what users remember. The choke points are where the money goes.

The early Web was sold as an escape from gatekeepers. It created new ones. Search. Browsers. Marketplaces. Mobile operating systems. Cloud platforms. Social media. The AI era is being sold with the same democratic glow.

And yet the deeper stack is already hardening. It is made of chips, power contracts, data centers, model weights, enterprise identity, workflow data, cloud credits, procurement channels, and the small tyrannies of default settings. The romance is in the chatbot. The control is somewhere colder, louder, and much more expensive.

That was true in the nineties, too. The Web looked like a page. The winners understood it was a stack.

The danger to OpenAI

OpenAI is the obvious Netscape figure because it supplied the first mass-market revelation. Before ChatGPT, artificial intelligence was a research field, a back-office tool, or a phrase executives used when they meant analytics with a larger budget. After ChatGPT, it was something anyone could talk to.

People underrate the power of the first interface that makes a new technology feel inevitable. Netscape did not invent the Internet. It made the Internet feel reachable. OpenAI did not invent the transformer. It made the transformer feel conversational.

But Netscape's story is not a founder myth. It is a warning label.

Netscape had the user's excitement but not enough control over distribution. Microsoft had the operating system. It could place Internet Explorer where users already lived. It could make the browser free. It could turn a product category into a feature.

The lesson was simple: if your rival owns the layer beneath you, your brilliance may become their menu option.

OpenAI is better protected than Netscape was, but not safely protected. It has a huge brand, astonishing usage, and deep ties to Microsoft. It also has the curse of being expensive in a way software companies used to avoid. Each improvement requires compute, chips, talent, energy, and capital. The old software dream was scaling with almost no marginal cost. AI can't do that.

That is why OpenAI's partnership with Microsoft is both strength and vulnerability. Microsoft gives it cloud computing infrastructure, enterprise access, and capital. Microsoft also sits close enough to learn from it, package it, hedge around it, and sell AI into the places where people already do work.

If Netscape's problem was that Microsoft stood underneath it, OpenAI's problem is subtler: Microsoft stands underneath it, beside it, and increasingly in front of the customer.

Netscape had the dazzling demo. Yahoo had the traffic. AOL had the subscribers. None of that was enough. The durable winners were the companies that turned their software layers into a control point and a tollbooth.

Nvidia may be doing exactly that.

The rise of Nvidia

In the nineties, Intel was the metronome inside the personal-computer boom. Microsoft owned the software platform. Intel owned the pace of the machine.

Nvidia occupies a similar place in AI, but the analogy understates the ambition. Nvidia is not merely selling chips into a boom. It is selling the industrial base of the boom: GPUs, networking, software libraries, developer habits, and a vision of the data center as an AI factory.

Every major AI player is both Nvidia customer and Nvidia escape artist. Google has TPUs. Amazon has Trainium. Microsoft is developing its own silicon. Everyone wants an alternative. The problem is that wanting one does not create an ecosystem.

Nvidia's position today may be the purest example of moving up the stack from below. A chip company becomes a systems company. A systems company becomes a software company. A software company becomes a developer environment. A developer environment becomes a tax on ambition.

For now, Nvidia is the toll collector.

The AI market will not resolve into one winner. Platform shifts rarely do. The Web did not produce one winner. It produced layers of power. Microsoft kept the desktop. Google won search. Amazon won commerce and cloud. Apple won mobile hardware and the app economy. Meta won social attention.

AI will do the same.

Microsoft may become the default enterprise AI company, not because every Copilot is brilliant, but because Microsoft sits where work already happens. Nvidia may remain the dominant compute toll collector. Amazon will likely win much of the infrastructure layer. Google must reinvent search while defending it. Meta will use AI to extend attention. Apple may yet turn personal AI into a device-native experience.

The likely losers are the companies attached to the wrong layer.

The changing landscape of AI

In the nineties, AOL looked invincible because it owned access. Broadband made that access less special. Yahoo looked inevitable because it owned attention. Search made that attention less decisive. Netscape looked revolutionary because it owned the browser. Microsoft made the browser a dependency of the operating system.

In AI, the same demotions will happen. Some model companies will become features. Some application companies will become demos. Some incumbents will decorate old products with AI and call it transformation, which is the corporate version of putting a spoiler on a minivan.

The mistake, in every technological boom, is to confuse the moment of wonder with the arrangement of power that follows it.

The wonder is sincere. The arrangement is not.

The early Web made people feel as if they had slipped the old gatekeepers. Then came the search box, the app store, the marketplace, the cloud account, the login, the subscription, the default. Each solved a real inconvenience. Each left behind a narrower path.

AI will likely travel the same road, only faster and with a larger electricity bill. It will begin as a conversation and mature into an administrative system for human intention: what we ask, what we buy, what we write, whom we trust, which choices are shown, and which never quite appear.

The future does not usually arrive wearing chains. It arrives offering to save time.

We've tested, reviewed, and ranked the best business cloud storage services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

AI investment is accelerating across enterprises.

Budgets are increasing, boardrooms are asking hard questions, and the answer they’re getting back is token costs, prompt counts, and copilot deployment numbers.

Those aren't business metrics. They're activity logs.

Most organizations have settled on those numbers because they’re easy to track and show up well in reports. The problem is that those numbers measure the activity, not the results.

They show how much AI is being used, not whether the business is doing better since implementing it.

This is where token maxxing starts. It happens when organizations begin rewarding those who use AI the most, rather than what it actually delivers.

By optimizing for the wrong thing, we’re quietly setting AI programs up to fail.

When metrics become the mission

There’s a simple truth in business: people optimize for what gets measured, and we’re seeing that with AI right now. Understandably, because teams begin to generate more output when that is what leadership tends to see and reward.

Over time, the measure becomes the mission, and the focus on real outcomes starts to drift. The result is a surge in output that looks impressive on paper but has little to do with faster decisions or better products.

Teams with the highest usage or spend start to become the benchmark that others feel they need to match. But metrics like token spend should be evaluated the same way any other business investment is.

If we committed the same budget to a new sales tool or a go-to-market campaign, the first question from the board would be: how did revenue increase, and did it move the business forward in a measurable way?

High token consumption with no clear line to a business outcome is not a success story. Without that connection, it is just an additional cost. Meanwhile, the organizations that figure this out first will move faster with less spend — and that gap compounds.

AI did not create the problem; it exposed it

We’ve seen this pattern before. In the early days of cloud computing, companies moved fast and invested heavily without rethinking how work was structured. Costs went up, but the outcomes didn’t always follow.

That wasn't a cloud problem. And this isn't an AI problem. It never was.

The reality is that most companies already had a structural issue long before AI arrived. The work that matters most doesn’t live in a single system. It stretches across teams, tools and departments, held together by people stitching context together and filling in the gaps between systems. That’s always been the case. What AI has done is expose it, and this is where the problem runs deeper than just metrics.

AI makes individuals faster within the tools they already use. But when that work still sits inside disconnected systems, speed doesn’t fix the problem; it makes the gaps more visible. Teams move faster, but not necessarily together, and that’s why usage metrics can look so strong. More prompts, more output, more activity. On the surface, it looks like progress, but without a clear connection to outcomes, that activity can quickly become noise.

The problem isn’t just that we’re measuring the wrong thing. It’s also that most organizations don’t have a clear view of outcomes in the first place.

What good measurement looks like

Stop asking how much AI is being used and start asking what has changed because of it. A few shifts that will follow:

Output volume does not equal business value. Treat the two as interchangeable, and your AI program is already starting to drift.

Start with the outcome, not the tool. Instead of asking, “How many prompts did we run?” ask, “What did we achieve?”

Measure what crosses teams. If impact stays inside a single function, you have a useful tool — not a competitive advantage.

Here are a couple of examples: for engineering teams, this could mean focusing less on lines of code or pull requests and more on what actually reaches production and delivers value to customers. For marketing, it might be focusing on whether campaigns launch faster or land better.

Ultimately, AI that makes one person’s morning easier is a useful tool, but real value comes from AI that changes how the whole company gets work done. Without that, you end up with intelligent tools that operate in isolation. Helpful in moments but limited in impact.

Redefine what good looks like

Right now, most of the AI conversation is still focused on individual productivity — how much faster one person can write, code, analyze, or create. That matters. But it’s only part of the story.

The bigger opportunity sits at the organizational level and looks at how work actually moves across teams, systems and the company as a whole. AI can absolutely improve how businesses operate, but only if we are willing to ask harder questions about what we are actually measuring and why.

The companies that pull ahead won’t be the ones using AI the most. They’ll be the ones who changed what they measure when AI showed up. That’s the only metric that matters.

Use the best business cloud storage to manage your data.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The competition in artificial intelligence between the United States and China extends beyond a binary race for supremacy. The current landscape is more accurately characterized as "complementary competition."

While the United States and China compete, they occupy distinct positions within the global AI value chain, each possessing unique strengths that do not fully overlap.

This complementary competition is shaping the global AI industry through differentiated yet interconnected advantages.

The United States maintains leadership in foundational aspects of AI, including frontier model development, advanced semiconductors, cloud computing, basic research, and the broader developer ecosystem. Companies such as OpenAI, Anthropic, Google, and Nvidia significantly influence the architecture of the global AI industry.

For example, NVIDIA’s dominance in high-end GPUs provides the United States with a significant advantage in the computing layer. Additionally, leading universities, advanced research laboratories, and robust venture capital networks continue to supply AI development with talent, innovation, and funding.

According to Stanford’s 2025 AI Index Report, the United States attracted 109.1 billion dollars in private AI investment in 2024, and American institutions produced 40 of the world’s 50 most notable AI models.

This concentration of capital, research, and platform-building capacity reinforces the United States' position as a primary provider of the underlying AI architecture.

China's strengths in commercial contexts

In contrast, China is increasingly recognized for its strength in large-scale deployment of AI within industrial and commercial contexts. The Chinese industrial system is extensive, encompassing manufacturing, logistics, energy, automotive, electronics, and urban infrastructure. These sectors generate numerous real-world use cases for AI deployment, testing, and iteration.

China further benefits from tightly integrated supply chains and industrial clusters. For instance, in Shenzhen, hardware suppliers, software teams, factories, and logistics networks are highly interconnected, enabling rapid progression from concept to prototype to iteration. This integration allows China to embed AI into economic processes effectively, rather than confining its application to laboratory research or consumer chatbots.

This is why the U.S.-China AI competition is better understood as complementary rather than a zero-sum game. The U.S. provides many of the key enabling technologies, research breakthroughs, and software foundations. China excels at translating AI into industrial systems, business processes, and scaled commercialization. The two sides still compete intensely, especially over chips, standards, talent, and platform influence.

But they also pressure each other in different directions: the U.S. advances raise the technological frontier, while China’s deployment capacity pushes AI toward broader practice. Together, they shape the pace and structure of global AI development. This broader dynamic also elucidates why agentic AI products, such as Atoms, have disruptive potential.

A significant shift

This shift is particularly significant for small businesses and individual founders. For example, a local service company can develop internal booking tools, a startup can rapidly test landing pages and advertising campaigns, and a consumer brand can assess product demand before making substantial investments. In each scenario, the primary value lies not only in automation but also in time savings, cost reduction, and improved coordination.

Within the framework of complementary competition, products such as Atoms occupy the intersection of American and Chinese strengths. These products rely on foundational model capabilities, cloud infrastructure, and agent architectures typically associated with the American AI ecosystem.

However, their greatest commercial impact may occur in contexts characterized by rapid deployment, cost-effective experimentation, and industry-specific integration, domains in which China holds considerable advantages. Thus, the future of global AI will likely be determined not solely by leadership in research or deployment speed, but by the ability to connect advanced intelligence with practical economic workflows effectively.

This is where complementary competition is most evident.

We've reviewed and ranked the best business cloud storage services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The business case for running things on-premise has always started with control.

Host it yourself, keep the data in your environment, avoid vendor lock-in. It's a reasonable instinct, and for a long time it was a reasonable answer.

The gap between what you could run internally and what was available externally was manageable. On-premise was a defensible choice.

AI is changing that.

The build-it-yourself case ignores almost everything that comes after: the people required to keep things running as AI models evolve, the license fees and compute costs that compound as the landscape shifts, the upgrade cycles that never quite arrive on schedule, and the work required to unpick decisions made against a technology landscape that looked completely different six months ago.

None of these costs are hidden, exactly. They're just easy to ignore when the initial business case is about build cost.

You can run AI on-premise - just not the best AI

The frontier models - the ones getting most of the headlines - can't be self-hosted. Their providers don't make them available for private deployment.

What you can license and run internally is constantly improving, but so is the frontier. Anthropic alone released over a dozen Claude models in under two years, and they're far from the only provider.

Self-hosting means slow release cycles. Upgrades are expensive and disruptive, so firms stay on versions longer than they should. The same is true of the hardware underneath.

Specialized AI chips go out of date fast. New GPU generations arrive every couple of years, each meaningfully better than the last, and each requiring fresh capital investment. Your model is behind, the silicon it's running on is behind, and upgrading either is a major project.

Models, licenses, infrastructure, tooling, people - none of it follows a predictable refresh cycle. In the current environment, "out of date" can mean within months. Each round of investment is made under pressure, with limited time to evaluate options properly.

The talent drain

To build and run AI tools on-premise, you need engineers who aren't working on what actually differentiates your business. They're keeping up with the AI. Tweaking tools as models evolve. Troubleshooting when things break. Managing the infrastructure. Evaluating new model releases as they come out.

When it comes to data processing and reconciliation, these things are required but they're not differentiating. They need to work, but significant engineering time spent on them won't give you an edge. It's expensive maintenance of something that isn't your business.

As the internal environment expands and the technology ages, the headcount required to manage it grows. These are expensive specialists, and most of what they do doesn't move the business forward.

Why AI belongs in a cloud-native world

The argument for cloud-native AI isn't really about cloud computing. It's about whether your architecture can keep pace with a technology that's moving faster than any internal release cycle can match.

In a cloud-native world, new model capabilities arrive as features, not projects. When something better appears at the frontier, the platform absorbs it. The compliance conversation doesn't restart. The security review doesn't go back to zero. The engineering team doesn't have to rebuild anything. The capability lands, and your operations team can use it the same day.

The control argument that drove firms to on-premise in the first place still matters - but it's no longer in tension with cloud-native deployment. Permissions, audit trails, governance, data sovereignty: all of it can be enforced just as rigorously in a properly architected cloud-native platform, often more so. The trade-off has shifted. Control no longer requires standing still.

The firms that recognize this early get a head start. Their engineers focus on what differentiates the business. Their operations teams get better tooling every quarter without a procurement cycle. The question of "are we keeping up?" stops being one anyone has to ask.

What changes when you work with a trusted partner

Shifting the burden of building, maintaining, securing and testing to a specialist partner means your resources stay focused where they should be, and your capability evolves with the market.

Platforms built on infrastructure like AWS Bedrock are designed to absorb new model capabilities as they emerge - including the frontier models that can't be self-hosted at all. The underlying architecture keeps pace so the firms using it don't have to.

When a better model becomes available, the platform adapts. No new project, no additional engineers, no unravelling months of integration work. Operations teams focus on what they're there to do.

Engineers focus on the things that differentiate the firm. And the question of "are we running the right model?" stops being a quarterly investment committee discussion and starts being a setting someone flips.

Use the best business cloud storage to manage your data.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

For the past year or so, predictions of SaaS's demise have dominated the tech conversation.

Headlines warned that AI tools would replace traditional software, engineers would become obsolete and the SaaS industry itself was headed for extinction.

The narrative became so widespread that Wall Street gave it a name: the ‘SaaSpocalypse’.

Fortunately, reality has proved far less dramatic.

While the SaaS market - once the darling of investors - has undoubtedly taken a substantial hit, it is a long way from dead. In fact, I would argue that software doesn't die; it evolves.

If history has taught us anything, then it’s that major technology shifts rarely eliminate software altogether. Instead, they reshape it, with the biggest winners often being those agile enough to adapt.

From the rise of the internet to the shift to cloud computing, time and time again we’ve seen established players demonstrate an ability to evolve alongside technological change.

The AI era is unlikely to be any different.

SaaS subsectors best positioned to unlock AI-enabled growth

And I’m not alone in this view. Over recent months a number of high-profile economists, technology historians and analysts have arrived at that same conclusion. JPMorgan continues to champion the SaaS subsectors best positioned to unlock AI-enabled growth. Meanwhile, Goldman Sachs projects the global app software market could still reach $780bn by 2030, with agentic AI ultimately expanding the market rather than shrinking it.

Put bluntly, AI isn’t killing SaaS but rather separating the bad from the good. For years, SaaS startups have benefited from intense market hype and lofty valuations, often with minimum regard given to operational efficiencies, product differentiation or true value creation. Today that playing field looks a lot different. In effect, AI is now acting as a filter, revealing which SaaS models are superior and offer a genuine point of differentiation and, in turn, those which have for far too long favored style over substance.

For SaaS founders then, it isn’t time to panic. In our experience, investors-in-the-know will be more than willing to overlook the misguided SaaS narrative and recognize a genuine high potential opportunity should it arise.

Take, for example, the mid-market. Although much focus tends to surround the startup scene, I would argue that established software vendors are actually best positioned for AI transformation. After all, they already possess powerful structural advantages, including proprietary customer data, deeply embedded workflows and long-term contracts.

Moreover, they often sit on years of customer data. In this way, unlike startups where the focus is typically product, these firms actually have something meaningful to transform.

Well positioned to use AI to strengthen their market position

Far from becoming obsolete overnight, many are well positioned to use AI to strengthen their market position. Take, for example, Keyword.com. Having early recognized that search itself is changing, it launched a product to track brand visibility across AI platforms like ChatGPT, Perplexity, and Gemini. That repositioning opened up an entirely new growth vector.

In terms of how founders can best navigate this new terrain, it’s important to look beyond product and focus on structural resilience. For a long time, SaaS benefited from assumptions such as high multiples on ARR, heavy adjustments for stock-based compensation, and a willingness to prioritize growth over almost everything else.

That is now being replaced by a much more grounded framework. Investors are increasingly looking at real profitability, on a GAAP basis, and asking harder questions about cost structure, for example around sales and marketing efficiency, and now AI-related compute costs as well.

At the same time, as founders will be aware AI is introducing new variables. Revenue may be less predictable if it’s usage-based, and margins can be more dynamic because of inference costs, meaning investors won’t simply rely on simple rules of thumb anymore.

Valuation is becoming less about applying a multiple and more about understanding the underlying business, with a focus on true earnings quality, defensibility and efficiency.

Another significant but largely underappreciated upshot of the current environment is the emergence of roll-up opportunities.

The SaaS landscape is becoming increasingly fragmented

As valuations normalize and weaker players struggle to differentiate, the SaaS landscape is becoming increasingly fragmented. This creates fertile ground for consolidation strategies, particularly for investors with deep sector expertise.

Importantly too, AI enhances this model by enabling efficiencies across portfolios, from automated customer support to streamlined marketing and product development.

For investors, this is opening up a different kind of opportunity in terms of not just backing individual high-growth companies but participating in broader aggregation strategies that can unlock value through scale and operational improvement.

This means that purposefully positioning your business for ease of integration and standardization could pay dividends.

The short of it? Software isn’t going anywhere. Just like the major tech trends before it, it is simply evolving from a market that is less about hype and surface-level metrics to a more practical, mature model driven by genuine business outcomes.

For founders, this may mean navigating a more complex operating environment. But it also creates an exciting opportunity to differentiate.

Not all SaaS firms are primed for this new investment landscape, so those who pay due diligence by prioritizing intelligence over features, operational prowess and AI interoperability to deliver clear, measurable value will be best positioned to succeed.

We've reviewed and ranked the best business cloud storage.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Meta Compute could see the company sell off its excess compute
• The company expects to spend $125-145 billion on AI and data centers this year
• SpaceX recently struck up two very lucrative deals to sell compute to Anthropic, Google Cloud

Meta is reportedly looking at selling the compute capacity it has created during its AI-induced data center expansions, and it could launch a business to rival AWS, Microsoft Azure and Google Cloud.

A report from Bloomberg claims the new businesses, reportedly internally named Meta Compute, would see the company rent out excess compute capacity.

Though Meta hasn't officially confirmed such plans, a cloud computing business could allow customers to rent GPUs for AI training and inference, access Meta's models or host their own models on Meta's infrastructure.

Is Meta going to launch its own cloud computing business?

Meta anticipates spending $125-145 billion on AI and data centers in 2026, and a cloud business could help offset some of the AI infrastructure costs it's faced up to this point. It would also provide extra revenue when the company's GPUs are sitting idle between workloads.

CEO Mark Zuckerberg himself even refused to write off the possibility. "It's definitely on the table," he told investors on an earnings call.

If Zuckerberg okays the business, it wouldn't be the first of its kind. SpaceX also recently struck up deals with Anthropic and Google Cloud to sell its excess capacity. It would likely see instant success as well, because even industry giants like Microsoft are struggling to meet their own demands.

GitHub recently had to turn to AWS for extra capacity, with Azure failing to meet demands in the short term.

Meta has been struggle with share prices for several months, and while this recent news didn't trigger a full recovery, share prices did rise around 9-10% following reports indicating that shareholders are feeling more confident in the company's massive AI-related spending habits.

Cloud computing should make businesses more focused on their business, not more focused on their cloud.

