Easemate.ai launched in 2025 with a simple pitch: one platform for everything AI.

It doesn't make you choose between a chat assistant, an image generator, or a video tool. You get all three, alongside study utilities, document readers, and image editing features. The range of supported models is equally wide, covering GPT, Gemini, Claude, DeepSeek, Grok, Kimi K2, and Qwen 3 on the chat side alone.

The creative side is where things get particularly ambitious. Easemate integrates image models including Nano Banana, Midjourney, Flux Kontext, GPT-4o, and Seedream, with a video catalogue stretching to Sora 2, Google Veo 3, Kling, Seedance, and Runway. Few platforms at this price point give you that many models in one place.

We've been reviewing B2B software and AI platforms at TechRadar Pro since 2012. Easemate sits in a crowded but useful category of multi-model AI aggregators that we've tracked closely. You can also check out our AI tool roundup for 2026 and deep dives into platforms like OpenClaw or Moltbook.

What is Easemate.ai?

Easemate.ai is a web-based AI platform that consolidates multiple AI models and task-specific tools into a single subscription.

Rather than routing you to one underlying model, it lets you switch between GPT, Gemini, Claude, DeepSeek, and others depending on what you need, without juggling separate accounts.

The platform divides its offering into four main areas: AI Chat, AI Study & Research, AI Photo, and AI Video. Within those, you'll find tools for ChatPDF, document summarization, math and science solvers, flashcard generation, image-to-video conversion, and YouTube transcript extraction.

It targets a broad user base (students, solo creators, freelancers, and small businesses) rather than positioning itself as a developer tool or enterprise solution. If you want a single dashboard that covers daily AI tasks without managing multiple subscriptions, Easemate's pitch is worth considering.

Easemate.ai: At a glance

Buy it if…

• You want multi-model AI chat without juggling accounts. Easemate puts GPT, Gemini, Claude, DeepSeek, and Grok 4 in one place, which saves real time if you regularly compare outputs across models.
• You need creative tools alongside a chat assistant. The combination of image generators and video models in one subscription is hard to match at Easemate's price point.
• You're a student or researcher on a budget. The free tier includes 30 sign-up credits, daily check-in bonuses, and up to 200K AI chat tokens per day.

Don't buy it if…

• You need consistent professional image or video output. Users report a "prompt drift" issue where the platform ignores specific instructions, alters faces, or changes scenes unexpectedly.
• Reliable customer support matters to you. Support is online-only and has drawn criticism for being slow to resolve credit-related problems.
• You want granular model control. Access to Veo 3 and Sora 2 comes through Easemate's own interface rather than direct API access, which limits parameter customization.

My time with Easemate.ai

Getting started on Easemate.ai is frictionless. The platform lets you try select tools without an account, and signing up takes under a minute. Once logged in, the 30 free credits appeared immediately, and the interface guided me clearly toward the main tool categories. For a platform with this many features, the navigation stayed surprisingly tidy.

Where I hit friction was in creative generation. I ran several image prompts through Nano Banana and Flux Kontext and found outputs solid roughly two-thirds of the time. There were noticeable cases where the platform deviated from my descriptions, and rerunning the same prompt sometimes produced very different results. Video tools showed similar inconsistency.

The value case is real at the Lite tier, though. For $8.90 in the first month, rising to $9.90 after that, you get 1,200 credits, access to up to 120 images and 60 videos per month, and multi-model AI chat. That's a fair deal for casual creative work or students managing multiple AI tasks, as long as you aren't expecting the precision of a dedicated tool.

Easemate.ai: Features

The AI Chat section covers GPT (multiple versions), Gemini, Claude 3 Haiku, DeepSeek, Grok 4, Kimi K2, and Qwen 3. For most conversational tasks (drafting, translating, summarizing), having that range in one tab is useful. The daily free token limit of 200K is also more generous than most comparable platforms.

The study and research tools are well-executed and clearly the original backbone of the platform. Math, physics, and chemistry solvers work step-by-step, making them practical for students rather than just returning a final answer. Flashcard and quiz generators, mind maps, and AI Scholar round out a toolkit that serves academic workflows more carefully than most multi-purpose AI platforms do.

On the image side, the model selection is broad: Nano Banana, Midjourney, Flux Kontext, GPT-4o, and Wan 2.5 are all accessible on paid plans. Nano Banana produces good commercial-style images; Flux Kontext handles text-in-image prompts reasonably well. The consistency problem persists, particularly with prompts involving specific faces or complex scenes.

Document tools perform well. ChatPDF, Chat Doc, and Chat PPT let you upload files and query them conversationally, with OCR support for scanned content. The YouTube summarization tool is a genuine highlight: paste a link and get structured notes with timestamps, which worked better than expected in testing.

Easemate.ai: User experience

The interface is clean and well-organized. Tool categories sit in a top navigation bar, each expanding into a dropdown with clearly labeled options. First-time users shouldn't need a tutorial to find their way around, and the browser-based experience works consistently across devices.

The credit system is where the UX gets murky. Different tools consume different credit amounts, and it's not always clear how many you're spending before you generate. A failed generation still costs credits, which user reviews flag repeatedly. Easemate's team has acknowledged this in public responses, but the system itself hasn't changed.

Easemate.ai: Customer support

Easemate offers support via email at support@easemate.ai and through a help portal on the website. There's no live chat, phone line, or priority tier for paid subscribers. Documentation covers pricing and credits at a surface level but doesn't go deep enough for troubleshooting edge cases.

User perception is mixed. As of early 2026, Easemate held an overall rating of 2.0 to 3.0 out of 5 stars with most review aggregators, with positive feedback on ease of use offset by complaints about reliability and support responsiveness. More recent reviewers show a wider range of experiences — solo creators praise the video output quality, while others report credits consumed by failed generations with no satisfying resolution.

Easemate.ai: Pricing

• Basic (free): 30 sign-up credits, daily check-in bonuses, up to 200K AI chat tokens per day, limited image and video access
• Lite ($8.90/month intro, then $9.90/month): 1,200 credits, up to 120 images and 60 videos per month, full model access; annual billing drops this to $7.49/month intro, then $8.49/month
• Pro ($19.90/month intro, then $24.90/month): 3,000 credits, up to 300 images and 150 videos; annual billing drops to $16.90/month intro, then $20.9/month

The free tier is actually usable for light AI chat and occasional image generation — it's not a locked demo. Lite is the sweet spot for individuals. Easemate also sells one-time credit packs that never expire, ranging from 500 credits at $4.90 up to 15,000 credits at $104.90, with discounts of 10–30% depending on bundle size. There's no API access or developer-tier pricing.

Easemate.ai alternatives you should consider

• ChatGPT Plus: At $20/month, you get GPT-4o, o4-mini, and DALL-E 3 in a more mature, reliable environment. A better choice if text generation and image creation are your primary needs.
• Perplexity Pro: Covers multi-model chat with web search grounding and document uploads. Weaker on creative generation but more dependable for research-heavy workflows.
• Adobe Firefly: Produces commercially safe, high-consistency image output with better prompt fidelity than Easemate's image tools, though it lacks the platform's broad AI chat and video coverage.

How I tested Easemate.ai

• Tested AI chat tools across GPT, Gemini, Claude, and DeepSeek for writing, summarizing, and translating tasks, comparing output quality and response speed across models.
• Generated images and video clips using Nano Banana, Flux Kontext, Kling, and Veo 3 with a range of prompt types to assess consistency and prompt adherence.
• Used document and study tools, uploading multi-page PDFs and PowerPoint files to test ChatPDF accuracy, and running math and science problems through the step-by-step solvers.

My testing involved hands-on use of Easemate.ai across its four main tool categories over several sessions, combined with a review of third-party user feedback on Trustpilot and review platforms to benchmark real-world reliability against my own observations. Pricing details were verified directly against the official Easemate.ai pricing page.

• Rotary Mouse replaces traditional mouse wheel with continuous motion
• Device claims faster navigation through documents and spreadsheets
• Users report smoother timeline scrubbing in editing software

The scroll wheel on a mouse has barely changed in three decades, and most people have stopped questioning whether it still works well.

However a device called the Rotary Mouse now argues it lets users move through documents, spreadsheets, and timelines at speeds that ordinary flicking simply cannot match.

The product's pitch centers on a single mechanical swap, replacing the familiar up-and-down wheel motion with a continuous rotary dial that the maker says feels closer to turning a knob than clicking a switch.

A Different Motion, A Different Claim

According to the campaign, the rotary input allows users to scroll or scrub through content up to 2.5 times faster than a standard mouse wheel, while also reducing the repetitive strain associated with constant flicking.

Instead of short up-and-down movements, users rotate the wheel like a small knob, controlling speed and direction through pressure and motion.

The device still supports conventional vertical scrolling, meaning users can switch between familiar input and rotary control without changing devices or habits.

It measures 119 mm by 64 mm by 40 mm in a matte black, ergonomic shell, weighs just 59g without its battery, and supports 2.4 GHz wireless connectivity.

The Rotary Mouse also features an optical sensor with switchable 800, 1200, and 1600 DPI sensitivity settings, allowing users to fine-tune cursor speed and precision.

Its left and right buttons are silenced through a middle button built into the rotary wheel itself.

The device ships with one AAA alkaline battery and a USB wireless receiver, and works across Windows, Linux, Mac OS, and Android.

Some testers describe the Rotary Mouse’s motion as more fluid and easier to control, especially when trying to stop at precise points on a timeline or page.

The design also introduces tactile feedback through clicks during rotation, which is intended to help users maintain control during faster movement.

Early use cases and mixed reactions

Early adopters have tested the device in video editing software where timeline scrubbing is a frequent task requiring fine control and repeated movement.

In programs such as Adobe Premiere Pro and DaVinci Resolve, the Rotary Mouse is used as a scrubbing tool, where turning the dial allows users to move through video timelines more smoothly than repeated scrolling

In simulation racing setups, the same rotary input can be mapped directly to steering control, turning the dial into a compact alternative to traditional wheels or keyboard-based steering.

This makes it useful in games such as Euro Truck Simulator 2, Assetto Corsa, and BeamNG.drive, where gradual steering inputs are required and can be replicated through controlled rotation.

The concept has also gained attention online, with thousands of upvotes across PC enthusiast communities discussing alternative input devices and ergonomic design ideas.

The creator, Melvin Wong, an electronics engineer with long experience in hardware development, says the idea came from reducing finger strain during long computer sessions.

Prototypes were built using 3D printing techniques before evolving into early production units.

He claims that continuous rotation reduced repetitive strain while unexpectedly improving navigation speed across large digital workspaces during testing phases.

However, independent verification of the claimed speed improvements remains limited, and real-world performance likely depends on user behaviour and application type.