That distinction has gotten lost.

A company that has moved its operations to the public cloud and then deployed significant engineering capacity to manage that environment has not simplified anything.

It has traded one operational cost for another.

Around 2012, moving to public cloud was genuinely good advice. It meant elastic scale on demand, no capital expenditure, and you only paid for what you used.

For most workloads, it was a meaningful step forward and the migration wave that followed made complete sense.

Things have changed.

The 2026 Flexera State of the Cloud report puts the current picture plainly: 73% of organizations say cloud has increased their operational complexity. And 31% of cloud spend, according to Finout, is being wasted.

There is a gulf between the promise of the public cloud and the reality, and that gulf is worth understanding.

Hyperscalers don't own outcomes

A useful analogy here is Home Depot. Asking a hyperscaler to run your business environment is a bit like asking Home Depot to build an extension on your house. They will sell you every part you could possibly need, deliver it to your door, and give you more selection than you can use.

But this analogy has a limit. Home Depot has moved well beyond lumber, and hyperscalers have done the same. You can buy managed services across the entire stack from them today, and their marketplaces run to tens of thousands of products. There is a lot you could put together from a hyperscaler.

But their commercial interests are not aligned with keeping things simple. When profit is tied to consumption, complexity is not a problem the model is built to solve. Every service you add is revenue for them. Every integration you maintain keeps you in the ecosystem. The model rewards sprawl, not efficiency.

The egress fees that make data migration elsewhere economically painful are the clearest expression of this. They ensure that even businesses that recognize the complexity problem find it expensive to do much about it. Simplicity, for the major cloud vendors, is a commercial risk.

When 78% of organizations say they are moving toward multi-cloud specifically to reduce single-vendor dependency, that is a market telling you the model stopped working for them.

The overspend problem

The overspend is just a part of the problem. Another cost is what happens to the people running these environments. When engineering teams spend their time integrating primitives and keeping a complex stack functional, they are not building products or moving the business forward. The cloud was supposed to create capacity, but for a lot of businesses it has consumed it instead.

Adding compliance needs on top compounds this. Around 43% of mid-market companies fail their first audit. The reason is rarely that the regulations are unclear. Maintaining audit-ready posture inside a general-purpose cloud environment is a continuous manual effort, and public cloud platforms give you the tools but leave the work to you. That burden lands on internal teams who are already stretched managing everything else the environment demands.

Cloud that helps businesses focus on outcome rather than maintenance requires an environment where the operational weight sits with the provider. Compliance should be managed as an architectural property rather than a continuous configuration task.

Performance should be maintained at the infrastructure level rather than something you tune around. Bills should be predictable because the model is built that way. These should not be premium features reserved for enterprises with unlimited budgets.

AI is making this more urgent. Around 98% of AI pilots currently fail to reach production. The reason is fairly straightforward: pilots run in isolated environments, and when it is time to move into production, the storage, networking, compliance posture, and compute requirements that real workloads demand are not in place.

If a business can operate on a managed, purpose-built cloud platform, it is significantly better positioned to take AI from pilot to production, because the underlying architecture is already there. They do not need to solve for an infrastructure problem before capturing the business benefit.

Simplicity is not a concession

The businesses pulling ahead have reframed how they think about cloud complexity. Complexity is not a sign of sophistication; it is an operational cost. Every hour an internal team spends managing it is an hour not spent on the actual work.

If your engineering team is spending a meaningful portion of its time keeping the cloud environment functional, the cloud is not working for you. That is not a problem inherent to cloud. It is a question of whether your provider's model is designed around your success or their consumption.

The 97% of mid-market organizations saying they need to move workloads away from hyperscale are already answering that question for themselves.

We've reviewed and rated the best business cloud storage.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Over the past two years, most of the noise around AI has focused on the model race – whose model is bigger, faster or scoring better on benchmarks.

But as AI moves from pilots into the core of products and workflows, a familiar pattern from the early days of cloud is re‑emerging: systems are more programmable than ever, but they are also much harder to run.

And that means we now know where the most important competition in AI is shifting: from who has the “best” model to who can operate AI reliably, efficiently, and safely at scale.

AI is now hitting operational limits, not model limits

When looking at real‑world telemetry from thousands of production systems, a clear picture starts to form. Nearly 1 in 20 AI requests fails once applications reach scale, and a majority of those failures now stem from capacity limits such as rate limits, quotas and concurrency caps, rather than from model bugs or poor accuracy. That is a very different story from the benchmark charts most teams used to obsess over.

The amount of data sent per request is also climbing. Across many production estates, median users have more than doubled their token usage, while heavy users have seen volumes grow several‑fold. That growth is both a symptom of more ambitious AI use cases and a direct driver of cost and IT infrastructure stress.

You can see the impact most clearly in what many teams now describe as GPU sprawl: fragmented fleets spread across clouds and on‑prem clusters. Some GPUs sit idle while others are consistently saturated, and there is very little correlation between where GPU hours are spent and where they create business value.

The result is familiar to anyone who lived through the early adoption of cloud computing – runaway spend, unpredictable performance and capacity crises that appear out of nowhere.

How this is playing out in APAC

Across Asia‑Pacific, and especially in ASEAN, we’re currently seeing structural pressures: AI adoption is accelerating, but operational maturity is uneven.

Singapore is further along on governance and observability, driven in part by regulatory expectations and a more mature cloud landscape. Meanwhile, markets such as Indonesia, Malaysia and Thailand are moving very fast on deployment, often pushing AI into customer‑facing services while operational practices catch up.

As organizations across these markets roll out multi‑model and agent‑based architectures, they are running into reliability issues, limited visibility and inconsistent model performance. Token usage is increasing quickly, but optimization practices, such as prompt caching and context engineering, are underutilized.

That gap between readiness and deployment is already creating operational and cost debt that will be harder to unwind later.

The four operational disciplines AI teams need

With the evolution of AI resembling the early days of cloud, the good news is that we can predict, at least a little, where things are headed.

Now, the question AI leaders should be asking is this: which disciplines distinguish the teams that will cope best with this complexity?

In my view, there are four that teams working with AI need to adopt to see sustainable success:

1. Establish visibility and attribution

You cannot operate what you cannot see, and AI is no exception.

Teams need to see how GPU hours and tokens map to specific applications, teams and use cases, so they can connect that usage to latency, error rates and user impact.

That makes it possible to separate business‑critical workloads from background noise, and provide clarity into which services are driving cost or consuming capacity.

When usage is visible and attributable on a single view, decisions about where to optimize, protect capacity or dial back become much less emotional and much more data‑driven.

2. Enforce control and guardrails

Without guardrails, AI systems will consume as much capacity as you give them.

Practical controls include rate limits and budget caps, along with safeguards on agent behavior to stop unbounded retries, loops and poorly bounded workflows from exhausting shared resources.

These controls are about making consumption predictable and ensuring that one runaway experiment cannot impact core production services.

Without this discipline, AI programs tend to hit economic limits long before they hit technical ones. You end up with impressive prototypes, but unsustainable unit economics.

3. Optimize GPU utilization before scaling supply

Most teams reach for more GPUs when what they really have is a utilization problem.

GPU instances already account for a significant share of compute costs, and that proportion only grows as organizations push deeper into training and inference at scale.

But idle or underutilized GPUs create the sense of a shortage even when there is headroom in the estate. In turn, many teams can see their overall GPU bill climbing, but cannot see which workloads are driving consumption, or pinpoint the steps needed to improve efficiency.

What we learned during the early days of cloud is that in these instances, overprovisioning becomes the safest default – but then spend balloons even when there is stranded capacity in the fleet.

Treating GPU infrastructure as a first‑class system means tracking utilization so that teams can distinguish genuine capacity shortages from misallocation or fragmentation. Then, they can decide whether to free up capacity or truly add more supply.

4. Design for efficiency at the application layer

High AI costs and rates of failure come from how applications are put together, not from the models themselves.

Inefficient patterns, poor routing across providers and unoptimized prompts all drive up token usage and increase the risk of timeouts, errors and inconsistent behavior.

But with proper visibility into prompts, agents and tools in production, teams can see how requests actually flow through the system and tune for quality, latency and cost in a controlled way.

That turns the application layer from a black box into a place where efficient engineering choices are deliberate, measurable and aligned with business outcomes.

What leaders should do in the new AI race

The early days of cloud taught us that programmability without operational discipline can be as much a liability as an advantage. AI is now at a similar inflection point: the winners will not just be those with access to the most powerful models, but those who treat AI as a long‑term engineering and operations capability.

A useful test for any organization is whether it can explain where AI spend goes, how agents behave in production and which workloads it would protect first if capacity were suddenly cut.

If the honest answer is “I don’t know yet”, then the next phase of the AI journey is clear: stop chasing the next model release, and focus on building the operational foundations that will help you scale AI safely and sustainably.

We've reviewed and ranked the best business cloud storage services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Although the technology powering them has evolved in performance and efficiency, the core principle of data centers hasn’t changed much in the past two decades.

Every new generation of infrastructure has been designed to shorten the distance to users, reducing latency and increasing reliability. However AI is putting serious strain on that model, and unprecedented demand for data centers is making companies think outside the box.

Terrestrial projects now face challenges from all directions, including land availability, cooling, power and water consumption, and local opposition.

Step in orbital data centers. Some pioneers are now looking to rocket data centers up into space, and the model keeps shifting further from concept and closer to reality.

While enormous technical hurdles remain, advocates argue orbital data centers could eventually complement terrestrial campuses by shifting some of the most energy-intensive AI workloads away from land-based constraints.

Networking is what will define good orbital compute

Some of the benefits include virtually unlimited space to expand, uninterrupted access to green solar energy and cooler environments that don’t require so much cooling.

But while sending compute into space tackles some of the problems we face today, it reverses one of the biggest fixes we’ve spent years improving – latency.

In today’s instant economy, a data center has no use if it can’t exchange information quickly and reliably. Terrestrial networks lean heavily on fiber networks and physical connections, but orbital data centers would need to improve wireless transmission across hundreds of miles.

As work on this progresses, Internet exchange operator DE-CIX CEO Ivo Ivanov says orbital compute should be viewed as another layer in our increasingly distributed digital ecosystem – not an outright replacement for terrestrial compute.

To better understand whether orbital data centers are just another ambitious experiment or whether they could be the next evolution of digital infrastructure, I spoke with Ivanov about the industry’s biggest challenges.

• You noted that SpaceX's IPO has brought the idea of AI data centers in space into the mainstream. Setting the hype aside, how close is this to real, and what's actually driving the biggest companies in the world to look at space?

The idea of placing data centers in orbit is the kind of thing that immediately captures people's imagination. It belongs in the same category of technological ambition as reusable rockets, commercial spaceflight, or landing on the Moon.

For decades it felt like something that belonged in science fiction. Today, for the first time, it's being discussed as a serious infrastructure proposition. But I don’t think the real story here is about space – it’s about AI, and how it’s altering our global trajectory.

Over the last few years, we've seen an extraordinary increase in demand for compute capacity. AI training clusters are growing larger, power requirements are rising, and in many regions the availability of energy, land, and cooling has become a genuine constraint.

These constraints are what drive innovation, and that’s what we’re seeing play out here.

In theory, space offers access to continuous solar energy and room to scale without the physical limitations we experience here on Earth. Having said that, we're still at a very early stage.

I don't think we're talking about replacing terrestrial infrastructure anytime soon. What we're seeing today is the beginning of a serious exploration into how orbital resources could complement the digital infrastructure ecosystem that already exists.

There is still work to be done, particularly around networking – putting an AI data center into orbit would be a milestone achievement, but it won’t count for much if we have to truck hard drives into space to do any training. For now, all anybody can say with confidence is that it's becoming part of a broader conversation about how society will support the next generation of AI workloads.

• What problems could orbital data centers potentially solve that terrestrial infrastructure struggles with today?

I think the broader answer is that orbital infrastructure has the potential to ease some of the physical constraints that have shaped and limited digital infrastructure for decades. We can’t simply keep building more data centers – they need access to land, power, cooling, transport links, and of course, connectivity.

As AI continues to grow, some of those resources will become harder to secure, especially at the pace the industry wants to move.

That's why the industry is starting to look at non-terrestrial alternatives. Suddenly, you're talking about access to potentially unlimited solar energy, plenty of space, and the opportunity to scale at a pace that matches our ambitions for AI.

But putting data centers into orbit, as impressive as it would be, is only the first step. Google’s Project Suncatcher is already exploring how power might be harvested from the sun, while the European Space Agency’s OFELIAS project is exploring how optical feeder links between Earth and orbit can be optimized to reduce latency and disruption. There are lots of moving parts.

• From a networking perspective, what challenges will emerge once data centers move into orbit?

The moment you put a data center into orbit, networking becomes the make-or-break factor. A data center on Earth can usually rely on a dense ecosystem of fiber routes, Internet Exchanges, cloud on-ramps, and interconnection facilities. In orbit, that ecosystem doesn't exist yet.

Every workload, every application, and every AI model depends on data moving between different locations, and suddenly those journeys become much more complex.

A lot of the discussion I’m hearing focuses on bandwidth or speed, but I think predictability is the more interesting challenge. AI systems need data to arrive quickly, but they also need it to arrive consistently.

Optical and laser-based communications have enormous potential, but they also introduce new variables like cloud cover, atmospheric turbulence, satellite handovers, and changing orbital positions that can all affect how data moves between Earth and space.

That's why I believe the real challenge isn't simply connecting an orbital data center to the ground, but creating an interconnection layer that makes orbital, terrestrial, cloud, and edge infrastructure behave as though they're part of the same ecosystem.

That's a much bigger networking challenge than people realize.

• Elon Musk said orbital infrastructure is "only a few milliseconds away." How much could those milliseconds matter, and is latency the biggest challenge?

Those milliseconds absolutely matter. AI systems are fundamentally data-driven, and every additional moment spent moving information between users, applications, models, and infrastructure affects performance and productivity.

That's one of the reasons the industry is investing so heavily in edge computing – the closer you can bring compute resources to the data, applications, and users they serve, the better the experience tends to be.

Certainly, low-Earth orbit is only milliseconds away, but for latency-sensitive use cases, the 20-40 ms it takes to reach the stratosphere will provide a significant challenge for some AI inference use cases. So not all applications can possibly be served from hundreds of kilometers above the ground.

Having said that, performance isn't just a question of latency. A network that consistently delivers twenty-five milliseconds of latency is often more valuable for many enterprise use-cases than one that fluctuates between twenty and a hundred.

Physics determines how quickly data can travel, but factors such as atmospheric conditions, satellite handovers, routing decisions, and network architecture all influence how predictable that experience is.

• If a data center is in orbit and the AI workloads it serves are on Earth, what actually has to happen for those two worlds to behave as one network?

The way I like to think about it is that users shouldn't have to think about where a workload is running. Whether an application is being served from a terrestrial data center, an edge location, a cloud region, or one day an orbital platform, the experience should feel seamless.

Achieving that is much harder than it sounds, because data has to move continuously between different environments, and those environments need to behave as though they're part of the same network, even when they're separated by hundreds or thousands of miles and, increasingly, by the boundary between Earth and space.

That's one of the reasons projects like the ESA's OFELIAS initiative, which I mentioned earlier, are so important. As part of that initiative, DE-CIX is working alongside partners including the German Aerospace Center to explore how optical feeder links between satellites and the ground can become more stable, efficient, and predictable.

Rather than simply building more connections, we need to create an infrastructure foundation that allows terrestrial, satellite, and orbital resources to work together as a single ecosystem.

We've spent decades building that kind of seamless interconnection on Earth, and now we need to take what we’ve learned and combine it with new technologies to bring non-terrestrial infrastructure into the picture.

• Where does this leave terrestrial data centers and interconnection? Will orbital compute compete with what exists on the ground, or depend on it?

I don't see this as a competition at all. Digital infrastructure has a long history of adding new layers rather than replacing existing ones. Cloud computing didn't eliminate enterprise data centers, edge computing hasn't eliminated centralized clouds, and satellite connectivity hasn't replaced terrestrial networks.

Instead, each technology tends to find the role it's best suited to, and the overall ecosystem becomes more capable as a result.

I think orbital compute will follow a similar path. Some AI workloads may benefit from being processed closer to abundant energy sources in orbit, while others will remain in terrestrial edge data centers because they require ultra-low latency inference, regulatory compliance, or proximity to users and applications.

It’s not about choosing one over the other, but creating an architecture where workloads can be placed wherever they make the most sense and where data can move seamlessly between them.

• What should the industry be doing now to prepare networks for orbital compute?

I think the first step is recognizing that orbital compute is more than a compute challenge. It's a networking one. The industry has spent decades building highly interconnected ecosystems on Earth, where data can move efficiently between enterprises, cloud providers, networks, content platforms, and users.

If compute begins expanding into orbit, we'll need to ensure the same level of interconnection exists between terrestrial, satellite, and orbital environments. Advances in satellite communications, laser-based networking, and distributed infrastructure are creating new possibilities that didn't exist a decade ago.

What I'd like to see now is greater collaboration between network operators, cloud providers, infrastructure companies, and space organizations.

The technologies themselves are advancing rapidly, but the challenge is ensuring they evolve as part of a coherent ecosystem rather than a collection of isolated systems.

• Looking ten years ahead, what does a fully interconnected ecosystem spanning terrestrial, edge, cloud, satellite, and orbital infrastructure actually look like? What technologies will make it work?

The most interesting thing about the future is that people probably won't think about infrastructure at all. Today, we spend a lot of time talking about whether something runs in a cloud region, an edge location, a data center, or on a particular network.

Ten years from now, if we succeed with orbital infrastructure, workloads will simply move to wherever they can be processed most efficiently, and data will flow across terrestrial, satellite, and orbital environments as naturally as it moves across continents today.

When we look back at the great infrastructure achievements of the past, whether it's railways, electricity grids, undersea cables, or the Internet itself, their success wasn't measured by the technology, but by how completely they disappeared into everyday life.

People stopped thinking about the infrastructure and started focusing on what it enabled. I think we'll see something similar here. The future of digital infrastructure will be defined by our ability to connect every layer of the ecosystem into something that feels seamless.

That's when we'll know we've succeeded.

Generative AI has moved quickly from experimentation into early production use in many enterprises. However, very few can confidently forecast what it’s going to cost them in six months.

For a technology that has consumed so much board-level attention and capital, that reflects a lack of certainty, and one that some technology leaders may privately recognize as true of their own organizations.

The spend is real and the direction is clear, but the number at the end of the year can remain genuinely uncertain.

To capture a glimmer of the confidence driving the infrastructure race, Amazon’s CEO has indicated it expects to spend heavily on IT infrastructure to support AI, with an estimated $200 billion in AI capital spending, arguing it is “not going to be conservative” in how it invests in the tech.

In practice, what makes AI different from the infrastructure investments that came before it is not the scale of the commitment but the nature of the consumption.

Cloud computing was unpredictable when it arrived too, but it eventually settled into patterns that finance teams could learn to model. AI hasn’t settled in the same way yet, and much of the reason comes down to how it is being used.

A great deal of enterprise AI use remains exploratory, which is part of what makes forecasting harder. And unlike cloud, which stayed largely within technical teams for years before spreading, AI is moving across the whole organisation almost immediately. That changes everything about how you try to govern it.

The limits of financial visibility

On the surface, some forms of AI appear to offer what earlier infrastructure lacked: clean, granular, real-time data and what it costs. But across the rapidly growing landscape of technology providers leveraging AI in some way, many do not.

In some cases, token-based pricing is precise in a way that early cloud billing never was, and for finance teams accustomed to working with far less, it can feel like a step in the right direction for solving the visibility problem.

We unfortunately still have a long way to go, since simply understanding what was spent last month tells you very little about what will be spent next quarter, particularly once adoption moves beyond the teams who originally shaped the business case.

One must consider that teams across legal, HR, and customer operations are not thinking about token economics (tokenomics). They’re only thinking about whether the tool works.

Cost exposure builds not through any single decision but through dozens of small expansions, each reason in isolation, none of them reflected in a comprehensive forecast. By the time anyone joins the dots, the demand curve has already moved.

Extending the disciplines that already exist

The organizations who are doing a better job managing AI spend have tenured experience managing consumption-based technology. IT asset management (ITAM) teams for example often have more experience dealing with more fixed constructs like users or seats, which makes the consumption-based nature of AI far more challenging.

FinOps teams on the other hand have grounded experience in managing consumption that originated in public cloud. FinOps teams may therefore better positioned to deal with the new tsunami of AI consumption and spending, ensuring that it is governed as adoption scales.

FinOps has also been broadening its scope beyond the initial roots in public cloud, with AI cost management now sitting firmly within that remit for many, a shift reflected in how the FinOps Foundation is increasingly incorporating AI into its guidance. Part of that expansion is about forecasting demand that behaves differently from conventional workloads.

There is also growing interest in whether AI itself can support FinOps practices, particularly in anomaly detection, optimization and, over time, forecasting, as consumption patterns become harder to model.

The challenge is applying FinOps practices early enough so that governance shapes how AI scales, rather than scrambling to restore control once spend has already outpaced oversight.

The compounding difficulty of legacy environments

For organizations whose technology estates were built around consistency, extending governance into AI is harder than it sounds.

AI-first organizations design with cost in mind from the beginning, treating inference the way they would any other product input, with economic constraints shaping architecture decisions before commitments are made.

Retrofitting AI into legacy infrastructure means something different. Existing commercial commitments and operating models do not adapt quickly to a consumption model that is inherently variable, and that friction has a direct bearing on cost.

The difficulty is often that AI is being introduced into environments built around very different assumptions about how demand behaves, and that is part of what makes forecasting harder.

The challenge is not simply new spend, but expenditure ballooning in environments where oversight and control are already difficult to maintain.

Organizations navigating this will tend to run controlled experiments before broad rollout and are deliberate about how adoption spreads. In practice, that is often about containing unmanaged adoption early, before usage patterns, costs and dependencies become harder to unwind.

That same exposure increasingly carries beyond internal governance. As AI appears more often in customer procurement conversations, questions that were once largely internal are starting to be probed externally too.

For organizations whose governance has not kept pace, those questions can force a level of clarity they may not yet be prepared to provide.

From activity metrics to business outcomes

Beyond governance and cost control, there remains a harder question, which is whether AI investment is producing meaningful business value. Most leadership teams are not yet in a position to answer that with confidence, and the metrics currently reaching the board are not making it easier.

Model usage, inference volumes and compute consumed describe activity without explaining value. It is easy to build a compelling board update from consumption data without addressing whether any of it is moving the business.

What gets closer to an answer is understanding whether individual inferences are delivering something a customer would pay for, or something that meaningfully reduces cost or risk.

Incremental business outcome per pound or dollar of AI spend is a harder measure to produce, but it is closer to the economics that matter because it requires a clearer position on what AI is actually delivering.

That is precisely where many organizations are still finding the work harder than it looks, particularly as AI deployment moves ahead of the models used to understand cost and value.

That disconnect matters more as the market expands, because where those economics remain unclear, cost exposure can build in ways that are harder to recognize early and harder to contain later.

For many enterprises, the challenge ahead is scaling AI without allowing spend to outrun the value it is meant to create.

The best cloud storage: tested, reviewed and rated by experts.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• 79% of data centers are exposed to climate hazards, 54% face chronic heat or drought
• Higher temperatures lead to more cooling demand, creating a vicious circle
• Component longevity is also at risk from heat and other impacts

According to new First Street research, four in five (79%) or the world's data center capacity is exposed to climate hazards like flooding, wildfire and winds, with more than half (54%) of data centers located in areas that face chronic heat or drought stress.

With hyperscalers investing billions to keep up with demands created by AI, First Street stresses that today's decisions could impact cloud computing for years to come.

But even without further investments, rising temperatures and the effects of climate change could put existing facilities under further pressure.

Hyperscalers are building data centers in at-risk areas

Besides the obvious risk of component damage, higher temperatures also mean that data centers need more cooling than they previously did, driving up electricity and water consumption. Because chips and other components operate best at an optimal temperature, higher temperatures could even cause them to fail sooner.

Flooding, winds and storms also threaten electrical systems, network connectivity and fiber networks, potentially leading to outages.

The revelation comes just weeks after a separate study from insurer MS Amlin found that half (56%) of all new projects are being located in disaster-prone areas.

"Most underwriting for real assets still uses historical data, but the climate is no longer behaving the way the historical record would predict," CEO Matthew Eby said.

But with part and utility costs soaring, companies are being forced to reduce costs elsewhere, including locating campuses in cheaper areas that are often more at-risk. With companies already having to consider compute capacity, power availability, connectivity and local opposition, "climate risk is becoming an increasingly important determinant of long-term performance," First Street warns.

"Together, these risks affect NOI stability, cash flow durability, and long-term asset performance," the company added.

With demand for artificial intelligence straining supply chains across the entire development cycle, the tech sector has never been under so much pressure to perform.

But the race to build even more AI tools, to train frontier models and to automate workflows has led to a global construction boom, with hyperscalers investing hundreds of billions in huge data center projects that are themselves under social and environmental scrutiny.

Companies now face backlash over resource use – electricity and water consumption, land occupation and grid expansion are some of the biggest challenges hyperscalers are now having to address, besides tackling strained supply chains.

We’re starting to see on-device, edge and local compute emerge as a viable alternative to cloud compute, and the benefits are broad. For example, besides tackling objections to large campuses, it also delivers lower latency connections and predictable costs for enterprise customers.

AI and cloud have been synonymous, but owning edge AI could be the next competitive advantage

Tighter integrations into hybrid and on-prem deployments could also be seen as the next progression of AI, because while generative AI chatbots and basic productivity tools are well served within browsers, workflow automation and full context requires us to rethink the infrastructure layer.

For Amit Shah, co-founder and CEO of InstaLILY AI, competitive advantage now comes in the form of owned intelligence, where company systems can learn from organizational operations, workflows and knowledge.

The company’s Small Data Center approach claims to have already cut logistics routing times from 15 minutes to three, and reduced field-team training time by 60% for industrial operators.

To better understand whether the future of enterprise AI is indeed becoming more distributed, I spoke with Shah about cloud’s limits, why enterprise-grade AI has different needs to consumer tools, and the role hyperscalers could play in this evolution.

• InstaLILY launched what it calls "The Small Data Center" approach. How is that different from edge installations that have been around for years? Is the secret sauce the middleware then?

Edge installations have historically been meant for single-purpose devices running narrow inferences at the perimeter. Our “Small Data Center” operates differently with a full intelligence stack.

Our reasoning, workers, and governance all run privately, close to where work happens, and connected to the cloud as one system.

Powered by the same InstaBrain, an intelligence layer built from proprietary enterprise knowledge, with InstaWorkers™, AI workers that execute directly inside existing systems that reason the cloud runs locally that centrally executes on-site and the same InstaControl governs both.

The secret sauce isn't middleware as we stopped treating cloud and edge as a tradeoff. Deep reasoning belongs where centralized computation makes sense and high-frequency operational execution belongs closer to the work. The intelligence layer knows the difference, that is the shift.

• What's wrong with relying exclusively on "massive remote cloud infrastructure"? For all intent and purposes, the fact that they offer redundancy by default and operate an OPEX model make them a perfect combination for businesses of any size.

There’s nothing wrong with relying exclusively on a massive remote cloud infrastructure as long as your work lives in a browser tab. The hyperscale cloud is excellent at elastic reasoning and pristine redundancy. Though it’s a poor fit for operational execution in the physical economy.

The assumption that industrial AI will simply live in the cloud ignores how industrial operations actually work. Factories, warehouses, and logistics networks operate under tight latency requirements, inconsistent connectivity, and relentless pressure to control costs.

Even when connectivity isn't an issue, a generic model endpoint lacks the operational context that matters most, which are company-specific catalogs, workflows, exception logic, and decades of institutional knowledge.

No matter how capable the model becomes, manufacturers won't hand critical decisions to systems they can't govern, audit, or ultimately trust. OPEX and redundancy are real benefits, but they solve the wrong problem when the workflow itself doesn't live in the cloud.

• We have had distributed computing for decades now: from Blockchain to P2P, from bit-torrent to Skype. What's different this time around? Is AI amplifying the need for something different and acting as a catalyst?

Earlier waves of distributed systems moved files, transactions, or compute cycles around networks. This time around, computing moves intelligence through a categorical change.

AI is the catalyst because it is the first workload where value compounds at the edge. Every decision, exception, and workflow contributes to a private intelligence layer that becomes more capable over time.

Previous distributed technologies helped organizations share resources more efficiently because they didn't create proprietary knowledge. BitTorrent doesn’t get smarter the more you use it although the intelligence layer does.

The next era of enterprise competition won't be defined by who has access to AI but instead will be defined by who owns the intelligence their operations create.

• If distributed computing is such a boon for all players in the AI ecosystem, why aren't we seeing hyperscalers putting their weight behind this technology set?

Economics reward centralized consumption. Distributed inference compresses per-token and complicates a roadmap built around ever-larger central training runs. They aren’t ignoring it. They’re moving carefully because cannibalizing centralized inference is uncomfortable when it is their core business.

The pull is coming from the physical economy outward, not from hyperscalers inward. The companies leaning in hardest are those whose customers feel the pain of cloud-only architectures most acutely, such as manufacturers, industrial operators, field service businesses and logistics networks. Anyone whose work doesn't happen in a browser tab.

• You've witnessed the evolution of AI (or rather generative AI) as an integral part of it. How do you see it evolving over the next 5 years? PS: Are we in an AI-induced bubble?

The defining divide in enterprise software over the next five years will be between companies that rent intelligence versus companies that own it. The frontier-model arms race continues, but value will accrue to the layer that turns model capability into operational execution.

Autonomous AI moves from suggestion to action, from interface to infrastructure, and from a tool you use to a system that runs work.

The capital environment is certainly exuberant, but the underlying technology shift is not. This kind of exuberance is how every major platform transition in history has started.

The long-term winners will be the companies that build operational intelligence into a compounding asset, not those that merely bought the most GPUs.

Satellite constellations are already transforming global connectivity, extending the Internet’s reach to nearly every corner of the planet. Now, as conversations turn toward compute in orbit, the focus is expanding from connectivity alone to how distributed infrastructure will shape the future of digital services.

But what does it really mean for connectivity when infrastructure extends beyond Earth? The answer is crucial, because space isn’t just another deployment site, it introduces fundamentally different opportunities and constraints that challenge how modern digital services are designed and delivered.

Understanding Space Infrastructure

Space will not replicate the data center environments we know today. Constraints like launch capacity, radiation, power generation, cooling, and physical size mean orbital compute resources will likely be smaller and more specialized than hyperscale infrastructure on Earth.

Rather than lifting existing data centers into orbit, space is likely to inspire new forms of distributed compute and storage nodes, designed to complement terrestrial systems.

Because these distributed resources need to work together seamlessly, connectivity becomes the critical enabler. In other words, the real shift isn’t about relocating data centers to space. It’s about extending the distributed architecture of the Internet itself.

The opportunity lies in distribution

For decades, application architecture has evolved to improve user experience. Content delivery networks moved content closer to users. Cloud platforms introduced elasticity and geographic redundancy. Edge computing pushed processing toward the point of data creation. Space introduces another dimension to this model.

Rather than relocating entire applications, organizations will likely distribute functions across multiple environments, including terrestrial clouds, edge infrastructure, and space based platforms, all depending on what makes architectural sense.

Satellite-based Earth observation systems, for example, generate enormous volumes of imagery and sensor data. Processing portions of that data in orbit before transmitting results back to Earth could reduce bandwidth requirements and accelerate insights.

Other use cases may involve distributing cached content, supporting connectivity in remote environments, or enabling services for moving platforms such as ships, aircraft, and industrial operations.

What matters most is that space becomes another place where parts of a service can run, alongside existing terrestrial and cloud infrastructure.

Orbital connectivity follows different rules

Designing systems that rely on orbital infrastructure also requires understanding how satellite connectivity differs from terrestrial networks.

Satellite constellations are dynamic by nature. Satellites move continuously, requiring constant handoffs between spacecraft and ground stations. Network paths change continuously as satellites move and the routing topology shifts with them.

Even when performance is strong, these dynamics introduce behaviors that traditional network architectures were not built to accommodate. Latency profiles differ too. Low-Earth orbit constellations operate at a fraction of the altitude of geostationary satellites, which matters, but a 20–40ms round-trip is still not terrestrial fiber.

Bandwidth characteristics also differ. Satellite connectivity is often asymmetric, with significantly more capacity for downloading data than sending it back upstream. Power constraints and radio transmission requirements make large uplink transfers more expensive than their terrestrial equivalents.

These realities mean that not every workload belongs in space. Instead, the most effective architectures will carefully consider which components should run where, based on latency sensitivity, data volume, and operational constraints.

The internet is extending its reach

Today's Internet already spans continents and oceans through vast terrestrial and subsea infrastructure. Satellite networks aren't a future concept, they are operational now, delivering connectivity globally and extending reach to places where traditional infrastructure is difficult or impossible. Future orbital compute may yet introduce new processing and storage capabilities as well.

For service architects, this means digital platforms may soon operate across a combination of terrestrial fiber infrastructure, subsea cable systems, wireless access networks, hyperscale cloud environments, edge compute platforms, and satellite constellations or other orbital systems.

From a user’s perspective, however, none of this complexity is visible. They simply expect the service to work. But when a single user interaction may traverse a local access network, a regional ISP backbone, a subsea cable system, a cloud region, and potentially a satellite link, performance issues can emerge anywhere along that path - from congestion on a terrestrial route to disruptions at a satellite ground station.

So while space based compute infrastructure may still be years away from large scale deployment, the architectural questions it raises are already relevant today.

Understanding how these pieces interact will be the key to building resilient digital services, whether the infrastructure supporting them sits in a data center, at the network edge, or hundreds of kilometers above the Earth.

We've featured the best cloud computing provider.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Oracle says reports a Microsoft deal had broken down were "inaccurate"
• The two companies regularly explore further partnership options
• Microsoft is actively using rival clouds to support its customers

Oracle says that details reported about its cloud deal with Microsoft are "inaccurate" (via Reuters), and that Microsoft remains both an OCI customer and a partner as the two companies continue to collaborate.

"We have a tremendously collaborative and fruitful partnership, where we often talk about ways we ⁠can expand ​upon our ongoing work together," a company spokesperson said.

Business Insider had reported that a deal between the two companies was under pressure due to security and compliance concerns.

Microsoft and Oracle are still working together

Per earlier reports, the agreement could have been worth more than $3 billion, making it one of the largest cloud infrastructure leasing deals between any hyperscalers. The aim was to give Microsoft access to additional GPU and compute, which it could distribute between its own customers and for AI training and inference.

Microsoft has already had to turn to AWS for extra compute as it struggles to meet demand for GitHub's AI tools, with commits expected to increase 14x in the space of just one year.

According to the report, Microsoft was struggling with OCI's lack of Federal Risk and Authorization Management Program (FedRAMP), which it only offers to government cloud customers. Amazon and Google's public clouds, on the other hand, do have FedRAMP.

Oracle recently announced a 47% rise in cloud revenues – more than twice the growth of the entire business, which was already a very impressive 21%. At $9.9 billion, its cloud business is now worth more than half of its entire revenues.

Microsoft has not commented on the matter.

As physical AI enters our homes, workspaces and public infrastructure, it will have a transformative effect. Autonomous vehicles will become the norm on our streets, factories and warehouses will move to full automation, AI-enabled devices will assist in surgeries and medical procedures, and greater intelligence will be embedded into domestic devices.

Such is the emerging significance of physical AI, Gartner has identified it as a top strategic trend that will shape enterprise priorities over the next five years. There is no doubt the opportunities are great. But are organizations ready to roll out autonomous robots and drones, self-driving vehicles and industrial automation at scale?

Project leaders are finding that the deployment of AI in physical spaces, where they will coexist with humans, is very different from deployment of AI in an abstract cloud computing environment. Physical AI is requiring machines and systems to perceive what’s happening around them, interpret context and act autonomously in the real world.

For obvious reasons, these deployments must be proven safe and reliable. To successfully achieve this, leaders are required to overcome numerous practical complications, such as the constraints on edge devices, regulatory compliance and environmental considerations.

In addition to this, project leaders also need to convince their senior leadership teams that physical AI can be scaled across operations.

This will require them to show that the ongoing operational costs are manageable – and that a clear return on investment, be that through improved uptime, energy optimization or workforce efficiency, is evident. If they fail to demonstrate this, projects will never get past the pilot phase.

Embrace AI from the outset

To address these challenges, the first step for leaders is to ensure physical AI solutions and their benefits are factored in at the outset of any project. When organizations fail to include AI at the earliest stage – during the design and development of any product or operational environment – it creates challenges.

This typically results in fragmentation across hardware, firmware, applications and cloud computing – and results in a build-up of technical debt and diminishing returns. Siloed operational assets also result in disjointed workflows, operational bottlenecks and suboptimal performance.

Where this is the case, we often see organizations struggle to innovate and pivot whenever new commercial opportunities arise, such as through new smart consumer devices, factory robotics or in-vehicle infotainment.

Gartner estimates that the organizations taking a proactive approach in reducing, what it refers to as, “AI debt” will mature up to 500% faster over the next three years.

Enable edge inference

In contrast to cloud AI deployments, physical AI requires organizations to integrate real-time edge inference with several computing layers. Specific solutions will need to be engineered to compensate for the numerous hard constraints encountered on edge devices, including compute capacity, memory, power consumption, thermal limits and form factor.

These constraints typically force deliberate trade-offs in model size, update frequency, hardware selection and inference strategy. As edge capabilities continue to advance, these constraints can increasingly be addressed. Low power GPUs and specialized AI accelerators are expanding the range of workloads that can be executed locally.

Techniques such as model compression and quantization also help reduce computational demand while maintaining acceptable performance.

In particularly constrained environments, distributed edge architectures can be used to offload specific tasks to nearby devices. With these advances, what matters less is where intelligence runs, and more how deliberately edge constraints are engineered from the outset.

This will increase reliability, reduce reliance on cloud computing and lower the ongoing operational costs.

Run simulations

These edge engineering solutions will provide organizations with a proof of concept. But, to enable these to scale, project leaders also need to test scenarios and understand second-order impacts across operations. They will want to do this without disrupting production, compromising safety or committing capital prematurely.

Project leaders can derisk investments and validate their decisions, however, by leveraging advanced simulation platforms, such as NVIDIA’s Omniverse. This enables them to create digital twins of factories, assets and workflows, and allows teams to explore “what-if” scenarios.

Simulations allow teams to assess performance and identify constraints early. In energy intensive environments, for example, teams can assess power usage and sustainability trade-offs. This enables leaders to evaluate costs, right size capital investment, accelerate planning cycles and align stakeholders around a shared view of the future.

Build confidence

The use of simulations also helps to identify quick wins that will help leaders to demonstrate early success. This will provide crucial evidence that the technology is safe and reliable, but also that it can provide a clear return on investment.

This should act as the first phase of a staged rollout program. With physical AI, it is advisable that organizations take an incremental approach, as it will help to build confidence in the project among the senior leadership team – and remove the hesitancy that can hold projects back and prevent them from scaling.

To further instill confidence, project leaders should simultaneously roll out a structured organizational change management project too. This will prepare stakeholders and the workforce for the impact of physical AI within their operations.

Lead organizational change

The skill sets required in a physical AI project are different to those needed in a cloud AI deployment. Organizations need deeper expertise in embedded systems, real-time software and lower-level programming languages. As a result, there may be a need to augment workforces and evolve organizational structures.

To encourage acceptance of the technology, a clear communication strategy will also be necessary - one that explains how physical AI will provide value, and how the deployment will impact individual roles and processes. It may also be necessary to provide additional training and ongoing support throughout the roll out process.

Physical AI can no longer be considered a futuristic concept – it’s already transforming the world around us. It’s enabling organizations to innovate, go to market faster and seize commercial opportunities. It is also helping to optimize operational workflows, increase productivity and reduce costs.

If organizations want to take advantage and accelerate adoption, however, they must develop the solutions that work for their specific needs and derisk their deployment strategies. When they do this, organizations typically find they can scale physical AI quickly and reap the benefits sooner.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The enterprise narrative around public cloud computing has been overly idealized, treating it as an intangible, failsafe resource.

Boardrooms frequently discuss geographic regions and compute instances as if they exist purely in the ‘digital ether’, divorced from the vulnerable reality of concrete, copper wiring and power grids.

However, as the enterprise landscape grows increasingly complex, that illusion is rapidly deteriorating.

World events have reiterated the reality that ‘the cloud’ is networking and compute, all around us; powered by tangible grids on the ground. Should those assets encounter disruption for any reason, the digital economy they foundationally support grinds to a halt.

The cloud, and public cloud in particular, has been the answer to many enterprise challenges and served as a powerful accelerant for growth and innovation.

And whilst the enterprise shift toward public cloud continues at pace (with industry analysts expecting 90% of organizations to adopt a hybrid cloud approach by next year), this high rate of adoption risks a critical vulnerability when adopted ineffectively, most notably through the ‘Single-Cloud Trap’.

Despite resilient enthusiasm with cloud adoption amongst organizations, more than half are likely to remain unsatisfied, failing to achieve their anticipated results.

This dissatisfaction frequently stems from common causal factors such as infrastructural siloing and a fundamental lack of integration. For UK businesses operating under the same mindset, this presents a notable disconnect between their strategic cloud investments and robust operational resilience.

Today’s Cloud Status Quo

In today’s environment, the risk profile for enterprises reliant on cloud computing is higher than traditional models accounted for. The recent discourse around advanced AI models, and the intense subsequent discussions regarding their potential cybersecurity implications, certainly suggest that the threat landscape has irreversibly evolved.

Beyond just redundancy planning around local power outages or software bugs, a deeper focus on outages triggered by sophisticated, AI-driven threats capable of degrading an entire geographic region are more front of mind than ever.

Such an evolution makes a disconnected, multi-cloud strategy problematic. Standard architectural wisdom dictates that distributing workloads across isolated data centers within a given territory provides sufficient redundancy. While this mitigates localized faults, it does not offer protection against macro-level disruptions.

If a vendor suffers a severe, widespread outage, the very diagnostic dashboards and control panels enterprises rely on to manage the crisis are frequently hosted on that same compromised IT infrastructure, When a provider’s entire infrastructure within a region is compromised, their management tools often fail alongside their servers, leaving IT teams flying blind.