The Rotary Mouse is currently in crowdfunding on Kickstarter, where it has raised $3,654 against a $14,794 goal from 61 backers, with 18 days left before the campaign ends.

The team also says it has already sold more than 30 pre-production Founder’s Edition units and is aiming to begin shipping the final product by December 2026.

Disclaimer: We do not recommend or endorse any crowdfunding project. All crowdfunding campaigns carry inherent risks, including the possibility of delays, changes, or non-delivery of products. Potential backers should carefully evaluate the details and proceed at their own discretion.

• Stduy claims AI adoption outpaces governance, creating long-term code maintainability challenges
• Saved developer time is now being spent on reviewing, validating and governing AI
• Accountability and trust are now more important than speed and productivity

Even though 91% of organizations have two or more AI coding tools in active use, four-fifths (79%) of the more than 1,500 developers surveyed by GitLab believe software delivery hasn’t accelerated at the same pace as developer productivity, implying challenges along the workflow could be diminishing returns.

Around three in four believe developers are writing code faster (78%) and producing higher-quality code (73%), but a new report from the coding platform believes there’s much more to AI than speed alone.

GitLab describes this as an ‘AI paradox’, where developer time is being taken up reviewing, validating and governing AI despite its promised productivity impacts.

Is AI just shifting problems downstream?

In fact, 85% agree that the biggest constraint now is code review and validation, rather than code creation, proving that problems have simply been moved downstream instead of removed entirely.

But now that AI is deeply engrained within developer workflows, two in five (43%) now struggle to distinguish AI-generated code from human-written code, making it difficult to maintain security and quality in the long term. Three in four (73%) say they’re concerned about the long-term maintainability of AI code.

And it’s this visibility that’s causing one of the biggest headaches for coders, with a third (34%) now unable to determine whether AI-generated code played a role in an incident.

“The events of the past few months, including supply chain attacks, reliability issues, and regulators tightening expectations around AI traceability and provenance are making clear that speed without control is a liability, not an advantage,” Chief Product and Marketing Officer Manav Khurana commented.

Most companies (92%) now experience some form of governance challenges when it comes to vibe coding, and four in five admit they’ve adopted AI coding tools faster than they’ve implemented governance policies.

Trust to become a key differentiator – not speed

While coding has become a major use case for generative AI since its mainstream introduction in late 2022, policies are still falling behind. But developers know this, and 91% now plan to invest in governance over the next year, with 98% allocating budget specifically for this.

Looking ahead, developers are preparing to enter a new era of AI, where generating code is no longer a priority. Instead, it’s all about governance.

“Organisations that will ship trusted software faster are the ones building the foundations of accountability with context, traceability, and governance baked into the platform, not just bolted on after the fact,” Khurana added.

But reading between the lines, it’s been clear that developers haven’t been entirely comfortable with AI for some while. Stack Overflow’s 2025 developer survey found that many are turning their backs on the tech due to privacy, pricing and quality reasons. Nearly 46% distrust AI to some degree, compared with just 33% who trust it.

Developers have now reached a point where AI has very clear use cases, but GitLab’s report clearly shows that traceability, accountability and trust will become competitive advantages in the future.

• This portable display combines work and entertainment in a single compact device
• Connectivity hub aims to replace traditional laptop docking stations entirely
• InnLead Monhub’s 60Hz refresh rate may limit the fast-paced gaming experience quality

InnLead Monhub has launched a 15.6-inch FHD 1080P portable screen hub designed for multi-device use across work, travel, and entertainment contexts.

This monitor uses a 1920 x 1080 resolution at 141 PPI with LCD IPS technology for general productivity and media consumption on compact portable setups.

It combines IPS panel visuals, 60Hz refresh rate, 280 cd/m² brightness, and 1000:1 contrast for mixed productivity and media tasks usage under typical indoor conditions.

Connectivity and display structure under scrutiny

The InnLead Monhub claims to turn any workspace into a flexible setup with connectivity that includes HDMI 2.0, USB 3.0, dual Type-C, USB, and TF or microSD card support.

The brightness of the display reaches between 280 cd/m² and 250 cd/m², alongside 1000:1 contrast and 16.7M color support for varied viewing conditions.

This portable monitor supports plug-and-play connections, allowing laptops, consoles, and smartphones to extend displays without the complex setup steps required.

Marketed partly as a business monitor, the device emphasizes remote productivity, yet reliance on a single 60Hz panel may limit more demanding workflows.

It comes with a light travel weight of around 730g, although the 1075g standard weight may still feel noticeable inside compact bags.

Crowdfunding uncertainty and use case debate

The InnLead Monhub is currently on crowdfunding through Indiegogo, a stage where promised specs and final retail hardware often diverge.

SmartDock, the branding behind the device, pitches one screen as a replacement for separate laptop docks, console adapters, and phone stands.

Ava InnLead, the project's creator, frames the pitch around mobility, though that promise has not been tested outside the company's own marketing.

“With Monhub, we bring high-resolution visuals, seamless multi-device connectivity, and ultimate portability — empowering you to work, game, and create anytime, anywhere,” the company said

While casual gamers may find the 60Hz refresh rate acceptable, competitive players accustomed to faster panels will likely notice the difference immediately.

The device also claims broad compatibility with Mac, PC, consoles, and mobile hardware, a list that sounds simple until different operating systems and drivers get involved.

The built-in stand is described as sturdy and adjustable, though how it performs after repeated daily setup and breakdown remains untested.

As with most Indiegogo campaigns, Monhub's real test will not be its specification sheet but whether the finished product matches what backers were shown.

Disclaimer: We do not recommend or endorse any crowdfunding project. All crowdfunding campaigns carry inherent risks, including the possibility of delays, changes, or non-delivery of products. Potential backers should carefully evaluate the details and proceed at their own discretion.

Following up on my review of the Seagate IronWolf 8TB, for balance purposes, it's good to look at what WD is selling into the same retail space.

The WD40EFPX is the model I’ll be looking at in this review, and it has an especially interesting history I’ll dive into in depth. This specific model was first introduced on 14 September 2022, representing one of WD's more recent iterations in the Red Plus line-up. It ships with a significantly upgraded 256MB cache compared to the 64MB of the original WD40EFRX.

The WD40EFPX uses CMR technology rather than SMR, which is particularly significant in RAID environments. CMR drives deliver superior performance during array rebuilds and avoid the write cliff issues that plague SMR drives under sustained workloads.

It runs at 5400 RPM, connects via SATA 6Gb/s, and is rated for NAS systems with up to 8 bays. The workload rating is 180TB per year, suited to backups, file sharing, media streaming, and similar tasks in compact tower NAS units. The 4TB model uses air rather than helium as the internal atmosphere, unlike the higher capacity 12TB and 10TB designs that use helium and not air.

The 4TB model sells for just under $200, whereas its Red Pro brother is closer to $300. That might seem like a good deal, but in December of 2025, it was $150. That said, it’s $20 less than the Seagate IronWolf 4TB.

If you are buying lots of drives, then you would be better off investing in larger capacities that lower the cost per TB, but for a four-bay NAS where 12TB of RAID 5 space is enough, then these might be the smart option.

In performance and durability terms, the WD40EFPX isn’t the best NAS drive we've tested, but it gets the job done.

WD Red Plus 4TB: Price

This is as complete a list of current Red Plus models as I could assemble. I left out the old 1TB, 2TB, and 3TB models because they’re just old stock, and I also didn’t include the 8TB WD80EFZZ (128MB cache) since it appears to be discontinued.

As is evident from the cost per TB, the best value is the 8TB WD80EFPX, and the worst is the 12TB WD120EFBX, which uses helium where all other drives use air. The 4TB model isn’t a bargain, especially when you consider that only a few months ago it could be found for less than $150.

All the prices quoted here are from Amazon.com, and it may be possible to find these drives cheaper elsewhere.

The typical asking price for the Seagate 4TB IronWolf is $219, or that’s what B&H Video wants for one. This is $1 more than the Toshiba N300 4TB from the same retailer, and it’s a 7200RPM drive.

Overall, the WD Red Plus 4TB seems competitively priced, if that’s something you can actually say about drives in the current economy.

WD Red Plus 4TB: Design

Red Plus drives are available from 1TB up to 14TB. The 4TB WD40EFPX sits at the lower end of the range. With these smaller capacities, air-filled construction is standard. Drives pushing into double figures move to helium-sealed designs and, at higher capacities, step up to 7200 RPM class motors.

The Red Plus sits below the Red Pro in the family hierarchy. Red Pro is rated for up to 24-bay systems and carries a 300TB per year workload rating, while Red Plus covers systems with up to 8 bays.

From an external design perspective, there isn’t much to talk about here – it’s a 3.5-inch SATA mechanism that we’ve seen on computers for over twenty years. Most of the more interesting aspects of this design are inside and have WD buzzwords attached to them.

These include an adaptive compensation system with a shaft fixed on both sides, and three-dimensional balancing minimises the negative effects of vibration, particularly in multi-drive enclosures, called 3D Active Balance Plus. Other drive makers have something similar, but call it something else.

WD Red drives also include a multi-axis shock sensor that automatically detects subtle shock events and dynamic fly-height technology, allowing each read/write function to compensate and protect data. I suspect that was originally designed for laptop drives, but anything that avoids the heads from coming into contact with the recording surface is useful.

But the best feature of these drives isn’t a physical feature; it's software.

NASware 3.0 enables seamless integration, robust data protection, and optimal performance for NAS systems operating under heavy demand. It fine-tunes drive parameters to match NAS system workloads for optimum performance. This firmware also includes NAS-specific time-limited error recovery settings, optimised spin-up timing to reduce array-wide power surges, and adaptive thermal throttling.

So how does this hardware compare to its primary competitor, the Seagate IronWolf 4TB (ST4000VN006)?

These two drives occupy the same market position and share a remarkably similar specification sheet. Both run at 5400 RPM, use a SATA 6Gb/s interface, carry a 256MB cache, and are rated for up to 8-bay NAS systems. Both use CMR recording and carry a 180TB per year workload rating. MTBF on both is rated at 1 million hours, and both carry a 3-year warranty. They are literally brothers from different mothers.

Active power draw is 4.7W for the WD40EFPX and 4.8W for the ST4000VN006. IronWolf idle power is 3.96W, and standby drops to 0.5W. WD does not publish a specific idle figure prominently for the WD40EFPX, though the drives behave comparably in practice.

The difference is negligible in a real-world NAS. Across a four-drive array running 24/7 for a year, the delta between these two drives would amount to pennies on an electricity bill. Power is effectively a tie.

The WD Red Plus is consistently described as quieter than the IronWolf in real-world NAS deployments. Both drives include rotational vibration sensors at the 4TB capacity level. However, my testing consistently placed the Red Plus as being the quieter drive, and this aspect could be extra useful in home and near-desk environments.

But what NAS customers are most interested in is reliability because, as I can attest, having drives fail in a working NAS can be stressful.

The WD Red Plus carries a non-recoverable read error rate of 1 in 10^14, while the IronWolf quotes 1 in 10^15. The IronWolf's figure is ten times lower, meaning statistically fewer uncorrectable errors per bits read. In practice, this rarely matters at the 4TB capacity level, and real-world failure rates across both brands are broadly comparable.

Real-world failure data from Backblaze between 2022 and 2024 shows WD Red Plus 8TB and above models carrying an annualised failure rate of approximately 1.4 to 1.8%. Comparable IronWolf figures tend to hover in a similar bracket, making reliability broadly comparable across the two brands. I haven’t got data for the 4TB models, but since they are similar enough to the 8TB models, it’s not unreasonable to assume that the hardware has much the same reliability.

There are, however, a few performance differences I’ll mention later, and Seagate has a health management system.

IronWolf drives include IronWolf Health Management, which works with leading NAS systems to provide prevention, intervention, and recovery recommendations to ensure peak system health. It is enabled on Synology, QNAP, Asustor, and other major platforms. It provides drive health telemetry beyond standard S.M.A.R.T. data, giving users early warning of potential issues.

IronWolf also includes three years of complimentary Rescue Data Recovery Services, with an industry-leading recovery rate of 95% in the event of accidental data corruption or drive damage.

WD Red Plus has no equivalent to either of these features. There is no bundled data recovery service and no proprietary NAS health management system beyond standard NASware 3.0 and S.M.A.R.T. compatibility.

In WD’s defence, many NAS makers are introducing their own AI logic to monitor drive health, regardless of brand, but IronWolf Health Management is one of the reasons these drives have been so successful.

To understand more about this drive, let’s cover the lineage that brought us to the WD40EFPX.

WD Red Plus 4TB: History of the Red Plus range

The WD Red NAS HDD series arrived in 2012. It was a direct response to the growing popularity of home and small business NAS systems. Regular desktop hard drives were designed with single-use deployment in mind, and NAS use demanded something different.

Faster read and write speeds, higher workload tolerance, and suitability for multi-drive, always-on environments.

The WD Red line launched with 1TB, 2TB, and 3TB models. The 4TB variant expanded the range, giving home users and small businesses the full 4TB maximum for up to five-bay NAS units. The original 4TB model carried the model number WD40EFRX.

A unit manufactured in September 2013 used a Marvell 88i9446-NDB2 dual-core drive controller, a Hynix 64MB DDR2 cache IC, and NASware 2.0 firmware. Western Digital described the rotational speed as "IntelliPower" rather than publishing an RPM figure — a marketing term that obscured what was, in practice, a variable-speed 5400 RPM class design.

The drive shipped with TLER (Time Limited Error Recovery) enabled from the factory, which was an important distinction from consumer desktop drives. It made the WD Red family genuinely RAID-compatible, something that mattered greatly in multi-bay arrays where a drive that spins on an error for too long can be rejected by the RAID controller.

Over the following years, WD evolved the NASware firmware, introduced new form factors, and added the WD Red Pro tier for more demanding environments. The original WD Red remained the affordable CMR option for home and SOHO users throughout this period, and the WD40EFRX remained in the lineup for several years.

A key hardware progression during this era was the increase in cache size. Across the Red Plus lineup, cache memory has grown from 64MB in early models to 256MB in current ones. The WD40EFRX shipped with just 64MB, which by today's standards looks overly modest.

Then came the most significant and damaging chapter in WD Red history, and this directly shaped how the Red Plus line came into existence.

Western Digital began shipping SMR (Shingled Magnetic Recording) drives from early 2017, but declined to disclose this to customers, even when directly asked. The EFRX suffix was quietly retired and replaced by the EFAX suffix. What buyers did not know was that the new drives used a fundamentally different recording method.

The larger EFAX drives at 8TB and above remained CMR. The smaller capacities at 6TB and below transitioned to SMR. EFRX drives were removed from production pipelines.

Users began experiencing problems adding the new WD Red NAS drives to RAID arrays. SMR drives were not intended for random write workloads, and NAS rebuild operations like resilvering in ZFS terminology use exactly that kind of sustained random writing.

RAID resilvering tends to overload the cache on SMR drives, sending them into minutes-long pauses. Faulty firmware on the WD40EFAX also caused drives to return IDNF S.M.A.R.T. errors under intensive workloads, which RAID controllers typically interpreted as drive failure.

All my NAS were once populated with these drives, and successively they died like dominoes, until none of them survived. What annoyed me, and others, was that while internally WD knew why the death rate on these drives was high, sales staff were still pushing the line that all WD drives were CMR and that the company would make it "very clear" if SMR technology were used.

In 2020, consumers discovered what was happening. A class action lawsuit followed in the United States. The lawsuit alleged that WD had surreptitiously introduced SMR technology into WD Red NAS drives without disclosure, in an effort to reduce costs while keeping the selling price unchanged.

In an attempt to mitigate the fallout from this sorry exercise, a structural reorganisation of the entire Red lineup was performed by WD.

The WD Red name was retained for device-managed SMR drives across 2TB, 3TB, 4TB, and 6TB capacities, and it was positioned as suitable only for lighter SOHO workloads. A new WD Red Plus brand was created to cover all CMR drives across all capacities from 1TB to 14TB. WD Red Plus was specifically aimed at more write-intensive workloads, including ZFS.

WD described Red Plus as "the new name for conventional magnetic recording (CMR)-based NAS drives in the WD Red family." Critically, the WD40EFRX model number was retained and repurposed as the WD Red Plus 4TB, allowing existing owners to verify that their older drives were CMR by checking the model-number suffix.

During the transition period, WD Red Plus devices were sometimes delivered with a "WD Red" label, but the model number confirmed the CMR identity. This caused ongoing confusion in the channel.

Following the rebrand, there were two 4TB model numbers for Red Plus drives.

The WD40EFZX appeared as a transitional model, shipping with a 128MB cache and confirmed CMR recording. It was described as a 5400 RPM SATA 6Gb/s CMR drive with NASware 3.0 technology and an MTBF of up to 1 million hours.

Alongside that is the focus of this review, the WD40EFPX. As I’ve already mentioned in the introduction, this drive ships with an upgraded 256MB cache and uses CMR technology rather than SMR and has the same 5400 RPM rotational speed.

These drives closed the chapter on the hidden SMR debacle, and as a result, WD’s standing with those deploying NAS storage has recovered.

WD Red Plus 4TB: Performance

Testing drives for NAS use is fraught with problems because NASes are configured to hide the actual performance speeds from you, using their memory as a cache.

As an example of this, I connected a single WD40EFPX to a TerraMaster F4-425 Pro, accessed it over a 2.5GbE LAN, and achieved read/write speeds of 296 MB/s.

As this drive is rated by WD at 180 MB/s, those numbers are plainly artificial.

Therefore, ironically, rather than using a NAS to evaluate the speed of this drive, I resorted to testing it on a regular PC.

Here are my results:

When you ask the sorts of questions that AJA, ATTO and CrystalDiskMark have the answers, then the Red Plus 4TB is remarkably similar to the IronWolf 4TB, with perhaps a tiny edge to the WD drive. That said, those results are all within variance and hardly conclusive.

PCMark and Winsat come down more on the Red Plus side, and it’s especially interesting that read time with sequential writes is lower on that drive than the IronWolf. And, that extends into latency, which is generally better on WD.

With time short for testing, I wouldn’t call these results definitive and on a different PC, they might be entirely reversed.

But they do strongly suggest that the slight price premium that WD is asking might be worth it, although there isn’t any practical way of testing the resilience of the drive over the long run. And that’s specifically where customers have had an issue with this brand previously.

WD Red Plus 4TB: Final verdict

If Western Digital could move on to Nasware 4.0, or something that brings it closer to the Health Management technology that Seagate already has, then they might see a migration back to the days before they tried to slide SMR drives past their customers.

Because the Red Plus is a bit quicker than the IronWolf equivalent, even if it costs proportionally more.

But frankly, the cost of NAS drives of this capacity is way too high to actually encourage people to buy them, even if the underlying technology seems sound.

To paraphrase Monty Python, if Nvidia and the AI evangelists hadn’t artificially nailed storage to its perch, drives of this capacity would be pushing up the daisies by now.

But instead, we have the indefensible exercise in which Seagate, WD and Toshiba profit massively from phantom demand, and every day is Christmas at their factories.

If you must buy drives for a deployment, then go with the 8TB and 10TB models, as they offer the best value, if that means anything when HDDs are nearly as expensive per TB as you could buy SSDs capacity at one point in the past two years.

For more storage solutions, we've reviewed the best NAS devices.

• US Secret Service personnel are using personal devices while conducting official business
• Personal devices are not secured against the threats faced by Secret Service members
• But government-issued devices aren't equipped for the needs of Secret Service members either

The Department of Homeland Security inspector general has released a new report which claims the US Secret Service is refusing to use government-furnished equipment (GFE), such as smartphones, because they are not suitable for mission operations.

The report states GFE fails to “ensure real-time, continuous protection from cyberattacks by foreign adversaries or individuals” with the equipment found to contain multiple third-party apps with security vulnerabilities that could expose communications.

In order to be able to perform effectively, Secret Service members are using personal devices to communicate with law enforcement and each other during missions, but many personal devices are not secured against the threats faced during the protection of America’s VIPs.

US government struggles to secure issued phones

But using personal devices in professional operations is also highly unsecure. These devices often contain the whereabouts of Secret Service personnel and the targets they are protecting during missions at home and abroad.

Furthermore, the devices only have the consumer level of cyber protections. As they are not managed or operated by the US government, there is very little protection against commercially available spyware or malware.

In some cases, personnel used their personal devices as a hotspot for their GFE, or used their personal devices to access websites otherwise blocked on their GFE.

The report explains: “If a personal device is jailbroken, infected with malicious code, or not up to date on security software, an adversary could intercept device communication. Outdated and vulnerable apps could enable malicious actors to conduct surveillance, track locations, or record employees’ communications. Connecting to unsecured networks may also allow cybercriminals to access data or install malware.”

The main culprit behind Secret Service personnel choosing not to use GFEs was found to be the Secret Service’s Office of the Chief Information Officer (OCIO). According to the report, “GFE mobile devices lacked mission-critical capabilities because Secret Service OCIO’s process for assessing and approving requests did not always correctly identify operational needs.”