Compounding the issue, during a crisis often enterprise continuity ‘runbooks’ remain as static documentation that can quickly become obsolete during the course of a developing incident. When a major service provider goes dark, IT personnel are thrust into a reactive scramble, desperately attempting to manually stitch together alternative routing and establish secure connections on the fly, all while the business loses revenue and customer trust.

And recent large-scale outages across the globe have demonstrated that recovery in such scenarios isn’t a matter of minutes but rather days. This delay occurs because the vital connections (between different cloud providers) had not been established prior to the crisis. Attempting to build a secure network bridge while the primary infrastructure is failing is a recipe for prolonged downtime.

The Path Forward

So, how do organizations meet these challenges head-on? The path forward requires a fundamental shift in how enterprises define reliability. True resilience today is defined by the secure fluidity of the connections between multiple providers and a robust, managed adoption of Disaster Recovery-as-a-Service (DRaaS) to mitigate issues that can occur.

The perception of networking must evolve from a ‘utility’ to a vital foundation for business autonomy. There are a few key steps organizations should keep in mind to help accelerate this shift:

1. Connectivity between cloud environments need to be established during periods of stability. These network paths should be ready, encrypted, allowing critical traffic to be redirected in seconds if needed. Waiting for a crisis to begin connecting to a secondary cloud provider invites failure.

2. Independent observability, managed from a safe, distant geography, is the only way to maintain a clear view of the situation when local infrastructure crumbles. If your monitoring tools and command center are hosted in the same region as your actual business data, they will likely fail when you need them most.

3. Resilience must transition from static documentation to automated, executable code. A modern disaster recovery strategy should allow a failover to be triggered seamlessly, removing human error and the need for improvisation during a critical incident.

In the modern cloud-led world, organizational infrastructure must become as ambitious and resilient as the economy it supports. The ultimate goal is not to merely ‘exist’ on the cloud, but to be architecturally elevated above any single point of failure. By overcoming this once ‘great disconnect’, organisations will be better prepared for new and emerging threats.

We've featured the best cloud storage.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Cybercrime isn’t random. Groups like ShinyHunters don’t just target vulnerable systems. They target moments in time when any disruption hurts most. From an attacker’s perspective, timing creates leverage.

Likewise, ransomware operators have long targeted hospitals because downtime has immediate consequences. When healthcare systems go down, hospitals can’t function normally and, in some cases, can’t deliver critical care.

Attackers understand this pressure and know it collapses timelines. It turns long internal debates into urgent decisions.

That’s why the recent attack on Canvas occurred during final exam season. For many schools, Canvas is where exams live, grades are finalized, and communication flows between students and faculty.

When a system like this goes offline in the middle of finals week, there’s no graceful fallback.

Stress Is the Point

This is the typical ShinyHunters playbook, but with flair: Students, already stressed during exam season, found Canvas inaccessible. Some were in the middle of exams. Others were locked out hours before critical deadlines. That uncertainty matters. Do you retake the test? Does it get rescheduled? Do thousands of students get to pass?

These aren’t theoretical problems. They hit instantly, and they ripple outward, from students to faculty to administrators to IT teams. Customers felt the impact while they were already stressed. Attackers know that stress amplifies pressure, and pressure drives response.

What made this incident especially effective wasn’t just the outage itself, but how visible it was. This wasn’t a quiet breach discovered by a SOC at 2 a.m.

Students logged in and saw a warning message and ransomware note attributed to ShinyHunters (“rooting your systems since 19”). saw it firsthand and flocked to TikTok. Schools couldn’t quietly contain the issue while they assessed the situation.

This kind of visibility strips organizations of one of their most helpful tools during incidents: time.

Forced Transparency Works

Modern extortion has evolved. It’s no longer just about stealing data and sending a ransom note quietly to IT and security teams. Groups increasingly aim to involve the victim’s own users in the pressure campaign.

By disrupting Canvas during finals, ShinyHunters made the impact unavoidable. Every login attempt and every student exam delay amplified the attack. That dynamic mirrors what attackers have learned in other sectors. Public facing disruption forces faster decisions than quiet, private threats ever could.

This Is a Familiar Playbook

While the scale of the Canvas attack is significant, the strategy behind it isn’t new. ShinyHunters has a long history of targeting large platforms with broad user bases, especially in cloud computing and SaaS environments, where a single compromise can affect thousands of downstream customers.

Online learning platforms are particularly attractive for that reason. They centralize massive amounts of data, serve large populations, and operate on rigid schedules. Finals happen when they happen. Grades close when they close.

There’s also a structural challenge at play. Learning platforms often include multiple account tiers, integrations, and access paths. Even when core systems are well protected, peripheral or lower security accounts can create exposure. Attackers only need a foothold to gain initial access.

Why Education Is a Prime Target Now

For a long time, education was treated as a lower risk sector. That assumption no longer holds. Schools and universities operate like digital enterprises. They rely heavily on third party platforms, store sensitive personal data, and include complex identity management environments, often with constrained security budgets and decentralized oversight.

At the same time, the tolerance for downtime is shrinking. Exams, graduation timelines, financial aid deadlines, visas, and accreditation requirements don’t wait for incident response plans to play out.

Attackers pay attention to those constraints. They also understand the sensitivity of education data. Learning platforms don’t just store names and emails. They contain internal messages, accommodation requests, instructor feedback, and deeply personal communications. Even limited exposure can create real risk for students and staff.

The Lesson From the Canvas Attack

The Canvas incident isn’t solely a story about one platform or one threat group. It’s a signal about how attackers, in general, think.

Attackers have long targeted holidays and weekends. They’re also getting very good about targeting moments, any points in time where disruption is likely to maximize confusion and minimize options for victims.

For education leaders and IT teams, the takeaway is uncomfortable but clear. Learning infrastructure should be considered mission critical. Decisions about account security, identity controls, monitoring, and third party risk don’t just affect systems. They affect real people trying to complete exams, graduate, and move on with their lives.

Finals week is already stressful. Attackers know that. And they’re increasingly willing to exploit it.

We list the best online cybersecurity courses.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

OpenAI's recent financial disclosures paint a bleak picture for the industry's economic model. The company missed internal revenue and user targets whilst projecting losses that could reach $14 billion in 2026.

According to The Wall Street Journal, CFO Sarah Friar has warned internally that ballooning compute costs may outpace incoming revenue, raising questions about the company's ability to fund future computing contracts.

These headlines are symptoms of a broader structural dysfunction across the sector, where companies inflate valuations to unprecedented levels and buy compute capacity far beyond what they can efficiently deploy. At an $852 billion valuation, OpenAI's revenue growth cannot keep pace with the IT infrastructure costs required to justify that figure.

OpenAI generates revenue primarily through API access fees and ChatGPT subscriptions. But every user query costs GPU time, meaning margins on inference are razor-thin or negative.

Revenue scales with subscribers, yet compute costs scale with usage intensity. Billions go toward training next-generation models that won't produce returns for months or years.

Yet while these companies burn through capital acquiring GPU capacity they cannot profitably deploy, the hardware itself sits largely unused, processing power that could be advancing healthcare, education or development in markets that need it most.

The idle capacity paradox

The deeper problem is that the GPUs these companies are hoarding aren't even being put to work. The prevailing narrative from major AI companies is one of scarcity, used to justify premium pricing. But Cast AI's 2026 State of Kubernetes Optimisation Report, drawing on data from roughly 23,000 clusters across AWS, GCP and Azure, found that 95% of enterprise GPU capacity sits unused. Billions of dollars worth of compute, provisioned and paid for, rarely being put to productive work.

This means everyone outside the leading model companies and cloud computing giants is, in a sense, locked out. Startups, scaleups and institutions from London to Lagos cannot access compute at an acceptable price point. The hardware exists, as does the demand, but the current model has no mechanism to connect the two.

Why utilization stays so low

The reasons are structural. Companies buy GPUs for worst-case scenarios, the equivalent of building a motorway for rush hour and leaving it empty the other 23 hours of the day.

Then there's the hoarding problem. In this market, owning GPUs is a signal to investors. Large enterprises keep buying well beyond what they actually need to operate. The hardware becomes a balance sheet asset first and a productive tool second.

Training runs make it worse. AI model training is intensive but periodic. A company might push its GPU cluster hard for a few weeks, then leave it largely idle until the next training cycle. There's no incentive to let anyone else use that capacity in between. So it just sits there.

The result is that the biggest players hoard hardware they barely use, while everyone else gets priced out. The people this hurts most are the smaller teams who would actually put compute to productive use, developers in Nairobi or São Paulo building applications for markets that big tech has never prioritized. These are the teams that will turn AI into something practical and useful for ordinary people, and they're locked out by pricing designed to subsidize someone else's idle infrastructure.

Distributed alternatives

The persistent presence of idle capacity suggests the need for a different approach. Rather than concentrating GPU resources behind corporate walls, distributed compute networks connect underutilized hardware with developers who need it.

The mechanics are straightforward. GPU owners, whether data centers with spare capacity, companies with idle hardware, or individuals with powerful machines, plug into a network that verifies what each machine can do. When a developer needs compute, the network handles the matchmaking. A job requiring fast response times gets routed to high-performance hardware nearby. A batch training job that can run overnight gets spread across cheaper, more distributed machines. The developer doesn't need to know where the hardware physically sits.

The orchestration layer groups compatible hardware into clusters, so a developer can deploy a workload across dozens of different machines as if they were one unified resource. The economics follow naturally. Idle GPUs represent sunk costs that could generate returns if put to work. Developers get access at lower price points. Hardware owners monetize assets that would otherwise sit dormant.

More importantly, distributed networks reduce the systemic risk of depending on a handful of centralized providers. When compute access isn't tied to one company's quarterly performance, the supply chain becomes more resilient.

The write-down clock is ticking

The AI industry is due a correction. Companies have loaded their balance sheets with GPU assets priced for full utilization. Actual utilization is 5%. At some point, the books have to reflect reality. When that reckoning arrives, compute access for the wider industry contracts with it. The companies that depend on centralized providers will feel it first. The ones already connected to distributed alternatives won't.

The infrastructure exists. The demand exists. The only thing missing is a model that matches the two without requiring trillion-dollar valuations to function. That model is already being built. The remaining question is how many write-downs it takes before the rest of the industry pays attention.

70+ best AI tools tested.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Numerous EU businesses have backed the EC push for EU sovereignty
• OVHcloud, Nextcloud, Mastodon, ​Proton, Ecosia and more sign letter
• The cloud market is currently dominated by AWS, Microsoft, and Google

Thirteen European cloud providers have signed a letter backing the European Commission’s push to end European reliance on US companies, who have dominated the market for the past two decades.

The cloud providers, alongside NGOs and lawmakers, want to see local alternatives fostered and adopted by Europe-based businesses as the sovereignty movement grows.

European businesses are concerned about how US-based providers handle data and comply with European Union (EU) regulations.

Businesses call for EU sovereignty

In predictions made by Gartner earlier this year, European cloud Infrastructure-as-a-Service (IaaS) spend will surpass China and North America's, with spending set to triple from $6.7 billion in 2025 to over $23.1 billion in 2027.

"Our ​message is simple: Build European, buy European, protect European," said European Parliament lawmaker ​Alexandra Geese.

The call for EU cloud sovereignty has come from businesses across Europe, with numerous signatories signing the joint letter. Among the signatories are French ​cloud vendor OVHcloud, ⁠Germany's Nextcloud, social networks Mastodon and Monnett Social, Swiss privacy software company ​Proton, browser company Ecosia and Dutch quantum ​chip ⁠maker QuantWare.

Six civil groups including Defend Democracy and ⁠Save ​Social also signed the letter, alongside numerous European Parliament lawmakers.

“Technological sovereignty means that Europe has ​the capacity to freely design, understand, choose from different home-grown sources, ‌build, ⁠operate and effectively regulate the digital systems on which its society and economy rely,” the groups the joint open letter, seen by Reuters.

The European Commission announced in May that it would consider new rules on how US hyperscalers can process sensitive data from European governments, offering a chance for the comparatively smaller EU-based cloud providers to edge into the market and end US dominance.

• A Meta-owned cloud business would be "definitely on the table"
• Continued data center expansion will fuel the firm's ongoing AI efforts
• All compute is allocated, but any future surplus could be sold on to customers

At its annual shareholders' meeting, Meta CEO Mark Zuckerberg has revealed that a cloud business could be "definitely on the table" as the company continues to build out its data center footprint to support AI programs.

If Meta were to go down the cloud computing route, it would have to go up against very established hyperscalers. Amazon's business already occupies a third of the market, and Microsoft and Google jointly own another third.

Zuckerberg ultimately noted that, if Meta ends up with a surplus of compute infrastructure as part of its ongoing AI efforts, it could sell or rent that extra capacity to external customers.

Meta would consider selling excess compute via a cloud business

"Almost every week there are different companies that come to us from outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium to what we've bought it at," he added.

Although the company doesn't currently sell cloud to customers, it continues to spend big on artificial intelligence. AI-related capex for 2026 is now estimated to be between $125 billion and $145 billion. That's only a touch behind the capex estimates of Google parent company Alphabet ($175 billon to $185 billion), Microsoft ($190 billion) and Amazon ($200 billion).

Zuckerberg explained the reason that Meta hasn't launched its own cloud compute yet is that it currently has a use for all of the capacity it's set to build, but that the company has confidence in continued investments because if it weren't to use all of that capacity, it could ultimately sell it on to customers.

The UK’s critical national infrastructure (CNI) is entering a converged threat era - one where cyber, physical and aerial risks are overlapping attack surfaces targeting the same high-value assets. Furthermore, data centers have been formally designated as CNI, elevating both their strategic importance and their exposure to attack.

Historically, organizations responsible for critical assets have approached security in distinct domains, silos maybe. Cybersecurity teams protect networks and data, physical security teams manage access control and perimeter protection and airspace, in most cases, has remained largely ungoverned - until the past decade or so.

But the structure, and manner of threats, has evolved somewhat. According to the UK’s National Cyber Security Centre, (NCSC) the country is now facing around four nationally significant cyber incidents per week, many linked to hostile state activity. At the same time, geopolitical tensions are increasingly playing out through infrastructure disruption - both physical and digital.

We are living in the converged threat era, and attackers are not constrained by organizational silos any longer, rather they are actively exploiting the gaps between them, and hybrid threats - whether state-backed, or criminal - are combining multiple factors to achieve their objectives.

Hybrid threats in practice

A typical campaign might begin with cyber reconnaissance, identifying vulnerabilities in IT infrastructure or operational technology (OT) environments. This may be followed by physical surveillance, potentially conducted via drones, to map access points, movement patterns, or security weaknesses.

Drone incidents near sensitive UK sites have more than doubled year-on-year, demonstrating low-cost, accessible technology can now bypass traditional perimeter controls entirely. And the physical infrastructure underpinning the digital economy remains exposed - with 95% of global data traffic dependent on subsea cables vulnerable to disruption.

Securing distributed, high-value infrastructure

For CNI environments, which are often lightly staffed, and in remote or hard to reach locations, the security challenges have escalated. Data centers are a prime example. While they underpin the digital economy, many operate with minimal on-site personnel, relying on remote management and automation.

There are now between 11 and 12,000 data centers globally, supporting everything from financial services and cloud computing to AI workloads and critical government systems. In the UK alone, the sector generates around £4.7 billion in annual gross value added (GVA) and supports more than 43,000 jobs, with projections suggesting it could unlock a further £44 billion in economic impact by 2035. However, this has also made them increasingly attractive targets.

Recent incidents highlight this growing risk. In 2024 and 2025, several high-profile cyberattacks targeted data center operators and cloud providers, disrupting services and exposing vulnerabilities in interconnected systems.

Many modern facilities are designed to run on remote monitoring, automation and centralized control systems. While this delivers efficiency, it also creates conditions where threats can go undetected for longer periods. This is the same with other areas of CNI, including energy infrastructure, transport networks and logistics hubs with remote assets and reduced human presence all contributing to expanding risk.

State-backed groups increasingly operate through intermediaries, such as criminal networks, to allow them to obscure their involvement while still achieving their objectives. Likewise, those with a purely criminal intent are adopting techniques traditionally associated with nation states, further blurring the lines.

To combat this, we are now seeing how organizations approach potential threats that are coordinated, adaptive, and designed to exploit both technical vulnerabilities and organizational blind spots.

Despite this shift, many security strategies are fragmented, and physical security systems often operate independently of cybersecurity platforms. Access control data is not always integrated with wider threat intelligence, and airspace monitoring, where it exists, is rarely connected to ground-based systems outside governmental and military circles.

In converged threat environments, this becomes a liability, because an isolated access control system may detect unauthorized entry, but without integration into broader monitoring frameworks, it cannot provide context. Furthermore, a cybersecurity platform may identify unusual network activity, but without visibility of physical access points, it may miss a critical link in the chain.

The same applies to drone activity; detection alone is insufficient if it is not connected to incident response protocols or wider security systems.

A multi-layered, intelligence-led model

We’ve moved forward rapidly to tackle this, however, and security is evolving into a multi-layered, intelligence-led model, reflecting the reality of hybrid threats and converged risk.

At ground level, this now looks like robust perimeter protection and intelligent access control. Modern systems go beyond basic entry management, incorporating real-time data, identifying verification, and integrating with wider security platforms. The goal is a 360 view of control access, where we have a full understanding of it, who is on site, why, and whether that activity aligns with anticipated patterns.

The CNI community and security suppliers are also on the front foot in addressing the growing airspace gap. Counter-UAS (unmanned aerial systems) capabilities, including drone detection and mitigation, are becoming essential to CNI security.

Equally important is the digital layer, with continuous network monitoring of IT and OT environments providing visibility into network activity, system performance, and potential compromise. In sectors such as energy, transport and data centers, where operational systems are increasingly connected, this visibility is essential to maintaining both security and resilience.

This enables organizations to achieve real-time situational awareness, where data from multiple sources is collected and analyzed to provide a comprehensive view of risk. An access control event can be linked to network activity and drone detection can trigger automated responses across physical and digital systems - and ultimately, patterns can be identified.

This approach is increasingly aligned with the direction of policy and regulation across governments and international bodies. This includes NATO and the EU, which are placing greater emphasis on the protection of CNI, given the economic and political importance of such assets.

Perimeter defense alone is no longer sufficient, nor is a purely cyber-focused strategy. Instead, organizations will continue to adopt a layered approach that recognizes the interconnected nature of modern, hybrid threats, and infrastructure.

Protect yourself with the best firewall software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Artificial Intelligence (AI) and very large (hyperscale) datacenters are not new technologies. What is new is how they are both now commingled and supporting each other in an unprecedented growth spurt.

The datacenter (DC) industry is currently going through a period of fundamental structural change, driven mainly by the rapid transition from traditional cloud computing to AI computing.

Foundational to this transformation and a key enabler is fiber and the extreme fiber densification that is occurring within AI DCs.

This shift is fundamentally reimagining the "Physical Layer" from a passive utility into a strategic asset. While traditional IT infrastructure relied on hierarchical, north-south traffic flows, AI "factories" demand a massive, east-west parallel architecture where thousands of GPUs synchronize in real-time.

This architectural pivot creates a literal space-crunch; we are attempting to squeeze terabits of intelligence through physical pathways—conduits, trays, and ducts - that were never designed for such volume.

Consequently, the industry is moving past the "standard" 250-micron fiber toward ultra-thin, high-density 160-micron solutions. It is no longer just about the speed of light, but about the density of glass. To scale intelligence, we must first master the physics of the fiber that carries it.

The Density Dilemma

The difference in how AI workloads are processed compared to how traditional cloud workloads are processed is the basic driver of this increase in fiber volume. In a traditional cloud environment, tasks like web hosting, database management, and file storage are typically processed by Central Processing Units (CPUs) in a hierarchical fashion via access, aggregation and core layers.

Here, data mostly moves between the end-user, who may be located anywhere outside the DC, and the server within the DC. This type of traffic flow is described as ‘North-South’ and requires robust external gateways but relatively modest internal interconnects within the DC.

However, these traditional network architectures are ill-suited for the ‘all-to-all’ communication required by Graphics Processing Unit (GPU) or Accelerator clusters in AI DCs because they introduce latency and bandwidth bottlenecks at oversubscribed aggregation points; this cannot be tolerated by AI compute stacks.

AI factories use iterative, compute-heavy processes where thousands of GPUs must work in tandem as a single logical entity. This creates a huge amount of internal traffic because of the enormous parallel processing needs across the interconnected GPUs.

In such environments, the network must support the constant synchronization of the AI model parameters and the timely exchange of vast amounts of mathematical modelling data, creating a huge volume of internal traffic. This type of traffic flow is referred to as ‘East-West’ traffic and it now accounts for most of the movement of data within AI DCs.

The architecture to support large East-West traffic is very fiber dense because the number of East-West connections scales quadratically with the number of nodes (GPUs/Accelerators).

Why ‘Fiber-First’ is the New Standard

The scale of East-West traffic is orders of magnitude larger than the traffic moving North-South. Any link error or packet loss in this network is catastrophic; for instance, it can force the entire training batch to be restarted, leading to significant financial losses and delays in model development. Only fiber can meet the bandwidth and speed requirements needed for AI processing.

AI DC operators are also deploying network fabrics such as InfiniBand or high-performance Ethernet and these require lots more high-performance fiber than traditional CPU-based networks to meet AI performance metrics.