Additionally, the expected protocol for most Secret Service members was to use personal devices, so many avoided navigating the bureaucracy of requesting access to communications apps on their GFE, which in return created a blindspot for the OCIO who were not aware these apps were already being used at such a scale.

The report further found that no Secret Service GFE was equipped with Mobile Threat Defense software until August 2025, leaving them exposed to “malicious software,

cyberattacks, and other vulnerabilities.” Critical data was also retained on GFE devices after operatives returned from missions abroad, despite policy stating that devices should be wiped within 24 hours of returning to the US.

Ultimately, the report makes five key recommendations to the Secret Service in order to improve the security of its operators:

• Introduce a formal policy that ensures all GFE are issued with the required capabilities and software for each mission
• Ensure all employees complete the required cybersecurity training
• Ensure the Secret Service OCIO clearly communicates its guidance that personal devices are forbidden from use during official business
• Ensure controls are implemented to wipe devices in line with OCIO policy for returning personnel
• Subject all GFE mobile app code to an updated vulnerability testing policy

• One upward push turns your monitor into wall art instantly
• DuoShift replaces notifications with a physical, deliberate end-of-day gesture
• Compact apartments finally get a desk built for two lives

Compact living spaces have made it increasingly difficult for people to separate their professional and personal lives within the same four walls.

Seung Bin Bae, a Korean student designer, has created a dual-purpose workstation called DuoShift aimed at fixing one of remote work's most persistent problems.

DuoShift addresses this issue through a single physical motion rather than relying on software, apps, or scheduled reminders.

A single motion replaces years of software fixes

The desk operates in two distinct modes, the Work Mode and Life Mode, and switching between them only needs an upward shift of the screen.

In Work Mode, it functions as a standard productivity monitor, holding spreadsheets, browser tabs, and video calls during the day.

Pushing the screen upward shifts it into Life Mode, where it becomes a digital art frame instead.

This transformation simultaneously clears the desk surface beneath it, returning the space to non-work use entirely.

Unlike calendar apps or notification systems that attempt to impose discipline through software, DuoShift relies on a deliberate physical gesture to mark the day's end.

Bae's approach treats this transition as a ritual, similar to closing a laptop or changing out of work clothes after finishing a shift.

Visually, the design remains minimal and slim, intended to integrate into a living space rather than dominate it visually.

Why this matters beyond the desk itself

Compact urban living continues to expand rather than recede, and remote work remains common years after the pandemic reshaped daily routines.

Most monitor designs have not adapted to treat this overlap between living and working as a serious design problem.

DuoShift's customizable frame finishes also allow it to either blend into a room's interior or stand out as a deliberate design object.

Beyond its function, the product hints at commercial possibilities, including digital art subscriptions or collaborations with furniture and interior design brands.

Its modular structure additionally permits upgrades without full replacement, a detail aimed at reducing electronic waste and extending product lifespan over time.

As compelling as the concept is, it remains a student project without the manufacturing scale needed to reach a global audience.

Samsung, as one of South Korea's most powerful electronics brands, could realistically turn this idea into a mass-market product.

Without backing from a company of that size, an innovation this promising risks staying confined within South Korea's borders.

The project was named an Honoree in the Home & Living category at the Core77 Design Awards.

Whether the concept could scale into mass production through a partnership remains an open and unanswered question.

What DuoShift demonstrates, regardless of its commercial future, is that a single deliberate gesture can offer something software has consistently struggled to provide.

You've probably noticed the same handful of tasks eating your week: chasing invoices, answering customer questions, copying data between tools that don't talk to each other. Then "AI agents" started showing up everywhere, from tech newsletters to LinkedIn; you're left wondering whether any of it applies to a business your size. The honest answer is that it usually does, no coding skills required.

Maybe you've already tried a chatbot or a basic automation tool to cut down on repetitive admin. Both helped for a while, then hit a ceiling: chatbots answer a question and stop, while basic automation only runs the exact steps you set up originally. AI agent builders are built to close that gap, letting you create something that can finish a job from start to finish instead of stopping halfway.

What is an AI agent builder?

At its simplest, an AI agent builder is a platform that lets you design, deploy, and manage AI agents without coding them from scratch. The agents themselves run on large language models (LLMs) that can reason about a request, decide which tools to use, and carry out several steps toward a goal.

That's the part that separates an agent from a basic chatbot. A chatbot answers what you type into it. An agent can look up an order, update a record, and send a follow-up email from a single instruction, then report back once the job is done.

It helps to see the three side by side.

Most of the confusion around "agentic AI" comes from vendors stretching the word to cover all three. A genuine agent is the one that can act, not just answer. Plenty of products marketed as "agents" today are really just chatbots or workflow automations with a new label, so it pays to ask a vendor exactly what their agent can do on its own before you sign up.

How to build AI agents for business automation

Every agent builder, however polished, is made of the same basic pieces. You give it instructions in plain language or a visual flow. The platform connects those instructions to your tools and data through APIs, then handles the reasoning and orchestration behind the scenes.

Most platforms also add memory, so an agent can recall earlier steps in a task. They pair that with guardrails that limit what the agent is allowed to do without a person signing off first.

Builders generally fall into three categories. No-code tools use guided steps and templates built for people with no programming background. Low-code tools expose the underlying logic, so technical teams can adjust it without starting from zero.

Pro-code frameworks such as LangChain sit at the far end, handing developers full control over memory, tool access, and orchestration. IBM's research on enterprise builders suggests the right choice depends less on team size than on how much oversight the work needs once it's live. A small business automating a single email workflow rarely needs the same setup as a bank automating fraud checks.

Why should businesses consider using one?

The case for an agent builder comes down to time and reach. Agents complete actual work instead of just suggesting it. They do this continuously, without adding headcount.

A few concrete reasons businesses are adopting them:

• Faster from idea to working tool. Templates and guided builders mean a working agent can go live in days rather than months of custom development.
• Lower cost than hiring for every workflow. Consumption-based pricing on most platforms means you pay for what an agent actually does, not for a full-time salary.
• Non-technical teams can build their own. No-code builders put creation in the hands of the people who know the workflow best, not only IT.
• Scales without adding headcount. One agent can handle a large volume of routine requests at once, freeing staff for the work that actually needs a person.
• Agents finish the job, not just start it. An agent can complete several steps across different systems from one instruction, whereas a chatbot would simply answer and stop.
• It works for small teams too. You don't need an in-house engineering team to get value; many of the platforms below are built specifically for business owners who aren't developers.

The numbers back this up well. Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of 2026, up from under 5% in 2025. McKinsey has estimated that generative AI, the technology underpinning these agents, could add up to $4.4 trillion a year to the global economy.

None of this happens automatically, though. A 2025 IBM study of 2,000 CEOs found that only 25% of AI initiatives had delivered the ROI leaders expected. Just 16% had scaled across the whole business, which says less about the technology than about how carefully it was deployed.

Common business use cases

AI agents tend to cluster around a handful of jobs. The same patterns show up whether you run a five-person shop or a five-thousand-person company:

• Customer service: Answering routine questions, processing refunds, and escalating complex cases to a human when needed
• Sales and lead handling: Qualifying inbound leads, drafting outreach, and scheduling meetings
• Finance and operations: Processing invoices, chasing approvals, and flagging anomalies in expense reports
• IT and HR support: Resetting passwords, answering policy questions, and handling onboarding paperwork
• Internal knowledge search: Pulling answers from scattered documents, wikis, and past tickets so staff stop hunting for information
• Marketing and content support: Drafting first versions of emails, social posts, or product descriptions for a human to review and publish

Smaller businesses tend to start with one, often customer service or lead handling, before expanding once the first agent proves its worth.

Choosing the right AI agent builder

The right platform depends heavily on your existing tools and your team's technical comfort. If your business already runs on Microsoft 365, Copilot Studio lets you build agents inside that ecosystem, with internal agent use bundled into Microsoft 365 Copilot licenses and a standalone option for publishing agents outside it.

If your business lives in Salesforce, Agentforce builds agents directly on top of your CRM data, with a free Foundations tier to start and consumption-based credits beyond that. For broader, non-technical automation, tools such as Zapier Agents and Lindy let small teams connect an agent to email, spreadsheets, and thousands of other apps using plain-language prompts rather than code.

Larger organisations weighing governance more heavily often look at platforms such as IBM watsonx Orchestrate, which offers no-code, low-code, and pro-code paths in one place, with centralised oversight over what every agent is allowed to touch. Engineering-led teams that want full control, meanwhile, tend to build directly with open-source frameworks like LangChain rather than a packaged product.

As a general rule, the more deeply your business already runs on one ecosystem, like Microsoft or Salesforce, the more sense it makes to build your first agent inside that same ecosystem rather than bolting on a separate tool. Before you commit to any platform, it's worth getting clear answers to a short list of questions:

• Can it connect to the specific tools you already use?
• Does pricing scale predictably as usage grows?
• What happens when the agent gets something wrong?
• Who is responsible for reviewing that?

A few things to consider

Adoption is moving fast, but it isn't risk-free. Gartner predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027 because of rising costs, unclear business value, or weak risk controls. Gartner has also warned of "agent washing," where vendors rebrand older chatbots or automation tools as agents without the underlying capability to match.

An agent is only as good as the data it can see, too. If your records are scattered across spreadsheets, inboxes, and three different apps that don't sync, an agent will struggle no matter how good the platform is. Clean, accessible data is the unglamorous prerequisite that most successful deployments get right before anything else.

Security deserves the same scrutiny you'd give any other system with access to your data.

It's also worth thinking about lock-in before you build anything substantial. Agents built deep inside one vendor's ecosystem, like a CRM or productivity suite, are usually harder to move later than ones built on an open framework or with clear data-export options. That's a reasonable trade-off for the convenience many platforms offer, as long as you go in with eyes open about it.

Start small. Pick one workflow with clear, measurable value. Run it in a sandbox or a limited rollout before expanding it further.

Keep a person in the loop for anything customer-facing or financially sensitive until the agent has proven itself. Revisit that oversight regularly, rather than setting it once and walking away.

FAQ

• California government will have access to Anthropic's Claude with a 50% discount
• The technology will be used to improve workflows and cybersecurity
• Governor Newsom said Claude will be used "responsibly, transparently, and in service of people"

The government of California will now be able to use Anthropic’s Claude AI with a half price discount.

A press release published by California Governor Gavin Newsom says the state's government will also have access to free workforce training, expert GenAI technical assistance and workflow input from Anthropic developers.

“This partnership is about using technology the California way: responsibly, transparently, and in service of people. AI should not replace the human work of government; it should help our workers move faster, solve problems more effectively, and deliver better results for Californians,” Gov. Newson said.