To build a data center whose infrastructure will not be obsoleted at the next step-increase in speed (say from 800G to 1.6T), we have to adopt a fiber-first approach at the network design stage. Scalability is now a basic DC necessity.

It’s not ‘Speed of light’, It’s now ‘Speed to Light’

The pressure to deploy AI-ready capacity quickly is unending and high-density, pre-terminated fiber solutions are really the only deployment option available when building AI DCs. Instead of spending weeks on-site splicing multi-fiber elements, pre-terminated fiber solutions offer advantages such as:

Plug and Play: Significantly reducing deployment timelines (from weeks to days).

Reduction in Human Error: Pre-terminated solutions are factory assembled and tested.

Simplification: Make it easier to upgrade or swap components as technology evolves and cable management is more controlled.

Pre-terminated solutions are a key enabler of the speedy deployment of fiber networks, allowing for quicker commercial activation of AI DCs.

The Shift from Compute-Led to an Infrastructure-Led Design

Optimised fiber connectivity in an AI DC is no longer an afterthought; it’s now an important expectation within best-practice AI DC design. We are moving away from a world where we buy servers or AI compute pods and then figure out how to connect them.

Instead, we are now designing DC environments where fiber connectivity is key to determining the efficiency, speed of deployment, scalability, resilience, operational agility, and long-term readiness of the DC facility.

This shift is especially important as AI workloads demand denser interconnects, faster data movement, and lower latency across the compute environment. A well-planned fiber architecture helps reduce future rework, supports smoother upgrades, and ensures the DC can scale with evolving AI compute requirements.

We feature list the best Linux server distros.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

5G is deployed. Fiber is rolling out. AI tools are embedded in everyday professional life.

And yet millions of users still experience buffering, failed transactions, and AI assistants that stall before completing a simple query.

The infrastructure promise and the user reality remain stubbornly misaligned.

The reason is not bandwidth; it is measurement. We are still evaluating 21st-century networks with 20th-century monitoring logic.

QoS vs QoE: Why the Distinction Matters

Quality of Service (QoS) reflects what the network does: download speed, latency, and packet loss. Quality of Experience (QoE) reflects what the user actually feels: did the app load? Did the payment go through? Did the video stream uninterruptedly?

The Telecommunication Standardization Sector (ITU-T) defines QoE as "the degree of delight or annoyance” experienced by the user, intentionally shifting measurement to the human perspective.

A network that meets every technical benchmark can still fail to deliver a usable experience if the application layer, the Content Delivery Networks (CDN) routing path, or the cloud infrastructure between operator and end user introduces degradation.

This gap between technical performance and user perception is where operators lose loyalty and where traditional monitoring provides limited visibility.

Recognizing the Limits of Infrastructure Alone

The mobile industry has committed to an extraordinary level of investment in the pursuit of better connectivity. GSMA Intelligence estimates that operators will invest $1.5 trillion in capital-expenditure between 2023 and 2030, of which more than 90% is directed toward 5G. As of early 2024, 261 operators had launched commercial 5G across 101 countries.

The industry is also increasingly aware that infrastructure alone does not guarantee a good customer experience. Operators are integrating AI tools into network management, deploying 5G Standalone architectures that enable network slicing and quality-on-demand, and building API frameworks.

The direction is clear: the industry is moving toward experience-aware network management. The challenge is that this movement requires measurement frameworks capable of capturing experience, not just infrastructure performance.

AI Raises the Bar and Exposes Monitoring Gaps

The mass adoption of AI assistants, copilots, and generative tools introduces new experience metrics. Time to First Token (TTFT), query completion rates, and response streaming consistency. These determine whether an AI assistant is genuinely useful in a professional context. They are currently invisible to a traditional Network Operations Center.

A connection that meets every conventional QoS threshold can still make a large language model practically unusable. As enterprises embed AI into core workflows and as operators position AI connectivity as a monetization opportunity, the inability to measure AI-level QoE becomes both a commercial and a technical blind spot.

The Hidden Layer: CDNs and Cloud Infrastructure

One of the most underappreciated sources of experience degradation sits between the operator and the application. CDN and cloud computing infrastructure can cause buffering, slow loading, or stalled AI responses.

According to the Ericsson Mobility Report, video represents roughly 74% of global mobile network traffic as of 2024 and most of it is delivered through CDNs that no single operator controls end-to-end.

True QoE measurement must span the full stack, from the radio access network through CDN, cloud availability, and application responsiveness. Without this visibility, troubleshooting becomes guesswork.

Regulators Shifting from Coverage to User Experience

Regulators worldwide are moving toward experience-based oversight.

In the US, the FCC's Measuring Broadband America program uses crowdsourced measurements via the FCC Speed Test app to capture real-world performance across both rural and urban areas.

In India, TRAI’s MySpeed app performs a similar function, enabling citizens to submit real-device measurements that feed directly into regulatory analysis.

These initiatives share a common logic: the most credible measurement of network quality is what citizens actually experience, collected at scale, continuously, and independently.

Digital Inclusion: What You Cannot Measure, You Cannot Fix

The digital divide has always been a policy concern. Today, it is also a measurement challenge. According to the ITU, globally, 83% of urban residents use the internet compared to 48% of rural populations.

Critically, the urban-rural ratio has remained at 1.7 for four consecutive years, unchanged despite years of infrastructure investment.

Two users on the same operator, in the same city, can have radically different experiences depending on device, building, or time of connection. Without continuous and granular QoE monitoring, inclusion programs risk targeting the wrong areas.

From Measurement to Action: The Case for Full-Stack, Continuous QoE

None of this is achievable with synthetic lab-based testing. Emulators do not replicate device behavior under load, and controlled tests do not capture peak-hour congestion, CDN routing decisions made in production, or the compounding effect of multiple degradation factors across the delivery chain.

Measuring QoE credibly requires testing on real devices, live operator networks, and running actual applications continuously.

This methodological shift, from a “full-stack, always-on” approach, creates actionable intelligence for operators, regulators, and policymakers. It ties investment to measurable improvements in real user experience rather than theoretical performance metrics.

We list the best network monitoring tools.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Europe has a problem with Big Tech. And it's not abstract, theoretical, or something policymakers can quietly debate for another decade. It's happening right now, inside banks, hospitals, transport systems and government departments that cannot afford to fail.

A handful of US technology vendors now sit underneath Europe's most critical IT infrastructure. They decide when systems change, how much they cost to run, and what happens when something breaks. This is masked as progress. In reality, it's dependency and, in some cases, outright coercion.

Amazon, Microsoft and Google control more than 70 per cent of the European cloud computing market, while US firms supply the vast majority of enterprise software used across the continent.

When those vendors change their commercial models, customers don't get a vote. They do receive an invoice.

This is no longer about IT preference. It's about control.

The part no one likes to say out loud

For years, the software industry has pushed a simple story: change is good, upgrades are inevitable, and moving on a vendor's timetable is the price of staying "modern." That narrative has been repeated so often that many organizations have stopped questioning it.

But look closely, and a different picture emerges.

A stable system reaches the end of official support. The vendor announces a major upgrade or cloud migration. Licensing terms change. Costs rise. Security risks are quietly reframed as the customer's responsibility, unless they move.

Each step is presented as reasonable in isolation. Together, they form a funnel with only one exit: deeper lock-in.

At that point, this stops looking like a competitive market and starts to resemble a cartel. Customers technically have choices, but exercising them would mean rebuilding core systems under pressure, with limited internal expertise and no room for downtime. Vendors know this. That's why they push so hard.

I've sat across the table from CIOs who are told their perfectly stable platforms are now "legacy" overnight. I've spoken to boards facing seven-figure cost increases because a licensing model changed, not because their needs did. That's leverage, masked as innovation.

When dependency becomes dangerous

Technology failures aren't new. But the concentration of risk is.

Today, many organizations run critical operations within tightly integrated vendor ecosystems: cloud, databases, middleware, and core applications sourced from the same small group of providers. If one part fails, everything downstream feels it.

We've seen this play out repeatedly. A European travel company running tens of thousands of servers was pushed into a forced migration that would have increased cost and carbon emissions overnight. Instead, by stepping off the vendor's upgrade treadmill, it extended the life of its systems, avoided tens of thousands of tons of CO₂, and kept full operational control.

In another case, a media organization discovered that a core encryption standard had been deprecated, not because it was unsafe, but because it no longer aligned with a vendor's product roadmap. The choice presented was stark: rebuild fast or pay indefinitely. We developed the new standard into the existing system. They maintained compliance, avoided the disruption, and freed up resources for work that actually mattered.

These aren't edge cases. They're the logical outcome of a market where too much power sits on one side of the contract. That's why boards are now asking different questions about what systems they run, who controls them, and what leverage they've given away over time.

How lock-in quietly took hold

Vendor lock-in didn't arrive with a single bad decision. It crept in through decades of reasonable ones.

Software estates evolved through mergers, upgrades, integrations and bolt-ons. Contracts were signed under very different market conditions. Licensing language grew more complex as vendors consolidated and alternatives disappeared.

Over time, visibility was lost. Few organizations today can clearly map what they run, which components are mission-critical, and which obligations are contractual rather than technical.

This suits vendors. Environments that aren't fully understood are easy to control and hard to challenge.

This is why the sovereignty debate matters. It's not about nationalism or rejecting US technology. It's about whether customers are allowed to make rational decisions about their own systems – or whether those decisions are pre-made for them.

The baseline every CIO needs

Nobody is suggesting a mass exodus from Big Tech tomorrow. That's both unrealistic and unnecessary.

What is necessary is a reset.

The first step is clarity. Organizations need a true baseline of what they run, how it's used, and where contractual constraints are driving decisions. Without that, every renewal conversation is reactive.

Once that baseline exists, options reappear. Some organizations renegotiate contracts that were signed when the balance of power looked very different. Others choose to extend the life of stable systems instead of replacing them on an arbitrary timetable.

Software doesn't lose its value simply because a vendor says it should. Stability and reliability don't expire on a policy date. Recognizing that restores choice.

Taking back control

Across Europe, CIOs are already pushing back, and with data on their side.

They're entering renewal discussions with clear usage insight, a realistic view of operational risk, and defined red lines on cost and control. They're turning forced upgrades into negotiations. They're refusing to be passive.

The result won't be a sudden rupture with Big Tech. But the relationship is beginning to rebalance. Customers who understand their systems, question inherited assumptions, and stop equating vendor pressure with progress will regain leverage. Those who don't will keep paying for change that serves someone else's bottom line.

Europe's technology future doesn't depend on choosing the right vendor. It depends on whether customers remember they're allowed to say no, and have the courage to do it.

We've listed the best IT management tools.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Offering speed, flexibility, and the ability to scale without heavy upfront investment, the public cloud has for years been the model of efficiency. But as AI becomes embedded across every function of organizations, what once seemed like convenience now looks a lot more like a permanent cost burden.

That’s why many businesses are shifting from a cloud-first mindset toward a more balanced, hybrid approach, one that sees AI workloads brought back on-premise

Cloud used to be a major cost saver, but in 2026, the economics are changing quickly. Ingress and egress fees, combined with the premium charged for GPU compute cycles, have ballooned as more AI models run.

When 10% of top-line revenue goes to a cloud provider just to keep the lights on, organizations feel like they’re not simply renting infrastructure but paying a recurring tax on their own growth.

This is the state of play with the always-on nature of today’s AI models.

Frequent, high-volume tasks are driving cost increases. Enterprises are now using large language models (LLMs) to summarize internal meetings, scan customer support tickets, and run continuous retrieval-augmented generation (RAG) pipelines.

Individually, these API calls seem inexpensive. But at scale, they are a massive recurring expense. AI agents bring more complexity. These systems function more like digital employees, planning tasks, verifying outputs and retrying workflows.

From renting to owning

With public cloud pricing models, the more a team relies on AI, the more an organization pays. In other words, there’s a tax on realizing AI’s full potential.

On-prem infrastructure turns that upside-down. A one-time investment in high-performance hardware converts unpredictable monthly expenses into fixed, depreciable assets. Companies own the computing capability outright rather than paying exorbitant rent.

The cost of local hardware is often recouped quickly when compared to ongoing API usage or GPU rental fees, particularly for predictable, always-on workloads.

But cost is just part of the equation. Performance is the other.

In the cloud, workloads typically run on shared infrastructure. Organizations often operate on a “slice” of a server alongside other tenants, introducing latency, resource contention, and performance variability.

By contrast, local AI runs on dedicated hardware. There is no network lag, no shared queues, and no “noisy neighbor” interference. For end users, that translates into immediate responsiveness.

The governance imperative

Data sovereignty is another driver of the on-prem trend.

In a public cloud environment, sensitive data resides on third-party infrastructure, creating challenges for compliance, auditing, and intellectual property protection.

On-prem AI changes that dynamic. Prompts, proprietary training data, and outputs remain within the organization’s physical and logical boundaries. Compliance with frameworks like GDPR or HIPAA becomes more straightforward because data residency is guaranteed by design.

This also addresses growing concerns around “prompt leaks.” When employees input sensitive information into external AI systems, there is a risk of unintended persistence or exposure. Localized AI environments create a controlled, secure environment for experimentation and deployment.

Smaller, more efficient models are making this possible. Businesses do not need hyperscale infrastructure for every use case.

That’s why we are beginning to see the “rightsizing” of AI. Capable assistants can now run on systems with 64GB or 128GB of high-speed memory. What once required a large, expensive server can now be done with a compact, cost-effective workstation.

Hybrid model

This transition to on-prem AI does not mean abandoning the cloud.

For most forward-looking businesses, the right solution is a hybrid model. Cloud can be used more strategically, reserved for large-scale training jobs and burst workloads that require massive, synchronized GPU resources.

At the same time, local infrastructure handles agentic AI programs, internal copilots, and sensitive data analysis.

As a strategic hub rather than a peripheral, companies can build environments that are faster, more secure, and more cost-efficient than a cloud-only approach.

They can attain full control over their data, eliminate hidden costs such as egress fees, and offer their teams a better experience.

In the future, we will see one person directing a team of agents, and in an enterprise, hundreds or even thousands of agents may continuously plan, call tools, share context, verify results, and retry tasks — all of which drive token usage sharply higher. This is a fundamental shift in how AI is used.

Collectively, these trends point to the emergence of a “private AI” model.

The shift from cloud-first to hybrid and on-prem AI is being driven by a convergence of forces: economics, governance, and performance. In 2026, the question is no longer whether to use the cloud, but how to use it strategically while keeping control over the workflows that matter most.

We've featured the best cloud computing provider.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Modern businesses produce massive amounts of telemetry, and as systems grow more complex, the risk of failure rises with them.

Under normal conditions, observability platforms perform well, delivering responsive dashboards and reliable alerts.

But during incidents, when queries expand across longer time ranges and more stakeholders join the investigations, those same systems often slow down, revealing critical limitations when they matter most.

When the Architecture Doesn’t Match the Workload

The scale and complexity of modern telemetry are exposing a core architectural problem. Many observability platforms are designed for predictable, steady-state monitoring – not the unpredictable, exploratory queries required during incidents.

Monolithic, detection-oriented observability systems assume they can support both predefined monitoring and open-ended investigation. But in practice, these systems are optimized for workflows where questions are known in advance.

When unexpected issues arise, these systems struggle to handle ad hoc queries across large data sets and multiple teams.

This is not a feature gap. It is an architectural mismatch.

How Evolving Observability Environments Scale Complexity and Cost

Today’s digital environments produce far more telemetry than most organizations have dealt with in previous years, and it’s all driven by microservices, cloud-native infrastructure, and AI workloads. As data grows, teams rely on large historical datasets to investigate issues, often requiring concurrent access from multiple users.

At the same time, cloud economics have shifted. Cloud storage is relatively inexpensive, but compute (especially during large, exploratory queries) has become a dominant cost driver. Many observability platforms tightly couple storage, indexing, and cloud computing, forcing organizations to scale all three together.

This creates a structural inefficiency: costs rise with data volume, even when that data is only queried occasionally. Teams are left choosing between retaining less data, accepting slower investigations, or overprovisioning IT infrastructure for peak performance.

Why Observability is Moving Toward Decoupled, Event-Native Systems

To address these challenges, observability architectures are shifting away from tightly coupled systems toward a decoupled model in which storage, compute, and data visualization operate as independent layers. This separation introduces flexibility and allows organizations to store more telemetry, scale compute on demand, and access the same underlying data across multiple tools without duplication.

At the same time, the structure of the data itself is also evolving. Event-native systems treat events (such as application logs, user requests, and API calls) as the fundamental unit of analysis. Instead of relying on predefined indexing strategies, data is stored in formats optimized for large-scale scanning and high-cardinality queries.

This enables more flexible investigation and multiple teams can query the same data concurrently without significant performance degradation.

Systems built on this model, such as Apache Druid, have demonstrated the ability to support bursty, high-concurrency workloads while maintaining interactive performance at scale. This is critical for investigation workflows, where query patterns are unpredictable and often collaborative.

Event-native systems solve the data exploration problem, while decoupled observability architectures address scaling and cost issues. Modern and effective observability requires the right data model and architecture. This shift mirrors patterns seen in other data domains.

Business intelligence platforms, for example, initially relied on tightly coupled architectures before evolving toward decoupled systems. As data volumes grew, separating storage, processing, and visualization enabled each layer to scale independently and innovate more rapidly. Observability is now following a similar path.

Observability Warehouses: A Dedicated Data Layer

The next architectural evolution for observability systems introduces a purpose-built data layer that sits underneath traditional observability platforms like Splunk, Grafana, and Kibana.

Known as an Observability Warehouse, this allows organizations to store large amounts of telemetry, query it quickly when needed for investigations, and scale compute based on demand (instead of by data size). This improves cost efficiency and operational resilience.

Separating the data layer gives teams more flexibility to analyze the same event-native data across multiple tools, adapt to new technologies over time, and avoid being constrained by a single platform’s architecture. Retention strategies, query engines, and visualization layers can evolve independently.

As telemetry volumes continue to grow, investigations will become more collaborative and data-intensive. Organizations that align their observability architecture with how these workloads behave will be better positioned to respond quickly, control costs, and operate reliably at scale.

We've ranked the best ITSM tools.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

In the early 2000s, the architects of the internet faced a familiar-sounding modern problem: How do you build a system that handles massive, unpredictable demand without it breaking when any single part of it fails?

Their answer was to build a system of peer-to-peer networking. Rather than routing everything through central servers, P2P systems distributed load across thousands of individual nodes with no single point of failure, intelligence closer to the user, and resilience baked into the architecture rather than bolted on top.

It was a successful solution. P2P networks proved faster, more resilient, and more scalable than anything centralized IT infrastructure could match for distributed workloads.

Then, as the cloud computing era took hold, the hyperscale model became the dominant infrastructure logic of the last fifteen years. Its premise — aggregate everything into the largest possible data centers, optimize for unit cost, centralize without limit — made sense for many workloads.

But AI inference, the phase of AI that is now exploding in enterprise environments, operates on exactly the same principles that made P2P compelling in the first place.

Understanding why

Understanding why requires separating two phases of AI that are often conflated. Training a large model is a one-time, compute-intensive process. It runs well on centralized, aggregated infrastructure, and the hyperscale logic holds there. Inference is different.

Inference is every time the model is actually used: a fraud detection system flagging a transaction, a predictive maintenance system identifying a fault on the factory floor, a logistics platform recalculating routes in real time. These decisions happen continuously, in milliseconds, at the point where operations actually run.

Routing inference workloads to a distant hyperscale facility introduces latency that is simply incompatible with many of these use cases. A surgical assistance system cannot wait for a round trip to a data center in another region. Neither can an industrial safety system, an autonomous inspection drone, or a real-time customer service agent running on retail floor infrastructure.

McKinsey projects that global data center demand will more than triple by 2030, driven overwhelmingly by inference rather than training, and the infrastructure serving that demand needs to be built around what inference actually requires, which is compute close to where the decision happens.

P2P systems’ answer was to stop treating distribution as a problem and start treating it as the architecture. BitTorrent did not try to solve file transfer by building faster central servers, but it distributed the problem across thousands of nodes: each one close to a user, each one handling local demand locally.

When individual nodes dropped off, the system degraded at the margin. No central server going down took the whole network with it. The architecture assumed failure and built around it, outperforming centralized alternatives on speed, resilience, and scale simultaneously.

Edge computing

Edge computing applies the same logic to AI infrastructure. Smaller, modular compute facilities positioned close to where data is generated and consumed distribute the inference workload the way P2P distributed file transfer. Each site handles local decisions locally. The network as a whole becomes more resilient because no single facility carries the entire load.

Running that inference centrally also carries a cost that compounds with scale: Every time data moves out of a hyperscale cloud provider's network, organizations pay egress fees.

For AI workloads that require continuous data transfer between a central facility and distributed operational environments, those charges accumulate in ways that are easy to underestimate at the planning stage. Processing data locally at the edge — close to where it is generated — reduces the volume crossing the network in the first place.

A hardware shift is also changing the feasibility calculation at the device level. Neural processing units (NPUs) designed specifically for AI inference tasks are now embedded in smartphones, laptops, and industrial edge devices.

The compute required to run capable inference workloads has been falling steadily, and hardware that would have required a server rack a few years ago now fits in a handheld device.

As inference-capable hardware becomes cheaper and more physically compact, the assumption that every workload needs to route back to a centralized facility becomes harder to sustain.