Claude comes to California

The discount for Claude also extends to local governments at the city and county level, allowing state workers with lower budgets to gain access to cutting edge tech. Gov. Newson said the technology will primarily be used for drafting, summarization, and analysis, while also “supplementing day-to-day work and improving services for Californians.”

Californian state agencies will be able to access Claude through the Statewide Information Technology Shared Services (SITeS), a new portal that centralizes AI tools for government use. The Californian government has already worked alongside Anthropic to integrate Claude into numerous tools for state workers, such as the Engaged California tool that provides Californians with more of a voice in policymaking.

Claude Security and Claude Code are also being integrated into the workflows of the California Department of Technology and the California Governor's Office of Emergency Services in order to improve cybersecurity. The California DMV will use Claude to reduce wait times and improve services, and the Department of Healthcare Services will use Claude to assist in the state’s Medicaid program.

The state is home to 33 of the top 50 private AI companies in the world, including Anthropic. “As a California company, we feel a real responsibility to our home state. We’re honored to expand our partnership with California’s agencies and to put Claude to work for the people who keep this state running,” said Kate Jensen, Anthropic’s Head of Americas.

“Building AI responsibly and in service of people has been our approach from the start, and that’s exactly what this partnership puts into practice.”

While many of us wouldn’t think twice about buying software from a global hyperscale software vendor, open source software (OSS) has flown under the radar as a majorly important part of the software landscape for decades.

Recent shifts in European attitudes, fuelled by geopolitical tensions and technological sovereignty concerns, has seen an increased focus on the desire to run open source software and back local vendors. A major advantage is the open nature of its development – anyone can inspect, modify and use it, and companies can distribute their own versions without the limitations and expenses of vendor lock-in.

It’s so important that around 70% of modern software stacks are estimated to rely on open source components in some form – EXANTE sees it as a plumbing system that keeps software stacks together.

But that very benefit could also be a disadvantage for corporate customers, because small groups of volunteers don’t have the same requirements as large corporations. While companies invest heavily in cybersecurity and resilience, visibility into open source software can be alarmingly low.

AI is making underfunding even more pronounced

Vibe coding has further complicated matters, with developers now able to write code more quickly and malicious actors leaning on those same AI coding tools to discover vulnerabilities and develop exploits.

Without the commercial funding that hyperscale software receives, it puts OSS in a more vulnerable position. But given the importance of OSS for companies looking to build their own software stacks, many are now starting to see backing projects as an integral part of managing their risks.

To discuss why the responsibility, or burden, of open source software sustainability is increasingly being picked up by enterprise customers instead of individual developers and contributors, I spoke with EXANTE CTO Richard Forss, who explained how AI is changing software development and cybersecurity strategies globally, and why businesses should treat it as critical infrastructure.

EXANTE is also the company behind the Gecko Fund – a new €1 million grant programme to support critical open source software projects used across trading and financial data systems with grants of between €10,000 and €150,000 available – and the company believes that financial and technical support for open-source projects is now more important than ever.

“We believe the industry that benefits from these tools should play a role in sustaining them,” Gecko Fund founder Anatoly Knyazev asserted.

• How broadly is open-source software used across fintech, and, namely, brokers? How exactly is it used?

Open source is the foundation of modern fintech and brokerage businesses, but almost nobody talks about it. If you looked under the hood of any brokerage platform, ours included, you would find that about 70% of the critical stack is open-source.

It powers the databases, operating systems, messaging layers, cloud infrastructure – the bits that quietly move millions of transactions a day. Clients never see any of it, which is rather the point.

The reason we all use it is simple economics. Software is built in layers, and there is no sense in any one firm rebuilding the foundations from scratch. We use shared, battle-tested components for things like market data processing and risk infrastructure, and we spend our own engineering effort on the parts that are actually ours.

It lets teams move quickly and solve complex architectural and engineering puzzles that would otherwise be out of reach.

• Why does the underfunding problem in open-source software exist, despite it being so widely used across the global economy?

Because it works too well to notice. Open source behaves like a utility, the plumbing in the walls, and nobody thinks about the plumbing until it bursts. Millions of organisations rely on these tools every day, and for years, almost none of them paid a penny towards keeping them alive.

More accurately, open source drives an estimated $8.8 trillion in global economic value. Yet, nearly two-thirds of the developers responsible for maintaining these widely adopted systems receive very limited financial support.

Hundreds of thousands of organisations – from startups to huge enterprises – benefit from these tools every single day, but the actual responsibility for maintaining them stays with a handful of volunteers.

Everyone relies on the infrastructure, but historically, almost no organisation has taken responsibility for paying for its upkeep.

• Why should businesses care if critical open-source projects are maintained by very small teams or individual developers?

Because it is a single point of failure hiding in plain sight. We spend enormous amounts of money assessing vendor risk and operational resilience, scrutinising suppliers, war-gaming outages, and then the whole edifice can rest on a library maintained by one or two exhausted people you have never heard of.

Our first Gecko Fund grant went to Kryo, an open-source data serialisation tool used all over the world in high-performance and trading environments. For years, it has been kept going by two people on two different continents, with no outside funding at all. They do excellent work – and that is, of course, not the problem. The problem is that it is two people. If one of them burns out, or simply does not have a free evening to patch a serious flaw the moment it appears, the consequences do not stay neatly within their project. They ripple out into regulated markets that have no idea they were depending on them.

• How is AI changing the risk profile of open-source software and cybersecurity?

AI has put its foot on the accelerator, and it is pressing down on both cars at once. On our side, it genuinely helps. It spots bugs, speeds up code review, catches things early. The problem is that the people trying to break in have exactly the same tools.

Bad actors use AI to scan open-source codebases, identify vulnerabilities, and launch targeted exploits much faster than before. The time between a flaw being discovered and an active attack being launched is continuously shrinking.

This puts intense pressure on underfunded maintainers to deploy fixes instantly. Since AI scales these threats so effectively, the security of the underlying open-source ecosystem has become a systemic priority.

• Which parts of the open-source ecosystem are most vulnerable to underinvestment today?

It’s not the famous projects. Mostly the dull ones. The risk lives in the deep, unglamorous code: I’m talking about the low-level libraries, the developer tooling, the APIs – developer tools that operate entirely out of sight. The components that never trend on anything and never will.

That is precisely why they go unfunded. They do not make headlines, so they do not attract sponsors or donations, and yet they are wired into thousands of commercial products.

It is the paradox of the whole industry. The more essential a piece of software is, the more invisible it tends to be. When one of those hidden libraries fails, it does not fail for one company. It fails for everyone at once.

• How do you think this gap can be bridged? What measures can businesses take to mitigate possible risks?

Stop calling it a hobby and start treating it as infrastructure, because that is what it is. The first thing any firm should do is the unglamorous work of auditing its own supply chain and understanding exactly which open-source components it actually depends on. Most have never properly looked. Once you know what you are standing on, you have to hold it up.

But money is not the only currency – firms can contribute engineering time, security audits, testing environments, or documentation. The goal is to make the relationship mutual: if your business profits from these tools, your business should help keep them safe.

• What infrastructural changes can be made to make open-source software more sustainable?

The corporate world needs to stop expecting critical infrastructure to be maintained by someone giving up their weekends. That is the change. Everything else follows from it.

It is starting to shift. There are sovereign technology funds now, government-backed efforts, industry groups forming around open-source standards. That is encouraging.

But finance, of all sectors, ought to be leading rather than waiting, because finance has more to lose than most if this plumbing fails. If we formalise how companies contribute and build proper structures for funding this work, then the people holding up the machinery behind global commerce might finally get the stable backing they need to keep the markets running. They have earned it several times over.

HyperDrive Next TB5 Dock: 30-second review

While docking stations are rarely the most glamorous pieces of hardware on a desk, they do offer a generation shift in connectivity that can completely transform a professional workflow. That’s the thinking behind HyperDrive Next Thunderbolt 5 Dock. Take a single Thunderbolt 5 uplink cable, and with it convert a premium laptop or professional mini PC into a modular, desktop-class powerhouse.

Historically, compact Thunderbolt docks forced users to accept a hierarchy of compromises-sacrificing networking speeds, capping host charging, or forcing external storage enclosures to occupy precious downstream ports. Hyper systematically eliminates these constraints. Built around Intel's newest Barlow Ridge controller, the dock shifts dynamically between a symmetrical 80Gbps bidirectional layout and an asymmetrical 120Gbps downstream pipeline for monstrous multi-display or high-refresh configurations.

What elevates the HyperDrive Next over its closest competitors, like the Ugreen Maxidok 10-in-1, is its inclusion of premium internal features. Rather than reserving professional-grade utilities for a giant, corporate-sized footprint, Hyper fits an active-cooled M.2 NVMe PCIe Gen 4 slot and a 2.5GbE LAN port directly into this rugged chassis.

One snag is that at $399.99, which seems excessive for a dock with this number of ports. But for digital creatives, video editors, and engineering professionals seeking maximum bandwidth and a clutter-free desk, it might be worth it.

It also suffers from some of the same performance issues that I noticed on the UGREEN Maxidok 17-to-1 TB5 dock, and it doesn’t offer any native HDMI or DisplayPort outputs. That last limitation means that if you want to connect monitors, you will need to buy Thunderbolt to HDMI/DP adapters, and those aren’t cheap.

Because of some of these points, the HyperDrive Next Thunderbolt 5 Dock isn’t a candidate to join our best docking station collection. But it might be the right specification for some customers.

HyperDrive Next TB5 Dock: Price & availability

• How much does it cost? $399/€348.75
• When is it out? Available now
• Where can you get it? Direct from the HyperShop, Amazon.com, and authorised pro-audio/video retailers like B&H.

The HyperDrive Next Thunderbolt 5 Dock is available direct from Hyper, as well as retailers such as Amazon and B&H Photo.

The device sits firmly in the premium tiers of enterprise and prosumer desk hardware. At $399.99, it is explicitly targeted at those who can monetise the massive leaps in transfer speeds and potential display throughput.

When positioned against the broader marketplace, this is on the edge of the premium space for a dock that, in theory, has the features to justify it. For context, smaller "halfway house" Thunderbolt 5 docks like the Ugreen Maxidok 10-in-1 retail closer to $300 but cut out the internal storage slot, drop networking to standard 1GbE, and cap individual laptop charging to 100W.

To get equivalent feature parity from alternative brands, you typically have to move to massive 17-in-1 enterprise enclosures that carry bulkier footprints and larger PSUs.

The competitive landscape is interesting, since many makers still consider TB5 technology to be high-end rather than consumer products.

The key players that make this level of hardware are Anker, CalDigit, Kensington, OWC, and Ugreen.

However, at this time, only CalDigit, Kensington and UGREEN offer a dock with an M.2 slot

One that many will gravitate to is the CalDigit TS5 Plus, a dock that sports a 10GbE LAN port and 140W power profile. That 20-port option doesn’t have an M.2 slot, and it costs $499.99/£469.99 on Amazon.com.

I did notice that the Kensington SD7100T5 EQ Pro isn’t globally available, but it can be bought in the USA, if you have $449.99 burning a hole in your pocket. It’s a fully featured dock with three TB5 downstream ports, an M.2 slot, and a 2.5GbE LAN port.

The most recent arrival is the Ugreen Maxidok 17-to-1, a slightly cheaper option at $390, but it does have a DisplayPort output, M.2 and 2.5GbE LAN port.

While only undercut by the Ugreen Maxidok, the HyperDrive Next Thunderbolt 5 Dock is only a ten-port dock, and if it didn’t have the M.2 slot, I’d be comparing it to the likes of the Maxidok 10-to-1 option that costs only $250.

I should also point out that to make the most of this dock, you need Thunderbolt 5, though it will work with Thunderbolt 4, Thunderbolt 3, USB4 and even USB-C.

Buying this dock with a Thunderbolt 4 machine means paying a premium for capabilities you cannot yet access. The hardware is forward-looking, but the investment only pays off when the host catches up.

If you aren’t planning to upgrade to a machine with Thunderbolt 5 technology, then this hardware's best feature will mostly go unused.

• Value: 4 / 5

HyperDrive Next TB5 Dock: Specs

HyperDrive Next TB5 Dock: Design

• 3x TB downlinks
• No HDMI or DisplayPort
• M.2 slot on the underside

The success or failure of this design hinges on whether you intend to use a monitor attached to the dock. The cable/adapters to do this aren’t expensive, but they are an extra expense, and it would have been nice if Hyper had included at least one with the dock.

The design of this dock reminds me of others made by StarTech and OWC that use the same extruded aluminium tube with capped ends.

It’s not horrible, and it feels remarkably robust, but it's also not cutting edge or especially refined.

The HyperDrive Next TB5 dock is a horizontal desktop slab measuring 19.4 x 7.8 x 5.2 cm. It is compact without being cramped, and the low profile keeps it unobtrusive on a crowded desk.

What struck me first was how heavy it is, with the dock weighing an impressive 1.88kg (4.12 lbs). If you’ve had a dock that someone has knocked off your desk with their elbow in passing, this is not that type of dock. And, brushing this one at speed might result in a trip to the doctor.

Construction uses a mix of aluminium and recycled PCR plastics. Hyper states that 75% of the dock body is composed of post-consumer recycled material. The aluminium elements serve a thermal purpose as well as an aesthetic one, helping dissipate heat during extended running.

The front panel carries one TB5 downstream port, two USB-A ports at 10Gbps, a 3.5mm audio combo jack and a power button. The physical power button pays dividends in daily use. Powering the dock on and off without unplugging the host cable is a small convenience that quickly becomes an expectation.

Having a TB5 downlink on the front is nice, although some might get confused and think that it is the uplink.

The rear carries the host TB5 port, two further downstream TB5 ports, two USB-A ports, the 2.5GbE RJ-45 port, the DC power input and dual Kensington lock slots. The two-slot security provision, one nano and one standard, is a welcome professional-grade addition.

If you haven’t clicked already, Hyper traded away the dedicated monitor connection for the third TB5 downlink, meaning you will need adapters to connect any type of monitor.

The M.2 expansion bay is accessed via a removable panel on the underside of the chassis. Fitting a drive is straightforward. The bay supports full-size M.2 2280 modules at PCIe Gen 4 x4 or Gen 3 speeds, but it can also accept smaller M.2 form factors. Hyper also formally supports compatible PCIe M.2 accelerator cards here, including inference hardware such as the Hailo-8.

I’ll be frank and say that before Hyper mentioned such functionality, I was unaware that AI acceleration in the M.2 form factor was even a thing, but apparently it is, and this dock supports it.

• Design: 4 / 5

HyperDrive Next TB5 Dock: Features

• TB5 Bandwidth
• M.2 Slot
• 140W charging

Thunderbolt 5 offers 80Gbps symmetric bandwidth, handles simultaneous data, display and power delivery. When the host machine supports bandwidth boost mode, that figure rises to 120Gbps on the display path, enabling the full triple 4K or 8K output scenarios.

Display support is platform-dependent. On Windows and on the MacBook Pro M5 Pro or M5 Max, three external displays run at up to 4K 144Hz or a single 8K at 144Hz. Other macOS configurations and Chrome OS devices are limited to two extended displays. This reflects the display engine capabilities of those platforms rather than any shortcoming in the dock itself.

Unless you have a Thunderbolt-capable display, you will need appropriate cables or adapters to connect screens, and as many of these use the USB-C protocol, results may vary.

On a more positive note, Thunderbolt Share is the feature that sets this dock apart from most competition. Connecting two computers via TB cables allows direct file, peripheral and control sharing at speeds Hyper claims are up to 64 times faster than Gigabit Ethernet. No network infrastructure is needed. A work Mac and a personal Windows machine can share a keyboard, mouse and monitor without touching a cable.

That’s especially useful if you have a desktop computer and a laptop, since it avoids the need to buy yet another gadget to attach them simultaneously.

The M.2 slot deserves its own discussion. PCIe Gen 4 x4 bandwidth through a TB5 connection is theoretically sufficient for any current NVMe drive to operate near its rated speeds. The slot also supports PCIe-based AI accelerator modules. A Hailo-8 or similar inference card installed here allows a laptop to run local AI workloads without a desktop GPU.

Power delivery reaches 140W via EPR PD 3.1. This is the highest common tier for laptop charging, sufficient to keep high-performance MacBook Pros and demanding Windows workstations topped up even under load. The physical power button on the front allows the dock to be cycled without disturbing the host cable connection.

Port count totals ten across the front and rear. All four USB-A ports run at 10Gbps. Dual Kensington lock slots in nano and standard formats reflect Hyper's targeting of professional and enterprise environments.

• Features: 4 / 5

HyperDrive Next TB5 Dock: Performance

• Intel created performance issues
• 140W power dilemma

I ran into issues with this dock, and they were identical to those I previously experienced on the Ugreen Maxidok 17-to-1, hardware built on the same Intel Barlow Ridge controller.

That suggests neither Ugreen nor Hyper is the culprit in these cases, and that there is an issue with the controller's bandwidth allocation to both the 2.5GbE LAN port and the M.2 slot.

In both cases, neither interface performs as it should, reducing file transfers over the LAN port to about 60% of what I would normally expect and making the Gen4x4 appear to be a Gen3x4 or Gen4x2 slot.

And, like the Ugreen dock, I was able to get the full 2.5GbE LAN experience by plugging an inexpensive $26 USB to Ethernet Adapter 2.5GbE into the dock. Which means that it allocates less bandwidth to the LAN port than it does to 10Gbps USB Type-A ports.

The issue appears to be the same with the M.2 slot, and again, I was able to get better performance from a Corsair EX400U USB4 external SSD attached to one of the TB5 downlinks than I was for a Gen4 drive installed in the M.2 slot.

Those are issues it inherited, it appears, but there are some others that I need to mention that were choices Hyper made.

These relate to the power distribution, as this dock only has an 180W PSU, yet it claims it can charge at 140W. It can probably charge at 140W, assuming none of the other ports is drawing power. But if the M.2 is occupied, and something is pulling 15W from each of the Thunderbolt downlinks, there is a mathematical problem with getting those numbers and the overhead of running the dock from a total of 180W.

It’s interesting to note that other flagship docks come with 240W or even 330W in one case to avoid the power pinch, but not this one.

I feel bad talking about the performance issues on this dock because I don’t think the maker can be blamed, but equally, anyone handing over $400 for one of these needs to know what to expect.

• Performance: 3.5 / 5

HyperDrive Next TB5 Dock: Final verdict

The HyperDrive Next Thunderbolt 5 Dock is a strong entry into a rapidly maturing market. Hyper has delivered on the headline promises of TB5 bandwidth, triple 4K display output, Thunderbolt Share connectivity, a built-in M.2 expansion slot and 140W EPR charging. However, the devil is certainly in the details of many aspects.

The performance ceiling of the M.2 slot and the 2.5GbE port under load is a real limitation. But the context is that these shortfalls have appeared consistently across every dock built on the Intel Barlow Ridge TB5 controller family. They are not a Hyper problem; they are an Intel architecture problem. Buyers who understand that tradeoff will find this dock a genuinely capable daily driver, if it fits their specific workflow.

At $399.99, the dock is priced in the thick of the premium TB5 segment. The built-in M.2 bay and Thunderbolt Share support give it a differentiated position, but the lack of at least one dedicated monitor connection or any card slots could be showstoppers for some.

But for others, the inclusion of three TB5 downlinks opens up the dock to be a highly flexible option.

HyperDrive Next TB5 Dock: Report card

Should I buy a HyperDrive Next TB5 Dock?

Buy it if...

Don't buy it if...

Also consider

• Qualcomm Dragonfly AI200 AI accelerator rack is the first of multiple releases planned by the chip designer as it aims to score wins in the data center segment
• The upcoming Dragonfly AI250 accelerator leverages its proprietary High Bandwidth Compute (HBC) to offer a theoretical 18x the amount of bandwidth of its sibling
• Qualcomm's push comes amid an increasingly lucrative datacenter market grappling with memory shortages

It is no secret that the modern AI server ecosystem is dominated by Nvidia in most countries, even as China increasingly leans towards Huawei as its own home-grown provider for similar solutions.

Qualcomm may not be one of the first companies that come to mind when you think about AI data centers or the chips housed inside them, with many investors feeling it has missed the boat altogether in the server segment.

Qualcomm's recent Investor Day 2026 event was a reminder that it is not only still in the game but also has ambitions to carve out a large piece of an ever-increasing pie by taking a different route than most of its HBM-leveraging competitors.

An alternate ecosystem to Nvidia's industry standards?

Much of Qualcomm's Investor Day event focused on its plans to become a sizable player in the AI data center market, which is currently dominated by OEMs deploying a mix of Nvidia and AMD accelerators alongside custom silicon (ASIC) offerings from Google, Meta, Microsoft, and even Amazon's AWS.

It aims to do so by differentiating itself from the competition, relying on its own area of expertise to carve out an edge: efficient Low-Power Double Data Rate (LPDDR) memory stacked in a 3D array above its AI accelerators to drive the next generation of AI inference workloads.