Data sovereignty

As data sovereignty regulation is tightening across the EU, Southeast Asia, Latin America, and beyond, centralizing inference in a small number of facilities creates legal exposure.

For organizations operating across multiple jurisdictions, edge infrastructure resolves this by design: data is processed locally, within the relevant jurisdiction, without requiring complex legal and technical workarounds after the fact.

Finally, another important element is that power availability — not price — is becoming the binding constraint on data center capacity. In Northern Virginia, the world's densest cloud hub, utilities have projected connection timelines for large projects stretching up to seven years due to grid congestion.

Ireland's data centers now consume more than 20% of national electricity. These problems are the predictable result of concentrating enormous compute into a small number of locations, but the megawatt problem is more tractable when it does not need solving in one place.

Edge deployments, by distributing workloads across many smaller sites, spread the energy demand in a way that aligns better with available grid capacity.

None of this means hyperscale infrastructure is going away. Training workloads, large-scale data processing, and many enterprise applications will continue to run efficiently in centralized cloud environments.

The case for edge is not a case against cloud, but rather for matching infrastructure architecture to what workloads actually need.

The engineers who built P2P networks understood that distributing intelligence across the network made it stronger, not weaker.

As inference pushes AI out of the data center and into the places where businesses actually operate, that lesson is becoming increasingly relevant again.

We've rated and reviewed the best backup software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

CIOs and other executives need to be ready for another major change imperative in the form of AI… but what can they glean from past events? Smart leaders will learn from history and apply key lessons from the last landmark technology-enabled inflexion point: cloud computing and the rise of ecommerce.

I’m old enough to have lived through the cloud computing revolution and it was a remarkable time. The changes to the way we deployed and ran IT had significant impacts internally (elastic capacity, lower upfront costs, a smaller admin overhead, reduced ‘shelfware’) but also on broader business performance.

Cloud computing accelerated decision-making and lowered the risk and expenses associated with R&D and program and process trials. I believe that AI will be even more transformational and drive more impact, especially on the business side.

Let’s be honest here: all major change exercises have bumps on the road. Cloud certainly did and AI tools will not be an exception to that rule. Those who went through cloud-enabled business transformation, as I did at Accenture, need to apply wisdom, recognize familiar obstacles and opportunities, and act accordingly.

An incomplete shortlist of concerns will include:

- securing buy-in from stakeholders;

- managing up, across and down, including scope for retraining and new roles;

- not buying into propaganda or ‘tulip mania’ where hype exceeds reality and common sense;

- avoiding vendor/platform lock-in that restricts downstream flexibility;

- building necessary and complementary skills and partnerships externally;

- managing new vendors with new tariff structures or ways of working;

- re-engineering business processes to capitalize on technology change;

- and seeing this as a holistic change opportunity, not a technology silver bullet.

Let’s dive deeper into a few of these…

Understanding limitations

It’s OK not to know everything. In business, we do everything we can to look ahead, see around corners and anticipate next steps. But ultimately we rely in part on best guesses. That’s especially so in the case of AI (or cloud), the underlying technology is moving very fast. There’s nothing wrong with making incremental changes to plans and course-correcting on a regular basis. And, as goalposts inevitably move, it’s the only sensible approach.

It may get worse before it gets better. Any big change program carries risks. With cloud, we made mistakes that led to unexpected consequences like cost spikes and ‘bill shock.’ We didn’t always factor in what we would consume and how much that consumption level would cost so results sometimes got worse in the short term.

But everything we can know today tells us that investing in AI is a strategic competitive advantage. Some projects will fail and computing costs may come with a higher price tag, but these are short-term challenges that will ultimately lead to long-term progress and our ability to innovate, automate and analyze.

First-mover advantage isn’t always an advantage. In the cloud, there was industry, media and peer pressure to adopt quickly. We feared laggard status and missing out on a promised competitive advantage. Later, some former ‘laggards’ claimed a ‘second-mover advantage’ because they observed the frenzied action before reacting.

The truth of what was better (first-mover or second-mover) was somewhere in the middle. Some sectors and organizations moved fast successfully, while others were better off once the dust settled. There were significant variances caused by differences in industry, company profile, skills and risk appetite. Establish which camps you fall into to move at the right speed but make sure you are prepared to make big moves when the time is right.

Rethinking security and governance

Security and governance need a rethink. Many people underestimated the impacts on data privacy and regulatory compliance in the cloud. This led to cybersecurity compromises caused by misconfigurations and external risks due to the larger attack surface that was presented.

With AI, we need to accept that immense new power will also be used by malicious actors so we must mitigate risks caused by new processes and well-meaning employees and partners.

Vendor choice is different. New technology waves bring new vendors. Smart CIOs will look at the risk of using untested providers, analyze their backgrounds and likely futures, and make choices accordingly.

They will often be assisted by chief risk officers, identity, privacy and sovereignty experts, change veterans, and others with specialist domain expertise. Organizations must avoid being locked into vendors or platforms that restrict downstream movement and value.

Business process change is more important than technology change. Change can’t sit in an IT silo. It’s more important to understand new vistas of possibility and rethinking what business processes are ripe for AI-enabled change. Just as important is communication. So winning buy-in from above, below and across the organization is critical.

AI is exciting, fast-moving and, yes, scary. It’s fine not to get everything right but learning from history will help.

We've listed the best Request For Proposal (RFP) platforms.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

As cloud computing technology has matured and business needs have shifted, cloud adoption has accelerated rapidly over the past decade, reshaping how organizations build, scale and deliver digital services. But as cloud environments expand, so too does the complexity of securing them.

Today, many organizations are confronting what could be described as the “cloud complexity gap.” This gap is coming to be defined by the growing mismatch between the speed and sophistication of cloud-based threats, and security teams’ ability to contain them.

Recent research has revealed three reinforcing factors that have created a widening complexity gap: fragmented defenses, stretched teams and threats now operating at machine speed. Cloud environments have become far more complex than the traditional defenses designed to protect them.

Crucially, this is not an investment problem. 62% of organizations expect their cloud security budgets to increase in the next 12 months. Yet, despite this investment, 59% still rate their cloud security posture at initial or developing stages.

For all the money being spent, the maturity and effectiveness of cloud security is struggling to keep pace. Unless organizations rethink how they approach cloud security, this gap will continue to widen.

So, what are the key drivers of complexity in cloud environments – and how can they be addressed?

More tools create more risk

One of the clearest drivers of cloud complexity is tool sprawl, with nearly 70% of organizations now identifying this as a major obstacle to cloud security. As organizations expand their cloud environments, the natural response is often to deploy new security tools to tackle emerging risks. And with the surge in AI capability and adoption, there’s no shortage of said tools.

On paper, this seems sensible. More protection layers should mean stronger security. In practice, however, these tools are rarely implemented with deep integration. Instead, they create fragmented architectures where security data is trapped in silos, policy enforcement becomes inconsistent and visibility across environments is limited. Security teams are left stitching together insights from multiple dashboards to try to get a real-time picture of risk.

When tools don’t share context or operate from a centralized data model, organizations lose control. Regaining this control becomes more difficult because increasing the number of tools increases the number of entry points attackers can target. In dynamic cloud environments, these blind spots are prime targets for cybercriminals.

AI is accelerating the attacker advantage

At the same time, the rise of AI and automation software is reshaping the threat landscape. Threat actors can rapidly scan cloud environments for misconfigurations, map complex permission pathways and identify exposed data.

As the time between vulnerability, attack and response lessens, 66% of cybersecurity experts say they lack strong confidence in their ability to detect and respond to cloud threats in real time. As it stands, traditional, human-led processes alone can’t respond fast enough.

AI is also transforming how organizations operate. IT teams are rapidly introducing new AI tools and capabilities into their workflows, often under pressure to innovate quickly. But these tools can introduce new vulnerabilities and provide attackers with additional opportunities to exploit weak points when deployed without robust governance or oversight.

The challenge of hybrid and multi-cloud

Hybrid and multi-cloud environments promise flexibility and resilience. However, they also introduce a new layer of structural complexity that can be hard to control.

Nearly nine in ten organizations (88%) operate across multiple cloud models, combining public cloud providers, on-premises IT infrastructure, SaaS applications and a distributed workforce.

Each of these environments brings its own architecture, security controls, identity frameworks and configuration standards. Individually, these environments are already complex. Together, they create a fragmented ecosystem that is difficult to manage and even harder to secure.

As organizations scale their cloud infrastructure, the number of configurations, permissions and data pathways expands rapidly. Cloud infrastructure is designed to scale efficiently, but that does not mean that security automatically scales alongside it.

For security teams, the challenge is defending a constantly evolving network of interconnected systems, not a single perimeter. Hybrid and multi-cloud strategies can make it harder to triage threats and therefore respond accordingly across the full cloud environment.

Security teams under pressure

Compounding these technical challenges is the ongoing cybersecurity skills shortage.

Three quarters of organizations report a lack of qualified cybersecurity professionals, placing additional pressure on already stretched teams. However, simply hiring more people is unlikely to solve the problem. With attackers using AI and automation to expand the attack surface at speed, scaling headcount will do little to close the gap today.

Instead, organizations need to focus on reducing operational friction. Simplifying architectures, improving integration between tools and automating repetitive tasks can significantly improve efficiency. With better visibility and fewer disconnected systems to manage, security teams can focus expertise where it has the greatest impact.

Towards a unified cloud security model

Closing the cloud complexity gap ultimately requires a shift toward unified cloud security platforms. Many organizations currently run separate tools across network, cloud and application environments that were never designed to work together. This fragmented approach creates unnecessary complexity and weakens overall security posture.

A unified platform provides shared visibility across hybrid and multi-cloud environments, allowing organizations to apply consistent policies while coordinating detection and response across the entire infrastructure. This grants security teams a consolidated view of risk and the clarity required to stay ahead of emerging threats.

We've ranked the best endpoint protection software.

Recent geopolitical conflict has highlighted an increasingly important reality for the cybersecurity industry: modern conflict now unfolds across both physical and digital domains.

Alongside traditional military operations, cyber activity, information campaigns, and digital disruption have become regular parts of geopolitical confrontation.

For organizations around the world, the lesson extends beyond any single conflict. Cyber operations tied to geopolitical tensions rarely remain confined to national borders.

The IT infrastructure that powers global business - cloud platforms, telecoms networks, financial systems, and supply chains - is deeply interconnected. When instability happens in one region, the ripple effects can quickly reach organizations thousands of miles away.

This growing convergence between geopolitics and cybersecurity underscores a critical challenge: the global shortage of cyber and cloud skills needed to defend modern digital infrastructure.

Geopolitical tensions often trigger cyber activity

Periods of geopolitical instability have historically been accompanied by increased cyber activity. State-aligned threat groups, criminal networks, and politically motivated hacktivists frequently exploit moments of heightened tension to launch campaigns targeting governments, infrastructure providers, and private-sector organizations.

These operations vary widely in sophistication. Some involve advanced espionage or long-term infiltration carried out by highly capable threat actors. Others are less complex but still disruptive, such as distributed denial-of-service (DDoS) attacks, defacement campaigns, or the release of stolen data.

Crucially, organizations do not need to be directly involved in a geopolitical dispute to feel the impact. Shared infrastructure, third-party suppliers, and cloud computing providers create indirect pathways through which cyber activity can spread globally.

This means cybersecurity teams must prepare not only for highly sophisticated attacks, but also for waves of opportunistic disruption that can accompany geopolitical events.

The attack surface is expanding

One of the most important lessons from recent events is how tightly digital systems are now integrated with the physical infrastructure that societies depend on.

Energy systems, aviation networks, healthcare platforms, and financial services all rely on complex digital ecosystems that span multiple countries and suppliers. A disruption in one part of this ecosystem can cascade quickly across sectors and borders.

At the same time, organizations are rapidly adopting cloud technologies, automation software, and AI-enabled platforms, which continue to expand the attack surface.

As IT infrastructure grows more complex, defending it requires a workforce with increasingly specialized and continuously evolving technical skills.

AI is accelerating both attack and defense

Emerging technologies are also changing how cyber threats develop. Artificial intelligence and automation tools are making it easier for threat actors to identify vulnerabilities, conduct reconnaissance, and scale phishing or credential harvesting campaigns.

These technologies can accelerate attack cycles dramatically. Tasks that previously required significant manual effort - like scanning for exposed systems or generating targeted social engineering messages - can now be performed much more quickly.

However, AI also provides defenders with powerful capabilities to detect threats, analyze anomalies, and respond more effectively. The key differentiator is not simply the technology itself, but the people who know how to use it.

Which brings us to one of the most pressing cybersecurity issues facing organizations today.

The cybersecurity challenge is also a skills challenge

Despite rising cyber threats, the global cybersecurity workforce is significantly understaffed. Industry estimates suggest there are millions of unfilled cybersecurity roles worldwide, leaving organizations struggling to recruit and retain the expertise needed to protect increasingly complex digital environments.

But the issue goes beyond hiring dedicated security professionals. Cyber resilience now depends on the broader technology workforce as well. Developers, cloud engineers, IT administrators, and security teams must all understand how to build, deploy, and maintain secure systems. Without continuous upskilling across these roles, even well-funded security programs can struggle to keep pace with evolving threats.

Building cyber resilience through skills

As geopolitical tensions increasingly impact cyber activity, organizations must rethink how they approach resilience. Technology alone cannot solve cybersecurity challenges, and tools are only as effective as the people operating them.

Organizations that invest in developing cloud and cybersecurity skills across their workforce will be better positioned to detect threats earlier, respond faster, and adapt as the threat landscape evolves.

This means moving beyond reactive security measures and embedding cybersecurity capability into the broader technology workforce. Upskilling developers in secure coding, strengthening cloud security expertise, and ensuring security teams can effectively use emerging technologies like AI all contribute to a stronger defensive posture.

Closing thoughts

The growing intersection of geopolitics and cyber risk highlights a fundamental shift in how organizations must approach security. Cyber conflict is no longer something that happens only between nation-states. Its effects increasingly ripple through the digital infrastructure that businesses rely on every day.

While the geopolitical triggers may vary, the underlying lesson remains consistent: the organizations best prepared for an unpredictable threat landscape will be those that invest in developing the cybersecurity and cloud skills of their workforce.

In a world where digital systems underpin nearly every sector of the economy, cyber resilience ultimately depends on the people capable of defending them.

Check out our list of the best vibe coding tools.

For over 16 years, the Zero Trust cybersecurity model has upheld the principle of “never trust, always verify” and has become the universal standard across government and defense organizations.

But Zero Trust Access wasn’t built with Cross Domain Transfer – secure, policy-enforced data transfer between networks with varying security classifications – in mind.

That’s an issue, as modern cyber warfare runs on speedy intelligence sharing, quick coalition tempo and data-driven decision advantage.

Given this, it’s no surprise that more and more cyber leaders are recognizing that Zero Trust alone cannot secure operations that span multiple classification levels and allied nation environments.

A new strategic foundation to both enable and secure the flow of mission-critical information is taking shape.

Why cross-domain data sharing matters

In today’s high-octane international cyber battlespace, government and defense alliances such as NATO and operational initiatives such as the U.S. Combined Joint All-Domain Command and Control (CJADC2) are facing a complex problem.

As attack volumes increase, the ability to move data securely across domains and allies at mission speed for defense agencies to triage and respond becomes more crucial. At the same time, sharing and accessing sensitive data is a major operational risk for organizations at the frontlines of national security.

This conundrum requires more than Zero Trust Access to solve. Zero trust treats all network traffic as a potential breach and ensures that no users or systems are trusted by default. This suits securing IT infrastructure, but is less suited to data, especially when it comes to securing data that’s in transit.

Current conflicts demand fast, coalition-wide data sharing across multi-domain environments, such as land, cyber and cloud computing, in near real-time. In these environments, latency is not an inconvenience; it is an operational vulnerability that a sole reliance on Zero Trust opens the door to.

The sleepwalking risk

When weighing up how a sole reliance on Zero Trust causes problems, we need to consider wider factors that could contribute to a disastrous cyber security breach or lapse of cyber defenses. Without specialist mechanisms for data security and secure data exchange, national security agencies face a quiet, compounding erosion of operational tempo.

Data can’t move at mission speed, decision latency is the norm and intelligence gets stuck at classification boundaries awaiting manual transfer.

This can be seen as an internal friction for security agencies, but that risks downplaying how modern cyber warfare works in practice. Defense initiatives are predicated on the assumption that partners can share intelligence securely and at pace.

An organization that hasn't solved cross-domain data flow doesn't just put its own mission at risk; it degrades the collective tempo and trust of the entire coalition.

Shared identity frameworks and federated access policies across systems and partners alike have to be part of the solution. In a recent Everfox study, 64% of security leaders said they believe that secure data movement is the greatest barrier to achieving effective Zero Trust strategies.

In other words, Zero Trust environments struggle not because verification is unnecessary, but because verification alone does not solve cross-domain data orchestration. And when Zero Trust is implemented in isolation, it makes it harder to share data securely and collaborate.

The value of trust

To be battle-ready, national security organizations need to adopt a cohesive architecture that remedies these problems with data sharing at mission-speed and across multiple domains, partners and classification levels.

The solution is a data-centric model which applies security controls directly to data itself. This isn’t hypothetical. The aforementioned study found that nine in ten have adopted or are in the process of implementing a data-centric security approach.

What does this look like in practice? It entails integrating three frameworks: Zero Trust Architecture (ZTA), Data Centric Security (DCS) and Cross Domain Solutions (CDS). Each framework brings a key strength: ZTA provides continuous verification (the need for which is never going away), DCS applies security controls to data so intelligence can move securely across networks despite any variation in classification or trust level and CDS facilitates secure data exchange without compromising speed.

Together, this integrated architecture shifts security from static perimeter enforcement to dynamic, policy-driven data mobility, laying the groundwork and equipping organizations with the foundational tools for navigating the digital battlespace.

When deployed in concert, these frameworks shorten decision cycles and accelerate information exchange without diluting governance controls.

Where transformation begins

Zero Trust is an essential principle in government and defense organizations, but it often falls short in sharing data at mission speed. In modern conflict, battles are increasingly fought through digital networks where data is the decisive asset.

Navigating the digital battlespace requires the seamless movement of secure information across domains, and a sole reliance on Zero Trust doesn’t help to hit that requirement.

By complementing Zero Trust principles with DCS and CDS, national security organizations can usher in a new era of data-centric security based on the agility that modern mission success demands.

Check our list of the best data recovery services.

The quantum clock is ticking faster than we think.

Artificial intelligence dominates headlines as the technology most likely to reshape society—and potentially destabilize it. But while policymakers and security teams debate AI ethics, a potentially more devastating force is advancing in relative silence: quantum computing.

Google's Willow chip, unveiled in late 2024, completed a benchmark calculation in under five minutes that would take the world's fastest supercomputer ten septillion years. The quantum era is no longer a decade away—it's arriving in stages, right now.

The concern isn’t hypothetical. The accelerating convergence of AI, cloud computing, and early-stage quantum capabilities is already reshaping the threat landscape in ways most organizations are unprepared for.

The result is a growing “quantum clock” hanging over corporate America, one that is ticking down faster than legacy security models can adapt. This means increasing pressure on organizations to implement quantum-safe measures now, not later.

Why quantum changes the security equation

At its core, the issue is encryption. Nearly all modern digital security—everything from financial transactions and healthcare records to government communications—relies on cryptographic methods that assume attackers are limited by classical computing power.

Quantum computers, once sufficiently mature, break that assumption entirely. Algorithms that would take today’s fastest supercomputers thousands of years to crack could, in theory, be solved in minutes.

What makes this especially urgent is that adversaries aren't waiting for quantum maturity. Nation-state actors are already executing "harvest now, decrypt later" campaigns—intercepting and stockpiling encrypted data today, banking on future quantum systems to crack it open.

Intelligence agencies have warned that sensitive data with a long shelf life—trade secrets, defense communications, patient records—may already be compromised, years before a quantum computer ever touches it.

A breached foundation meets a future threat

Today's digital infrastructure is already porous. Years of breaches, credential theft, and supply-chain compromises mean attackers routinely begin from a position inside the environment.

When quantum-enabled decryption becomes practical, it won't arrive in a vacuum—it will land in an ecosystem already riddled with hidden access paths and dormant footholds.

This risk is amplified by how modern computing actually works.

The digital fabric no one fully sees

Most organizations are no longer running a single generation of technology. Instead, they operate a complex mix of legacy systems, containerized applications, serverless functions, and increasingly, AI agents that move autonomously across environments.

These workloads span multiple cloud providers, on-premises infrastructure, and hundreds of SaaS platforms stitched together by APIs, identities, and network connections few organizations fully understand.

This interconnected cloud environment is a constantly shifting web of dependencies, permissions, and data flows.

Security failures rarely occur because one system is unprotected; they happen at the seams— Recent incidents illustrate this: a SaaS integration drift that silently exposed customer data across trust boundaries, or an automation platform vulnerability that gave attackers lateral movement through orchestration workflows.

These aren't hypothetical. They're the kinds of gaps adversaries exploit daily.

Regulators have taken notice. Frameworks like CISA's Zero Trust Maturity Model 2.0 now mandate runtime proof of zero trust enforcement, not just policy documentation.

The EU's DORA and NIS2 directives require segmentation and full-path encryption. These aren't aspirational targets—they're compliance deadlines that most multi-cloud environments aren't prepared to meet.

AI accelerates the attack surface

AI accelerates this fragility. Autonomous AI tools now traverse systems at machine speed, chaining together actions across databases, APIs, and services in seconds.

A compromised agent—whether through stolen credentials, manipulated prompts, or poisoned training data—can escalate privileges and exfiltrate data across cloud boundaries before a human analyst even triages the first alert.

This is where the industry’s traditional mental model breaks down.

The perimeter is gone, but security hasn’t caught up

For decades, cybersecurity has been organized around the idea of a perimeter: trusted systems on the inside, threats on the outside. But in cloud and multicloud environments, that boundary no longer exists.

Every workload is potentially exposed because the threats live within the network fabric between those workloads. Every connection is a potential attack path. Even the infrastructure designed to enforce boundaries—VPNs, gateways, edge devices—has increasingly become a primary target for exploitation.

In a post-perimeter world, slogans about “zero trust” are no longer enough. The concept must move from policy statements into enforceable architecture—applied not just at login, but continuously, at the workload level, across every environment.

From slogans to enforceable architecture

This shift requires two foundational changes.

First, organizations must gain real visibility into east-west traffic: the lateral movement between workloads that attackers use to escalate privileges and reach high-value assets.

Most security tooling still focuses on north-south traffic—what enters and exits the environment—while missing the internal pathways where breaches actually unfold.

Second, security intent must be expressed in terms that reflect how modern systems operate. Static controls built around IP addresses and fixed infrastructure fail in environments dominated by ephemeral workloads and dynamic scaling.

Instead, access decisions must be tied to workload identity, function, and behavior regardless of where that workload happens to run.

Taken together, these capabilities define what the industry is beginning to call a cloud-native security fabric: a unifying layer that provides continuous visibility into east-west traffic, enforces consistent policy across every cloud and data center, and constrains lateral movement—without requiring organizations to rip and replace their existing infrastructure.

When paired with zero-trust principles applied directly to workloads—not just users—this model offers a way to contain breaches even when attackers gain an initial foothold.

Preparing for quantum without waiting for it

None of this eliminates the quantum threat. But it changes the equation.

Quantum computing may eventually break today’s encryption. Complexity, however, is already breaking today’s defenses. Organizations that wait for quantum-safe cryptography standards while ignoring architectural weaknesses risk being compromised long before quantum computers reach maturity.

In August 2024, NIST finalized its first three post-quantum encryption standards—a landmark step. Organizations should absolutely begin planning their cryptographic migration. But adopting new algorithms alone is not enough. Complexity is already breaking today's defenses.

Organizations that pursue quantum-safe cryptography while ignoring the architectural weaknesses underneath it are reinforcing the walls of a building with a crumbling foundation.

The risk isn’t sudden—it’s cumulative

The real danger isn’t an overnight “quantum apocalypse.” It’s a slow-burn failure—where harvested data, unchecked lateral movement, and unmanaged AI systems converge into a security crisis that feels sudden only in hindsight.

The path forward starts now: audit your encryption posture, map the lateral pathways through your cloud environments, and begin migrating to quantum-safe standards.

But don't stop at cryptography.

The organizations that will weather the quantum transition are those building security into the network fabric itself—where trust is enforced continuously, at the workload level, across every environment.

Because while AI may change how we work, quantum computing could change whether our digital foundations can be trusted at all.

We've featured the best endpoint protection.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

AI has quickly become part of the justification for many large software acquisitions.

According to PwC, around a third of the biggest deals completed in 2025 cited AI as part of their strategic rationale.

That momentum has carried into 2026, with AI tools shaping not just valuations, but the logic behind why organizations pursue acquisitions in the first place.

What’s changed is not just why deals are done, but what happens next. Once a deal closes, AI ambitions can conflict with operational reality.

Commitments to AI-led growth are frequently made before organizations have a reliable view of how their combined systems, data, and contracts actually fit together.

Boards expect value to materialize quickly. IT & Engineering teams, meanwhile, are still discovering what they've inherited. That gap between expectation and visibility is where many AI-driven deals begin to struggle.

Why “efficiency” becomes harder to deliver after AI-driven acquisitions

One of the earliest tensions in AI-driven M&A centers on efficiency. For deal teams, efficiency typically means margin improvement and speed. For technology leaders, it means keeping critical platforms stable, secure, and compliant while absorbing new systems and workloads.

AI widens that gap because much of the required investment arrives early. Spend on cloud computing infrastructure, data pipelines, tooling, and specialist talent often ramps up before organizations have resolved software overlap or decided which platforms will survive the deal.

In practice, this means AI investment is layered onto technology estates that are still fragmented. Duplicated systems remain in place, overlapping licenses can’t yet be retired, and contractual constraints limit how quickly rationalization can happen. Teams are asked to accelerate delivery while carrying more complexity than before.

That lack of visibility has a direct financial impact. In the UK, nearly one in every five pounds spent on SaaS is lost to unused or duplicate licenses, money that organizations under post-deal pressure cannot afford to waste.

AI does not create these inefficiencies, but it amplifies them by increasing IT infrastructure demand and placing greater strain on already fragmented technology estates.

The sunk cost trap sets in early

As integration progresses, organizations often encounter one of the hardest dynamics to escape in post deal technology strategy: the sunk cost fallacy.

Once investment has been committed to AI tooling, data migration, or platform alignment, it becomes increasingly difficult to revisit early assumptions.

Even when capabilities prove less mature than expected, data environments are more fragmented, or operating costs are materially higher than forecast, momentum tends to win out.

These outcomes are sometimes attributed to insufficient due diligence. In reality, no diligence process can fully surface the quality of data, the true cost of AI workloads at scale, or the operational dependencies that emerge after systems are combined.

Traditional integration challenges can take years to resolve. AI compresses this timeline. Misalignment often becoming visible much sooner, alongside the associated cost and risk.

If early signals are ignored, organizations can find themselves locked into decisions that are costly to unwind, with each passing quarter making it harder to change direction.

Clarity creates optionality

The organizations that deliver long-term form AI-driven acquisitions tend to slow down at the right moments. Rather than pushing ahead with integration before the basics are understood, the establish early clarity across their software, SaaS and cloud estates.

That clarity comes from knowing what software is actually being used, where licenses overlap, how SaaS and cloud commitments are structured, and how AI workloads will drive consumption over time.

Without it, AI investment is layered onto inefficiency, amplifying waste, rather than eliminating it.

For boards and executives, clarity creates leverage. A clear view of usage, entitlements, and contractual obligations allows leaders to challenge assumptions, prioritize rationalization, and pace investments before spend becomes locked in.

Deals deliver value when organizations remain willing to adapt as new information emerges. That includes revisiting integration plans, adjusting timelines, and setting realistic expectations for what AI can deliver in the near term.

AI shortens the distance between decision and consequence

AI does not change the fundamentals of software integration, but it accelerates the consequences of getting those fundamentals wrong. Spend accumulates faster, complexity compounds earlier, and the window to course correct narrows.

In that environment, the ability to see clearly, respond early, and continuously optimize becomes a defining factor in whether an acquisition delivers durable value or locks in long-term cost and risk.

Check out our list of the best data visualization tools.

The current state of 5G sits at an inflection point. As 5G standalone networks continue to grow in number, network slicing capabilities are becoming more of a reality.

The enterprise needs to take advantage of these capabilities such as predictable and reliable transport for business-critical applications, as well as the ability to support differentiated services over a single 5G connection

If enterprises can harness this power of network slicing, they’ll be able to improve business operations — with always-on critical business functions — in addition to maximizing their investments in 5G.

The answer to fully harnessing all that network slicing has to offer lies in a strategic combination of UE Route Selection Policy (URSP) and SD-WAN solutions.

The current state of network slicing

The attraction of network slicing lies in the promise of a dedicated service level agreement (SLA) for WAN networking — without a physical wire. 5G standalone networks have been a recent development, making network slicing a relatively new technology. The last WAN transport technology that was backed by an SLA was MPLS.

Recently, major carriers have begun to present a few network slicing options, which indicates immediate growth in network slicing possibilities for the near future. This will be important for multiple industries, such as public safety and retail.

In fact, currently, public safety agencies can take advantage of dedicated 5G network slices to ensure they have dedicated resources for public safety connectivity even in emergency situations when the cellular network is congested.

In industries such as retail, network slice service offerings will be important to provide dedicated resources for point-of-sale systems.

While these offerings represent exciting opportunities for enterprises, they only represent a small portion of the destination to which network slicing is heading.

The future of network slicing

Current network slicing offerings, like the ones discussed above, are single-slice offerings. This is the equivalent of purchasing a single connection for a dedicated function. This future is one of multi-slice offerings that create more than one dedicated network for critical functions within an enterprise.

For example, a hospital could use network slicing to create a slice specifically for telemedicine consultations, ensuring low latency and high bandwidth for real-time video communication between doctors and patients.

Another slice could be dedicated to critical medical devices such as remote patient monitoring systems, guaranteeing high reliability and security.

The multi-slice future of network slicing will provide enterprises a more definitive vision for how they will support further investment in AI-led innovation, 5G technology, and expanded cloud infrastructure. To capitalize on this future, however, they need the right technology that can maximize the future of network slicing.

The correlation between URPS, SD-WAN, and the future of network slicing

The implementation of multiple network slices on a single SIM card can only work if the intended applications are steered to the correct slices. It’s like taking a road trip of sorts. The only way to take the quicker route effectively is if the vehicles actually end up on that route.

For mobile phones, URSP has been the industry standard for sending the right traffic on the right network slice to the right applications in the cloud or data center. Let's say, for example, a consumer is looking for directions on a navigation app.

URSP helps the proper navigation data travel on the best path to the device so the user knows how to get to their destination.

In addition to traffic steering, URSP has the foundational technology to dynamically provision network slices on an enterprise’s modem. With each of these advantages, the question becomes, if URSP has the fundamental technology to direct data and provision network slices, where is the need for SD-WAN?

While URSP has great traffic navigation abilities for applications that reside on mobile phones, it has shortcomings when it comes to implementations on enterprise routers.

Today’s enterprises can use as many as thousands of applications and will therefore need technology that can recognize applications on a large scale and sufficiently direct traffic in a multi-slice environment.

SD-WAN (or software-defined wide area network) network technology has long been the industry standard in recognizing application traffic and transporting it across multiple WAN links.

As the enterprise steps into the world of multi-network slices, it will need SD-WAN technology that can act as the proverbial dispatcher, recognizing the application and sending it across the correct slice.

With this in mind, the foundation for a multi-slice future will not be either URPS or SD-WAN. The goal of enterprises should be to find solutions that can support both URPS and SD-WAN technology, creating a path for the most efficient use of a multi-slice framework.

Preparing for what’s next in 5G

Although network carriers have started rolling out network slicing solutions, multiple carriers have been hindered by the inability to acquire more mid-band spectrum (where 5G SA networks live). The acquisition of more 5G SA networks will translate to increased network slicing services in the future.

This wait in mid-bad acquisition was tied directly to the Federal Communications Commission’s inability to auction off this spectrum. However, due to the passing of the Trump Administration’s “Big Beautiful Bill” that the FCC has retained auctioning authority.

This should set the stage for more spectrum acquisitions, but any multi-slicing capabilities will remain in early stages in the near future.

As more spectrum is acquired and 5G standalone networks begin to roll out, enterprises have time to decide how to prepare for the network slicing future. They have the opportunity to properly analyze URPS and SD-WAN solutions that can create the best possible foundation for multiple network slices.

In turn, they will create a foundation for a robust and resilient network that can handle the enterprises’ growing dependence on large-scale AI and cloud computing technologies.

We've featured the best business smartphone.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

Digital Sovereignty has moved from a niche regulatory conversation to a strategic conversation for technology leaders and business executives.

Yet amidst boardroom discussions, media hype, and emerging regulation, a tangle of misconceptions has formed of what digital sovereignty is, what makes a true sovereign cloud and what it means for organizations of all sizes.

Far from being a theoretical construct or a compliance checkbox, sovereign cloud touches the core of how companies will secure, operate and govern their IT infrastructure in the decade ahead

Here are five persistent myths that are clouding judgement and why dispelling them is essential for any business aiming to compete and innovate responsibly.

Myth 1: Sovereign cloud is only about where data is stored

Sovereign cloud is often reduced to data residency, but digital sovereignty goes far beyond where data sits. True sovereignty applies to the entire digital estate, including the workloads, applications and platforms that process that data.

What matters is not just location, but who controls the infrastructure, who operates it, and which legal frameworks govern the systems in use. Simply hosting data in a particular geography does not guarantee sovereignty if control sits elsewhere.

In practice, digital sovereignty requires operational control across data, workloads and applications, from access and administration to monitoring and recovery.

For organizations operating across borders or handling sensitive data, failing to distinguish between data sovereignty and digital sovereignty can create hidden operational and legal risk.

Myth 2: Sovereign cloud is only relevant for governments or large enterprises

It’s tempting to see sovereign infrastructure as a “big business problem”, but that’s no longer true. Small and medium-sized enterprises increasingly face expectations from partners, customers, and regulators to demonstrate robust control over their digital assets.

Understanding where data resides, who can access it, and under which jurisdictions it lives isn’t an abstract compliance exercise. When SMEs can articulate this clearly, it smooths audits, client evaluations, insurance reviews, and commercial negotiations.

Sovereign thinking can help businesses reduce legal risk and unlock opportunities with more regulated customers, precisely because it brings clarity around control and accountability.

Myth 3: Adopting sovereign cloud means sacrificing performance or innovation and paying a premium

Some leaders worry that moving to sovereign cloud environments means giving up the agility and scale of public cloud platforms, or accepting significantly higher costs. In reality, sovereign infrastructure, performance and cost efficiency aren’t opposing forces.

Successful sovereign adoption involves leveraging hybrid architectures: combining local, compliant environments for sensitive workloads with broader cloud ecosystems for compute elasticity and AI experimentation.

This avoids the need for duplicated platforms, retrofitted controls or complex contractual workarounds that often drive cost elsewhere.

When sovereignty is built into infrastructure design from the outset, organizations can preserve innovation velocity, predictable performance and commercial efficiency, without treating compliance and control as an expensive afterthought.

Myth 4: Digital sovereignty is only about regulation

Regulation is an important driver of sovereign cloud adoption, but it is not the full picture. Digital sovereignty is just as much about operational resilience as it is about legal compliance.

When workloads, applications and management systems are tightly coupled to global platforms, local services can become exposed to failures originating far beyond their borders. Recent large-scale cloud outages have shown how issues in one geography can cascade across regions, even when data itself is hosted locally.

Digital sovereignty challenges this dependency. It focuses on architectural independence, ensuring that critical workloads, operations and recovery processes can function within defined jurisdictions, with local control and accountability.

The result is reduced systemic risk, improved uptime and greater resilience when conditions are least predictable.

Myth 5: Digital sovereignty only matters once the law says it must

There is currently no single law in the UK that explicitly mandates digital sovereignty. Many organizations therefore treat it as something to address only once clearer rules appear.

But UK policy is moving steadily in one direction. Proposed measures such as the Cyber Security and Resilience Bill, alongside growing focus on security, resilience and compliance across regulated sectors, point to tougher expectations around how digital systems are designed and operated.

Adopting digital sovereignty principles early, including operational control, workload isolation and continuous compliance, allows organizations to meet these evolving requirements without waiting for prescriptive mandates.

Rather than reacting to each new rule, sovereignty-aware architectures provide a resilient foundation that can adapt as obligations increase.

For organizations running critical or sensitive workloads, digital sovereignty is less about a single legal trigger and more about staying secure, compliant and resilient as regulatory pressure continues to rise.

Why sovereign cloud now matters to every business

Sovereign cloud isn’t an optional add-on or a checkbox on a compliance roadmap.

It is a shift in how organizations think about control, accountability and long-term resilience. In a world where geopolitical, legal and technological pressures intersect, the assumptions that once sufficed for digital infrastructure — “my data is safe because it’s in the cloud” — are no longer enough.

For business leaders, understanding and engaging with sovereign cloud thinking today means reducing risk, accelerating secure innovation and strengthening trust with stakeholders.

Sovereign cloud isn’t a future possibility, for many organizations, it’s already part of the competitive landscape.

We've featured the best cloud computing provider.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

For the past century, facial recognition technology (FRT) has existed largely in the realm of science fiction. From dystopian literature and film to speculative headlines and industry conjecture, FRT has long been portrayed as futuristic, invasive or experimental.

Yet behind the scenes, facial recognition has been quietly maturing, particularly over the past two decades.

In 2026, that maturation looks like it will reach a tipping point: FRT will no longer be perceived as cutting-edge or novel, but as a dependable, everyday enterprise technology.

This shift matters more than it might first appear. We talk about technologies becoming “boring”; it’s an important step, meaning the tech stops being treated as experimental, the hype starts to quieten down, and instead the actual solutions start to be relied upon as business-critical infrastructure.

Cloud computing followed this path. So did multi-factor authentication. AI is fast getting there. And facial recognition is now on the same trajectory.

The market signals are clear. For instance, the global facial recognition market size was valued at as estimated $8.83 billion in 2025, and this figure is projected to grow from $10.13 billion in 2026 to $30.52 billion by 2034 – a CAGR of 14.80% over the next eight years.

But a growth in investment and market value does not equal maturity. What marks 2026 as pivotal is how organizations are beginning to embed facial recognition into routine operations, not necessarily as a headline innovation but as a supporting layer that improves efficiency, security and decision-making.

FRT moves from pilot to production

Across sectors, FRT is moving out of pilot programs and into production environments. In border control, transport and travel, biometric identity checks are becoming a standard part of passenger flow management, helping organizations reduce friction while maintaining security.

In financial services, facial recognition is increasingly used to strengthen identity verification, protect against fraud and support remote onboarding, particularly as digital-only interactions become the norm.

In workplaces, healthcare environments and secure commercial premises, facial recognition is being deployed to manage access control and ensure that only authorized individuals enter sensitive areas.

What these use cases have in common is not novelty, but necessity. As organizations scale, operate across distributed environments and face increasingly sophisticated security threats, traditional methods of identity assurance are showing their limits.

Passwords can be stolen. Cards can be shared. Manual checks do not scale. Facial recognition, when implemented correctly, offers a friction-light alternative that fits modern operational realities.

However, becoming “boring” does not mean becoming invisible or unaccountable. On the contrary, as facial recognition becomes foundational, expectations around reliability, accuracy and governance rise sharply.

Businesses adopting FRT in 2026 will no longer be able to treat it as a specialist tool managed in isolation by IT teams. It will sit alongside core systems, subject to the same scrutiny as any other business-critical technology.

Expect scrutiny – embrace it

Scrutiny is a key word here. And this is where recent public debate offers important lessons.

The UK police have been using FRT with greater frequency. For instance, the Met Police announced in January that more than 100 wanted criminals were arrested by the Metropolitan Police within the first three months of a pioneering Live Facial Recognition (LFR) pilot in Croydon.

Meanwhile, Home Secretary Shabana Mahmood promptly followed the news by saying they the government would be going ahead with plans to expand facial recognition.

Yet this story came as campaigners have been appearing in the High Court, saying that the technology is expanding without adequate safeguards.

The debate in the policing sector mirrors broader concerns around privacy, bias and accuracy. They are not necessarily obstacles to adoption for businesses across other sectors, but they are certainly signals of where the FRT industry must continue to improve.

Facial data is inherently sensitive, and its use demands higher standards of protection and transparency. For business leaders, this means moving beyond a narrow focus on technical performance and considering the broader implications of deployment.

Accuracy, for example, is not a static metric. Performance can vary depending on lighting, camera quality, demographic diversity and operational context.

Organizations must understand that responsible deployment requires ongoing testing and monitoring, not one-off validation. Similarly, privacy cannot be bolted on after implementation. Principles such as data minimization, clear purpose limitation and secure storage need to be built into systems from the outset.

Trust must still be earned

Trust is another critical factor as facial recognition becomes mainstream. In enterprise environments, trust extends beyond end users to employees, partners, regulators and investors.

Businesses need to be able to explain why facial recognition is being used, what safeguards are in place, and how risks are managed. Transparency and accountability will increasingly differentiate responsible adopters from those who treat FRT as a black-box solution.

There is also a strategic dimension to this transition. As facial recognition becomes part of everyday operations, it shifts from being a purely technical decision to a business one.

Boards and senior leaders must understand how biometric technologies fit within their organization's risk framework, data governance strategy and long-term digital roadmap. In this sense, the “boring” phase of facial recognition is also the most demanding.

In 2026, the organizations that succeed with facial recognition will be those that treat it with the same discipline applied to other mature technologies.

They will select solutions based not only on capability, but on compliance, resilience and ethical design. They will involve legal, security and data protection teams early, rather than as an afterthought. And they will recognize that earning trust is an ongoing process, not a one-time achievement.

Facial recognition is no longer science fiction, and it is no longer experimental. Its transition into everyday business technology is well underway. So, the challenge is not simply whether it works, but how responsibly it is used.

If organizations get that right, facial recognition will fade into the background of daily operations – not because it is insignificant, but because it is reliable, well-governed and fit for purpose. And that, ultimately, is what technological maturity looks like.

We've featured the best privacy app for Android.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

As enterprises integrate AI agents into their workflows, a silent shift is taking place.