The near-memory compute architecture isn't exactly a new play in a market teeming with similar approaches, but the numbers are hard to argue with when it comes to Qualcomm's offerings.

Qualcomm's upcoming Dragonfly AI200 rack delivers 43 TB of LPDDR5X capacity and 414 TB/s of memory bandwidth per rack, built from accelerator cards each carrying 768 GB of LPDDR5X, which makes it an interesting offering, but much of the focus that hyperscalers will have will be on its Dragonfly AI250 sibling that incorporates High Bandwidth Compute (HBC) under the hood.

While it offers the same memory capacity per rack, its ability to leverage memory at up to 18x its sibling's bandwidth yields a theoretical peak memory bandwidth of up to 7.4 PB/s per rack, a far cry from the AI200's 0.4 PB/s.

The Dragonfly is positioned as an inference-centric accelerator for a reason; however, HBM is still better suited to certain tasks, such as training models rather than inference, making it the memory of choice for Nvidia's Blackwell and upcoming Rubin GPUs, as well as AMD's Instinct offerings.

With that being said, Qualcomm's solution is intriguing, even if the numbers are for specific use cases and its ability to court Hyperscaler giants such as Microsoft and Meta tends to indicate that it has a potential win, at least on paper, as AI datacenters continue to increase focus on inference-centric solutions to deploy their increasingly complex models to wider audiences.

The FIFA World Cup 2026 is well underway, with the group stages already generating huge amounts of action for fans across the world.

It’s been widely highlighted how this is the biggest edition of the tournament so far, with an expanded roster of 48 teams doing battle in 16 venues across three countries. But with such a sprawling reach, the technology needed to power such a crucial tournament has never been more important - and that’s where Lenovo comes in.

As FIFA’s official technology partner, Lenovo has provided the technical expertise to keep the tournament running, and power some of the most exciting new innovations such as AI Refcam - I went to find out more.

An unprecedented challenge

“We were given a challenge which is unprecedented for the game,” Ken Wong - EVP and President, SSG, Lenovo, tells me in Miami.

“For anyone that is going to be a single point of contact for such a complex project, there will be a lot of questions…but after two weeks of operations, I think we have delivered what we promised.”

Wong highlights how the technology being built and delivered to FIFA is actually Lenovo’s own in-house hybrid platform and delivery platforms, so the company was confident about the capabilities it could provide - and then, Wong says, hand it over to FIFA, “to make magic happen.”

As Wong notes, “keeping the lights on is a given”, and Lenovo has deployed more than 350 engineers to World Cup sites, including the stadiums and its Technology Command Center in Miami, with the company providing around 17,000 devices to ensure a resolute and powerful backbone for all the tournament’s operations.

The technology has (at least at the time of writing) been one of the biggest success stories of the World Cup 2026, with fans around the world praising the immersiveness of the Refcam and realistic VAR avatars.

“It's not only about technology - without outcome it is meaningless,” Wong notes, with the company looking to provide a fan experience “like never before”.

“Our role is to make sure based on our understanding of the technology in the platform - how we can maximize the flexibility and outcome for our customer? Now it's up to FIFA to use it.”

The "power of AI"

I ask Wong if the World Cup 2026 can be a good use case for responsible AI deployment, boosting everything from the fan experience to broadcast quality to more effective ground logistics and operations.

“Customers are often so focused on, do I have the most powerful model - but more and more, when AI is moving from POC to production and scale, the more people understand the importance of data,” he says.

Any complex operation that involves a massive amount of data can be a vital learning experience, Wong says - naming building or facilities operations all the way up to smart city oversight as possible use cases which can be extrapolated from the World Cup use case.

Lenovo has even established sports as one of its focus verticals, with the company also partnering with the likes of Formula 1, the Olympics, NHL, Serie A and more.

“We have confidence that we can provide a great outcome, for other organizations within the sports industry, be it a stadium or fans experience, or a whole federation,” he adds.

So with the tournament still underway at the time of writing, Lenovo might be hoping, paradoxically, that its technology stays out of the headlines for the time being - as that will mean everything is working smoothly, with no errors or controversies.

“We are the technology backbone of FIFA,” Wong notes, “our AI is powering the World Cup! To me, the real meaning is that we're part of the great game, but also it is proven now that Lenovo technology can be applied to the most complex operation in the world.”

• 54% of global data center capacity is at risk of temperature-related climate stress
• Before damage even occurs, operating costs can soar to keep up with cooling demands
• Future projects should focus on temperature projections and resource availability

A new First Street report has identified extreme heat as one of the most significant long-term risks that data centers face worldwide, with rising global temperatures, higher hardware power density and strained cooling demands all contributing to the problem.

More than half (54%) of the global data center capacity surveyed is now said to be located in markets that expect to face growing chronic climate risks like rising temperatures, more frequent heatwaves, drought and water stress.

But besides causing physical damage, rising temperatures can also add to operational costs simply by demanding more cooling capacity, the report argues.

Heat is a data center's worst enemy

Higher temperatures mean cooling systems need to work harder to keep servers within safe temperature ranges, leading to higher electricity consumption, lower cooling efficiency, greater wear on cooling equipment and potentially all server hardware if optimal temperatures aren't maintained, and higher operational and maintenance costs.

With cooling already one of the largest ongoing expenses for data centers, rising temperatures could compound the effects.

Summer 2026 is already proving to be a stress test for data centers in the Northern Hemisphere, with temperature records being broken across Europe and North America where around 50% and 46% of capacity respectively is located in chronic heat and drought regions. As much as 89% of Asia-Pacific capacity faces the same challenges.

Looking ahead, the report concludes that future data center investments should focus on future temperature projections rather than historical climate data, and the availability of power and water to support long-term cooling operations.

Companies could also look to locate campuses in low-risk areas, which have so far seen comparatively slower development.

"As digital infrastructure continues to expand globally, institutions that incorporate climate risk into site selection, underwriting, and capital allocation decisions will be better positioned to identify resilient opportunities and manage long-term exposure," the firm says.

• 119 malicious Edge extensions flew under the radar
• They installed harmful code days after extension installation
• It's proof that static code review is no longer sufficient

Microsoft says it has taken down 119 malicious extensions from the Edge Add-ons store after "proactive threat hunting" revealed a campaign that's been dubbed StegoAd.

As part of the program, the company also had to suspend more than 90 developer accounts associated with the dodgy activity.

Believed to have been active since at least 2021, it's believed that the malicious browser extensions had been downloaded a total of 2.6 million times.

Microsoft removes 119 'StegoAd' malicious extensions

The campaign was so broad that the extensions didn't just occupy one category: ad blockers, VPNs, video downloaders, translators and utility tools like PDF exporters were all ploys for the malicious extensions.

This particular campaign got its name from the type of tactic used – steganography is the name given to hiding malicious code inside seemingly harmless files. PNG images, SVG graphics and font files had hidden JavaScript embedded inside to bypass traditional antivirus tools and web filtering.

Once installed, Microsoft says they remained dormant for three to five days to avoid detection before going on to steal browser credentials, redirect users to malicious websites, manipulate affiliate links for financial gain, download additional malicious code and even communicate with C2 servers for updated instructions.

"The StegoAd campaign demonstrates that browser extensions remain a potent and evolving attack surface," Microsoft wrote, admitting that even its own safeguards had missed these dodgy extensions.

The report also concludes that static code review alone is no longer sufficient, because extensions and other installations can download malicious code long after they were first installed.

For developers themselves, Microsoft recommends being as clear as possible by not obscuring code, requesting only the necessary permissions to build trust, and report any suspected impersonation.

It’s clearly the wild, wild west when it comes to the race to build bigger and faster Artificial Intelligence systems.

The current backbone is built on Nvidia’s GPU architecture, allowing researchers and labs to build more and more sophisticated models that seem to leapfrog each other weekly.

The underlying technology required to achieve this is hundreds of thousands of these GPU cards sprawled across data centers throughout the US, drawing ever-increasing power and natural resources to push this frontier of human imagination.

In our rush towards our promise of a better future, are we laying the groundwork for this generation’s environmental calamity, much like previous generations had asbestos, microplastics, and lead?

Look no further than the SpaceX IPO, marketed as the largest IPO in history with a stunning $1.75 trillion initial valuation!

Buried deep inside, however, lies xAI – the AI company behind Grok. Grok runs from two massive data centers named Colossus 1 and 2.

What isn’t being discussed is that the NAACP is suing xAI for illegally installing air-polluting gas turbines in Mississippi that emit carcinogens into the air that residents breathe.

The question is not whether the power was needed, but what costs we are willing to accept, especially when, in this case, the demand releases toxic fumes into the air so people can interact with a chatbot.

An ominous situation

What makes the situation most ominous is that we are running a four-legged stool and waiting to see which breaks first. The first leg, of course, is computing power.

GPUs, while becoming more efficient, are at the same time getting denser and requiring more power. A traditional server rack used to draw about 5 kilowatts (kW) of power. GPUs upended that, with power draw climbing to 50 kW, 80 kW and now 140 kW per rack.

Indeed, Nvidia itself has proclaimed that its Kyber systems will draw an unheard-of 600 kW per rack by 2027. Whether this is true or not, the reality is that the power draw and density to run increasingly sophisticated hardware is going up; we just don’t know the timeframe of how we get there.

The second leg is what AI aficionados refer to as Jevon’s paradox. This is an economic principle that, as technology becomes cheaper to use and deploy, the use cases increase, so net consumption actually goes up even though the cost to run it keeps getting cheaper.

The third leg is that virtually all the free cash flow of megatech is going into deploying this IT infrastructure at an increasingly rapid rate. The projection for this year is $700 billion in capital expenditures, ballooning to over $1 trillion next year, and who knows after that.

At the same time, no one knows what limits Wall Street’s checkbook will be to fund additional expenditures for other companies. All this equipment must go into data centers somewhere.

Finally, we have a total lack of public policy on how to deploy these in sustainable ways, partially because there is a rush to get this equipment online, partly because no one really knows what the negative externalities could be, because no one has built up this much infrastructure at this scale and speed before.

The problem, of course, is if the part of the stool that deals with negative feedback to the environment and communities we live in is the first to break. It may not be easy to turn back once Pandora’s box has opened.

The need for policy sooner than later

Of course, we don’t have to wait for this all to happen. Thoughtful policy by the industry can get ahead of what could be a calamity by considering the “what ifs” before they happen.

Plus, by investing thoughtfully now, it stays ahead of what could be burdensome, time-consuming legislative efforts that lead to more regulation and compliance. In addition, consideration of things like total load (versus individual load) and the variance it might cause to the environment would be another consideration.