Security controls built on static access policies designed for predictable behavior are colliding with systems that reason instead of simply executing. AI agents, driven by outcomes rather than rules, are breaking the traditional identity and access management model.

Consider a retail company that deploys an AI sales assistant to analyze customer behavior and improve retention. The assistant doesn’t have access to personally identifiable information, it’s restricted by design.

Yet when asked to “find customers most likely to cancel premium subscriptions,” it correlates activity logs, support tickets, and purchase histories across multiple systems. This generates a list of specific users inferred through behavior patterns, spending habits, and churn probability.

No names or credit cards were exposed, but the agent effectively re-identified individuals through inference, reconstructing sensitive insights that the system was never meant to access and potentially exposing personal identifiable information (PII).

While it didn’t break access controls, it reasoned its way around systems to access information that it was not originally scoped to access.

When Context Becomes the Exploit

Unlike traditional software workflows, AI agents don’t follow deterministic logic; they act on intent. When an AI system’s goal is “maximize retention” or “reduce latency,” it makes autonomous decisions about what data or actions it needs to achieve that outcome. Each decision might be legitimate in isolation, but together, they can expose information far beyond the agent’s intended scope.

This is where context becomes an exploit surface. Traditional models focus on who can access what, assuming static boundaries. But in agentic systems, what matters is why the action occurs and how context changes as one agent invokes another. When intent flows across layers, each reinterpreting the goal, the original user context is lost and privilege boundaries blur.

The result isn’t a conventional breach; it’s a form of contextual privilege escalation where meaning, not access, becomes the attack vector.

Shortcomings of Deterministic Controls

Most organizations are learning that traditional RBAC (Role-Based Access Control) and ABAC (Attribute-Based Access Control) models can’t keep up with dynamic reasoning. In classical applications, you can trace every decision back to a code path. In AI agents, logic is emergent and adaptive. The same prompt can trigger different actions depending on environment, prior interactions, or perceived goals.

For example, a development agent trained to optimize cloud computing costs might start deleting logs used for audit purposes or backups. From a compliance perspective, that’s catastrophic, but from the agent’s reasoning, it’s efficient. The security model assumes determinism; the agent assumes autonomy.

This mismatch exposes a flaw in how we model permissions. RBAC and ABAC answer “Should user X access resource Y?” In an agentic ecosystem, the question becomes “Should agent X be able to access more than resource Y, and why would it need that additional access?” That’s not an access problem; it’s a reasoning problem.

The Rise of Contextual Drift

In distributed, multi-agent architectures, permissions evolve through interaction. Agents chain tasks, share outputs, and make assumptions based on others’ results. Over time, those assumptions accumulate, forming contextual drift, a gradual deviation from the agent’s original intent and authorized scope.

Imagine a marketing analytics agent summarizing user behavior, feeding its output to a financial forecasting agent, which uses it to predict regional revenue. Each agent only sees part of the process. But together, they’ve built a complete, unintended picture of customer financial data.

Every step followed policy. The aggregate effect broke it.

Contextual drift is the modern equivalent of configuration drift in DevOps, except here, it’s happening at the cognitive layer. The security system sees compliance; the agent network sees opportunity.

Governing Intent, Not Just Access

To address this new class of risk, organizations must shift from governing access to governing intent. A security framework for agentic systems should include:

Intent Binding: Every action must carry the originating user’s context, identity, purpose, and policy scope throughout the chain of execution.

Dynamic Authorization: Move beyond static entitlements. Decisions must adapt to context, sensitivity, and behavior at runtime.

Provenance Tracking: Keep a verifiable record of who initiated an action, which agents participated, and what data was touched.

Human-in-the-Loop Oversight: For high-risk actions, require verification, especially when agents act on behalf of users or systems.

Contextual Auditing: Replace flat logs with intent graphs that visualize how queries evolve into actions across agents.

Why Permissions Alone Are Flawed

Static permissions assume identity and intent remain constant. But agents operate in fluid, evolving contexts. They can spawn sub-agents, generate new workflows, or retrain on intermediate data, actions that continually redefine “access.”

By the time an identity system detects a security incident, a violation or breach has already occurred without a single permission being broken. That’s why visibility and attribution must come first. Before enforcing policy, you must map the agent graph: what exists, what’s connected, and who owns what.

Ironically, the same AI principles that challenge our controls can help restore them. Adaptive, policy-aware models can distinguish legitimate reasoning from suspicious inference. They can detect when an agent’s intent shifts or when contextual drift signals rising risk.

I tried 70+ best AI tools.

Sovereign cloud has become one of the most heavily used terms in enterprise technology.

Almost every major provider has gone to great lengths to market their version of it, promising that data is ‘residing locally’ or ‘ringfenced in Europe’, but the key question remains: what does true cloud sovereignty actually mean?

For governments, regulators and businesses handling sensitive workloads, distinguishing between what’s marketed as sovereign and what is sovereign in reality is vital.

The dominance of US hyperscalers in Europe makes this challenge even more pressing. Amazon Web Services, Microsoft Azure and Google Cloud together control more than two-thirds of the region’s cloud computing market, raising concerns about jurisdictional control and limits of so-called European cloud offerings.

As demand for sovereignty and resilience grows, the question is how Europe can build a more balanced and independent cloud ecosystem.

Defining true sovereignty

Sovereign cloud is not only a question of where data physically resides, but also of who has legal authority over it and the dependencies associated with it, including technology, supply chains and vendor lock-in, which may either enable or constrain freedom of action as sovereignty is understood.

Data residency answers the ‘where’ question, sovereignty addresses the ‘who’ and “up to which point”. This distinction is crucial when considering the different aspects of sovereignty. Take the question of legal authority for instance, in light of legislation such as the US CLOUD Act and Section 702 of the Foreign Intelligence Surveillance Act (FISA).

Both pieces of legislation permit US courts and agencies to require US-headquartered firms to provide access to data related to investigations via warrants or other processes, even when hosted abroad.

In practice, this means that a European bank, healthcare provider or government department relying on an American cloud operator may not easily determine whether the data of some of its European customers has been provided to US law enforcement or intelligence agencies, even if the data resided outside the US.

For organizations entrusted with sensitive information, that exposure is not theoretical. It is a live compliance and trust issue.

True sovereignty therefore requires more than local hosting. It demands that both the infrastructure and the jurisdiction are aligned with the customer’s own legal environment.

It also requires interoperability and portability across cloud environments, allowing organizations to choose where and how workloads run without being locked into a single provider.

It also requires transparency on the underlying technology and supply chain, as well as the ability to risk manage and make choices that reduce dependencies or concentration.

The lingering questions

This is why there are still many questions surrounding the sovereign offerings from global hyperscalers. Providers have all launched initiatives that stress enhanced European control or partnerships with local entities.

Yet because the parent companies remain subject to US law, customers remain concerned that there is a jurisdictional gap. Put simply, a sovereign wrapper around non-sovereign foundations does not fully resolve the issue in every case.

Customers may achieve greater assurances about data location or operational independence, but unless the operating entity is legally insulated from foreign jurisdiction and enables customers to make choices, the claim of sovereignty remains partial.

For critical workloads, such as those in public sector, regulated industries and AI applications, relying on US-controlled clouds introduces operational and compliance risk that cannot be fully mitigated by local hosting alone.

Why it matters now

The speed of market growth makes clarity urgent. The global sovereign cloud market size is projected to grow from USD 154.69 billion in 2025 to USD 823.91 billion by 2032. Europe alone accounted for around 37% of the global market in 2024.

This growth reflects rising demand for secure, trusted environments, particularly in Europe, where regulatory frameworks such as DORA, the GDPR and Data Act emphasize local control, risk management, supply chain transparency and concentration risk.

The European Union, for instance, has made digital sovereignty a strategic priority, while countries like Germany and the UK are exploring frameworks to ensure their critical data assets cannot be subject to overseas legal claims. The direction of travel is clear, sovereignty must be defined and enforced, not assumed.

Clear standards and stronger ecosystems

What’s missing today is a consistent framework defining what constitutes a sovereign cloud.

EU Member States have created cloud certification schemes like C5 in Germany and SecNumCloud in France that contain sovereignty criteria. DGIT, the European Commission IT service, has made a notable attempt in the context of procurement to define sovereignty in the form of a scale of requirements.

All these attempts, while welcome, demonstrate the fragmentation of the EU market. Customers are often left to navigate competing claims and complex technical language without clear standards for comparison.

A truly sovereign cloud should guarantee that data is controlled, accessed and governed exclusively within the jurisdiction of the customer.

Achieving this does not mean retreating from global innovation. It requires enabling local providers to deliver services that meet sovereignty criteria without compromise. European cloud service providers such as Redcentric and ANS are well positioned to fulfil this role.

By operating under local legal and compliance frameworks and investing locally, they can give organizations genuine control over their data. In many cases, this control is best realized through private cloud environments, where infrastructure, governance and operational authority can be aligned directly to sovereignty requirements.

Technology vendors have a role to play in supporting this ecosystem. By providing the platforms, infrastructure software, and interoperability frameworks, they can empower local providers without becoming operators themselves.

This distinction matters. It avoids entanglement with foreign jurisdictions while fostering an environment where sovereignty is embedded by design rather than bolted on afterwards.

A sovereign-by-design future

As the market evolves, the focus shifts from whether sovereign cloud exists to whether it truly meets customers’ needs. Local hosting and compliance claims must be tested against jurisdictional control. Cloud services must be designed for sovereignty from the ground up, with governance structures aligned to the data they protect.

In the end, the debate over sovereign cloud is about more than technology. It reflects broader questions of trust, independence, skill development, economic growth, resilience and control in the digital economy. For businesses and governments alike, cutting through the hype will be essential to ensure sovereignty is real, not rhetorical.

We've featured the best identity theft protection.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

Agentic AI is the latest buzz in boardrooms. Unlike generative AI tools, agentic AI act as autonomous agents that can reason, make decisions, and act across workflows to achieve goals. Done right, they promise to reduce manual work and unlock new levels of productivity.

But many early adopters of AI tools are struggling. Pilot projects stumble, costs escalate, and results fail to match expectations. The problem isn’t that agentic AI is overhyped, it is that businesses are moving too fast without the strategy, infrastructure, and data foundations required to make them work as intended.

And this isn’t surprising when you consider that 80% to 90% of all enterprise data is unstructured -- based on multiple analyst reports in recent years.

As someone who has built platforms through multiple waves of ‘intelligent automation,’ I’ve seen firsthand the same repeated patterns: technology alone doesn’t transform organizations -- alignment, governance, and cultural readiness do. The real breakthrough comes when innovation is grounded in trust and connected to business outcomes.

Where conventional AI might sort invoices, an agentic AI could approve payments, flag anomalies, and update compliance systems. That leap demands a contextual understanding of how data, processes, and rules fit together.

Too many organizations are treating agentic AI as bolt-on upgrade, as if they’re simply more advanced chatbots. The reality is more complex: agentic AI needs to be woven into the enterprise fabric, connected to the right data and workflows, and supported by governance. Without that foundation, autonomy quickly becomes chaos.

Infrastructure first

One of the biggest stumbling blocks is infrastructure. Many enterprises still run on siloed content repositories, legacy systems, and fragmented integrations. In these environments, agentic AI can’t access the full unstructured data they need to perform at their best.

In government, for example, content and processes are spread across different agencies, often using decades-old applications. Asking an AI agent to make decisions without integrating those systems is like asking it to assemble a puzzle with half the pieces missing.

Preparing for agentic AI requires investing in cloud-native foundations and interoperable content platforms that unify information and enable seamless connections across applications. Without this groundwork, agentic AI risk acting on partial or outdated information, and making flawed decisions as a result.

Bad data kills autonomy

Even with the right systems in place, poor data quality is a critical flaw. Agentic AI thrives on complete, accurate, and governed information. If datasets are inconsistent or scattered, agentic AI can’t make sound decisions.

Healthcare illustrates this challenge clearly. An agent supporting clinicians must pull from medical histories, lab results, and imaging data in real time. If one piece is missing or misaligned, the recommendations these agentic technologies produce could be flawed.

The lesson for early adopters is clear: start with a data audit and gain a firm understanding of where your unstructured data is. Know what you have, where it lives, and how it’s governed before handing decision-making power to AI.

Getting governance right

Another misconception is that agentic AI removes people from the loop. In reality, the most effective early use cases blend autonomy with oversight.

Take financial services. Agentic AI may verify documents and draft compliance reports, but humans still make the final call on high-risk cases, or how to proceed when a document is flagged by an agent. This balance accelerates workflows without eroding trust and accountability.

Strong governance must be embedded from the outset, covering regulation, ethics, and operational control. Without it, these agents risk amplifying bias, undermining trust, and exposing organizations to compliance failures.

Lessons from early adopters

The experiences of early adopters reveal three clear lessons.

The first, projects work best when they begin with a clear business outcome, not a fascination with the technology or jumping on a trend. Organizations that take time to define the processes they want to improve and the results they need to achieve are the ones seeing value.

Second, they invest early in the groundwork. Modern infrastructure and clean data may not grab headlines, but they are essential to making the headline-grabbing innovations possible.

And finally, they treat autonomy as something to scale gradually. The most effective implementations begin with human-in-the-loop models and only expand to greater autonomy once confidence and maturity grow. This approach builds trust in the technology while maintaining accountability.

These early lessons are already shaping a picture of maturity.

The shape of maturity

As agentic AI matures, it will move beyond isolated experiments and towards interconnected systems. The real breakthrough will come from agentic AI networks coordinating across workflows.

In a hospital, for example, one agent might surface patient histories, another manages scheduling, and a third flag billing issues; all contributing to a shared context that supports clinicians.

Proof-points will become a non-negotiable. Businesses will expect agents to show their work, like the data they used, the reasoning they followed, and the compliance checks they applied. Without this transparency, agentic AI won’t be trusted to handle sensitive or high-value work.

And the technology landscape itself will have to open up. Organizations will want the flexibility to integrate agentic AI powered by different models, switch providers as needs evolve, and scale across hybrid or multi-cloud environments. Flexibility and interoperability will be essential to protect long-term investments.

Beyond the hype

Far from failing, agentic AI is in its adolescence. Just as cloud computing went through a difficult transition phase before proving indispensable, agents too will require a period of adjustment.

The organizations that succeed will be those that prepare best, not adopt the fastest. By aligning strategy, modernizing infrastructure, cleaning data, and embedding governance, enterprises can move from experimentation to transformation.

With the right foundations, agentic AI can do far more than just automate tasks. It will enable genuinely intelligent systems that reshape how work gets done – and that could be the most significant shift in enterprise technology for a generation.

We've featured the best AI website builder.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

Cloud computing remains an increasingly important part of business and personal internet use - and with the rise of AI and the increasing need for huge amounts of data, cloud storage is in demand now more than ever.

There will be shifts in trends, as there always are, which will bring both complications and opportunities. For 2026, there are already huge investments lined up into hyperscaler capacity, with over $1 trillion promised into data centre construction by 2030 - but there is equally a push towards data sovereignty and geopatriation.

AI is undoubtedly changing the cloud landscape, with vast amounts of data and compute needed to train and run Large Language Models (LLMs). Bigger, non-proprietary, and all purpose models are less secure and less reliable for enterprises on the whole, so smaller, often locally stored models are becoming more popular - which means on-prem storage and compute will become once again more significant.

We spoke to a few cloud storage experts to find out what their predictions are for the cloud storage space going into 2026, and these are our top picks.

Sovereign cloud solutions

An overwhelming response from experts was in regards to the importance of cloud sovereignty. One of the biggest lessons from 2025 has been how the convenience and scalability of hyper-scalers, does not always necessarily outweigh the risks of security and outages.

“Demand for sovereign cloud solutions will rise, along with greater reliance on regional providers and renewed interest in on-premises or air-gapped data centers,” predicts Chintan Patel, Cisco EMEA CTO.

“A full overhaul of global infrastructure is unlikely, but selective migrations and diversified cloud strategies will become the norm, creating demand for local talent and skills.

Downtime caused historic issues in 2025, with the AWS incident in particular outlining the dramatic losses that come with giving up control of your information infrastructure. We’ve already seen a stagnation in the adoption of non-sovereign cloud storage, and the experts we’ve heard from seem to predict this will continue.

“Make no mistake: this is a paradigm shift," explains Emma Dennard, Vice President, Northern Europe at OVHcloud

“Organizations are realising that not having sovereign cloud - for certain workloads and applications – is an operational risk. As a result, more organizations are building cloud-native, portable systems by default. This means that moving apps between clouds is easier than ever, risks can be mitigated and costs can be optimized.”

Hyper-scaler hype cycle

It’s difficult to discuss predictions for cloud computing without mentioning AI. Cloud and AI are intertwined, and the dramatic increase in demand for cloud services is directly linked to the enormous amount of data and compute needed to run Large Language Models that have soared in popularity.

Most experts can recognize that AI (particularly Gen AI) is at the peak of the hype cycle right now, and that this level of general use and discourse is unlikely to continue long term.

But, that doesn't mean the technology won't continue on, explains CEO of Pulsant, Rob Coupland;

"As the hype settles, businesses are starting to evaluate real-world AI uses and determine what digital infrastructure is truly needed to support their AI goals. This also brings inference AI and sovereign AI into the picture, further complicating the landscape, with Edge computing emerging as a key beneficiary."

"Although hyperscale demand will no doubt continue, demand for specialised, inference-optimized storage platforms will become more significant."

The AI Bubble

There have been talks of an AI bubble for months, with experts warning of a ‘sharp market correction’ when this bursts. Considering a huge portion of the market, and indeed of the US economic growth can be attributed to artificial intelligence, the concerns that arise from the possibility of the bubble bursting are significant.

But, the financial concerns are not the only consideration. AI consumes huge amounts of power, and with energy sources across the globe already struggling to meet rising demand, this poses serious dangers for 2026 and beyond.

“Investment has poured into new data center developments to power global AI ambitions, but the energy systems required to support them are on their knees,” says Taco Engelaar, SVP and Managing Director of Neara.

“To increase capacity for new data centers, policymakers are proposing extensive grid expansion; but if history is anything to go by, public opposition could stop it in its tracks.”

Looking forward

There are lots of interesting predictions to take away for 2026, some positive, some a little more skeptical. What's certain is that cloud computing will be an integral part of enterprise moving forward, whether in a global or local capacity.

Increased demands for data and power brought about by AI are likely to strain public services and energy grids, with global outages and security incidents driving organizations towards local infrastructure with more transparency, controls, and stricter compliance.

• Nvidia has provided native GeForce Now support for Linux PCs and Fire TVs
• This comes amid the crisis causing high prices for GPUs and RAM kits
• It's another case of Nvidia further supporting Linux and ultimately Valve's SteamOS

CES 2026 is here, and while Nvidia hasn't made any new GPU announcements (as very much expected due to the ongoing RAM crisis), it has introduced new DLSS 4.5 upscaling technology for RTX GPU owners, and another big update for ideal for gamers without PCs.

Both Linux PCs and Amazon Fire TV sticks now have native Nvidia GeForce Now (GFN) support, allowing more gamers without gaming PCs (or with lesser-powered rigs) to dive in and enjoy RTX 4080 and RTX 5080-level frame rates (in select titles) when using the Ultimate membership, as announced at CES 2026.

This comes months after Nvidia delivered native support for GeForce Now on Valve's Steam Deck last year, providing an easier way for SteamOS handhelds to dive into GFN gaming without any convoluted setup. As of now, this has expanded to all Linux systems, which will help significantly for non-Steam Deck handhelds, and Linux laptops or desktop PCs.

What's arguably more important, though, is the native support for Fire TV sticks. This could prove a great alternative for accessing GFN and the world of PC gaming, especially given that the PC hardware market is in such a tough spot of late. You can't have failed to notice the skyrocketing prices of RAM and also storage to an extent.

I was never a big fan of the idea of cloud gaming until I used GFN in late 2025 when the RTX 5080 was officially added to the Ultimate tier of Nvidia's streaming service. To put it simply, I was surprised at how well GFN worked on my Lenovo Legion Go S handheld, with a commendably low latency and great image quality.

With native support for Amazon Fire TV sticks, this is now one of the cheapest ways to game via GFN in a no-hassle fashion. Of course, your gaming experience will still depend on the strength of your internet connection, which will need to be fast enough to ensure stable and smooth frame rates.

Analysis: Linux getting more attention from Nvidia is great news

With the Steam Machine theoretically on track for a potential launch early this year, GFN native support for Linux comes at the perfect time. It gives buyers of Valve's mini PC another option in terms of extra horsepower for games (internet connection dependent, of course).

Furthermore, Nvidia's recent focus on GFN support for SteamOS and Linux in general is hopefully a good indication of what's to come.

One of the biggest hopes for the Steam Machine is that it encourages developers and publishers to provide better support for SteamOS. This would include better anti-cheat support for more games, and also better driver support from Nvidia (particularly with DX12 games), and more native applications like Discord that can function fully in game mode.

It's a step in the right direction, and definite praise goes to Nvidia on this occasion – long may that continue.

TechRadar will be extensively covering this year's CES, and will bring you all of the big announcements as they happen. Head over to our CES 2026 live news page for the latest stories and our hands-on verdicts on everything we've seen.

You can also ask us a question about the show in our CES 2026 live Q&A and we’ll do our best to answer it.

And don’t forget to follow us on TikTok for the latest from the CES show floor!