For example, a single independent data center building with a closed-loop water system may not draw that much water (other than initial fill, makeup needs, or during maintenance). However, if you start looking at the compound effects of how many additional buildings are going onto the campus, and how many campuses will be built in the next 10 years, then the impact could be considered before things go awry.

In the fast-moving race that is the AI space, the trickle-down effects on communities and the environment are something we should be looking at today.

Rather than letting companies run roughshod over people and cities, the data centers that they portend to inhabit can be good corporate citizens, which is important because these facilities typically have a useful life measured in decades.

The plan is for the data centers to be around for a long time, and so should everything else around them in a sustainable way.

We've listed 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

According to reports, ChatGPT hit the one-billion-active-user mark in May. This makes it the fastest app to achieve this feat to date. I think this is enough for you to realize that it’s more important than ever to get your brand featured/ mentioned/ cited on ChatGPT and other LLMs online.

Unfortunately, this is not as straightforward, and before you start cooking up a new strategy to address this new era of search, you must understand where you stand with your brand right now.

This is where you start tracking how your brand is being perceived by these LLM models and how often you get picked over your competitors. This guide is here to answer exactly that.

For this, I tried a bunch of different methods to find the most economical and credible methods to track your business’s visibility in AI search results. Here’s everything that I found legitimate and actually working well.

What does AI visibility mean?

Before you pick a tool, you need to know which question you're answering. Most people merge three very different things into one word ("visibility") and then wonder why their data is a mess.

There are three questions, and they need three different methods:

• Are you mentioned or cited in AI answers? When someone asks ChatGPT or Perplexity a question in your category, does your brand show up in the response? This is the one most people mean by "AI visibility."
• Are people clicking through from AI tools to your site? A mention and a click are not the same thing. You can get named in a hundred answers and still get zero visits, because the user got what they needed and moved on.
• Are AI bots even crawling your content? If LLM crawlers like GPTBot and PerplexityBot can't reach your pages, you're not eligible to be cited in the first place. This is the plumbing nobody checks.

Keep these separate in your head. A tool that nails question ONE tells you nothing about question TWO. I've watched smart marketers celebrate a rising "mention rate" while their actual AI traffic sat flat, because they were measuring two different things and treating them as one. Now, the methods.

Tracking AI visibility in AI search results: Free methods

Here is how you can get started with tracking your business in AI search results for free.

Google Search Console (GSC)

Google Search Console added a dedicated generative-AI report on June 3, 2026, and it's the freshest free tool you have. For the first time, it breaks down how often your pages appear inside AI Overviews, AI Mode, and the AI features in Discover, instead of burying that data in your overall numbers.

This report measures the impressions, broken down by page, country, device, and date. So you can finally see which of your URLs Google's AI is actually pulling into answers, and watch that trend over time.

Now the honest caveats, because this isn't magic. It shows impressions only. There's no click data, click-through rate, or query data yet (Google says more is coming soon). The three AI surfaces are blended into one bucket, so you can't separate AI Overviews from AI Mode. And it's rolling out to a subset of sites, within the UK first, so it might not be in your account on the day you read this.

Check anyway. If it's there, set your baseline now.

A manual prompt log

The oldest method is still the cheapest, and it works on every platform. Write 10 to 30 questions your buyers actually ask, run them across ChatGPT, Gemini, Perplexity, and Claude on a fixed schedule, and log the results in a Google Sheet.

Try to pick your prompts and then leave them alone. AI answers wobble run to run, so a single check is noise. Same prompt, same day next week, every week. The trend over a month is the signal, not any one result.

Is it tedious? Yes. Will it cost you a dollar? No.

Plus, it forces you to actually read the answers, which teaches you more about why you're getting left out than any dashboard ever will. If you only do one thing this week, do this.

Tracking AI visibility in AI search results: Custom workflow

If the manual log gets old (it will), you can automate it without paying for a full visibility platform using a workflow automation tool like n8n.

It is a workflow automation tool, and its template library has ready-made workflows that do exactly what the manual log does on autopilot. The pattern is that it reads your prompts from a Google Sheet, sends each one to ChatGPT, Gemini, and Perplexity, has a step check for every answer for your brand, and writes the results back to your sheet on a schedule.

Browse n8n's template search for "brand visibility" and you'll find a few versions to start from.

The cost for this process is genuinely small. You bring your own API keys, and on cheap models, the OpenAI and Gemini calls run for a pretty small amount. Perplexity is the main line item, because its Sonar API adds a per-request search fee of roughly $5 to $14 per 1,000 requests on top of tokens. For a 25-prompt list run weekly, you're looking at low digits per month.

Tracking AI visibility in AI search results: Using A paid tool

If you'd rather not build anything, a purpose-made tool collapses all of the above into one screen.

Getting these tools set up is generally fast. Typically, you start by entering some brand information such as your website, name, and industry. The tool will then work on your behalf, returning insightful information such as:

• Your "share of voice" across different AI platforms
• Performance breakdown based on model (such as ChatGPT, Gemini, and Claude)
• Which websites and brands are being referenced where you want to be
• Gaps where competitors aren't mentioned
• Grading your website content for AI readiness

When is paying worth it? When you're tracking this across several brands, or you need a report someone else will read, or your time is worth more than the subscription.

For a single brand on a tight budget, the free tier plus a prompt log will take you a long way first. There's no shame in starting free and upgrading when the manual work starts eating your week.

So which one should you actually use?

Start with where you are. If you have no budget, turn on the Google Search Console AI report today, then run a manual prompt log every week. That combination answers all three questions: who Google's AI surfaces, who clicks through, and who gets mentioned across the chatbots.

If the manual log gets painful, automate it with an n8n workflow for the chatbots. You'll spend a few dollars a month and save yourself hours.

If you're doing this for more than one brand, or you need clean reports without the building, a subscription tool earns its keep.

One last thing, because it's the mistake I see most. Don't track AI visibility once and call it done. The whole reason the timestamp matters, the whole reason you sample weekly, is that AI answers change constantly.

The number you pull today is a snapshot. The line you build over a month is the actual story. The brands building that baseline now are the ones who'll know exactly where they stand when AI search becomes the front door to everything.

Frequently Asked Questions

Organizations are under mounting pressure to do more with less, particularly in highly regulated industries and the public sector.

Budgets remain tight, with recent research highlighting that up to 43% of finance leaders cite tight budgets as their top barrier to achieving goals.

Leadership teams are being asked to modernize operations while maintaining service levels.

In response, many have turned to AI tools.

The logic makes sense. Large language Models and AI-powered automation tools promise faster workflows, reduced administration and meaningful productivity gains for resource-constrained organizations.

Yet productivity gains alone do not automatically translate into operational change.

The AI paradox

This is the emerging AI job paradox. Organizations are investing in AI to create capacity, but many lack the workforce flexibility needed to absorb, redeploy or realize those gains in practice.

In many cases, the tools and technology are working exactly as intended. Employees are completing tasks faster, administrative workloads are shrinking, and teams are identifying new efficiencies. The challenge is that most organizations still operate within workforce structures designed for a different economic environment.

For years, workforce planning relied heavily on natural attrition as a mechanism for change. Employees would move roles, retire or move to other organizations, creating space for organizations to reshape teams and redistribute work. But that model is now under pressure.

In slower labor markets, employees are moving less frequently, reducing organizations' ability to restructure organically across many sectors, particularly public services and regulated industries where stability is often prioritized. At the same time, organizations are operating under headcount limits, making large-scale restructuring politically, financially, or operationally difficult.

The result is a workforce environment that is less flexible than many AI strategies assume.

This creates a disconnect at the heart of current AI adoption. Organizations can generate efficiency gains through automation, but they often lack a clear mechanism to convert those gains into meaningful organizational capacity. If an AI tool reduces the time needed to complete a task by 30%, what happens next? In many cases, the answer is surprisingly unclear.

The appearance of transformation without altering outcomes

When the employee remains in the same role, within the same structure, work may become faster, but the organisation itself does not materially change. Productivity increases are identified in theory but struggle to appear in financial performance, service delivery improvements, or workforce optimization.

This is why many early AI programs are creating the appearance of transformation without altering operational outcomes. The risk is that organizations begin to treat AI as a workaround rather than a catalyst for redesign.

The risk is that organizations begin to treat AI as a workaround rather than a catalyst for redesign. That approach may deliver short-term improvements, but limits the long-term value organizations can extract from the technology.

As AI tools reduce administrative effort and streamline repetitive work, organizations gradually accumulate pockets of excess capacity across departments. Without a strategy to redeploy that capacity, the gains are often diluted through inefficiency, duplicated work, or simply absorbed back into existing processes.

In effect, organizations become more efficient at the task level while remaining unchanged at the operational level.

The missing link of capacity governance

For many organizations, the missing link is capacity governance. Capacity governance means actively managing the operational impact of productivity gains rather than assuming efficiencies will naturally convert into better outcomes. It requires organizations to treat workforce transformation as an operational discipline, not simply a technology initiative.

That involves asking difficult questions about which roles are being reshaped, how can newly created capacity be redirected and which departments are facing growing demand. AI is creating a reality where work should be reorganized, reprioritized and organizations should be able to completely reshape structure and prioritize accordingly. Assessing organizational design and structure is as important as diving into technology and tools being used to creative proactivity and productivity.

Leading organizations are beginning to approach AI adoption in this way by redesigning work around AI-augmented tasks. In practice, this means breaking roles down into component activities and identifying which tasks are best handled by AI, which require human judgement, and where employees can shift toward higher-value work.

For example, a compliance professional may spend less time reviewing routine IT documentation and more time handling complex cases that require interpretation and decision-making. Customer service teams may automate repetitive interactions while focusing human effort on vulnerable or high-priority users. Operational staff may use AI to accelerate reporting and analysis while dedicating more time to strategic planning.

These organizations are actively redesigning workflows and managing workforce capacity in response to the changes AI creates.

An increasingly important shift

That shift will become increasingly important over the next several years. Across regulated sectors, demographic pressures, budget constraints, and rising service expectations are colliding at the same time as rapid advances in AI capability.

Organizations cannot rely indefinitely on incremental efficiency gains layered onto outdated workforce structures. Nor can they assume AI alone will solve structural productivity challenges.

Without operational redesign, many institutions risk creating a form of productivity stagnation where technology improves individual output but fails to generate meaningful organizational transformation.

There is a broader strategic implication. As AI adoption accelerates, organizations that successfully govern and redeploy capacity will gain a significant operational advantage. They will be able to respond faster to demand shifts, move talent into critical areas more effectively, and create more adaptive workforce models.

Those that fail to address the workforce dimension of AI may find themselves trapped between rising expectations and rigid organizational structures. This is why the future of AI adoption is likely to depend less on the sophistication of the models themselves and more on how organizations choose to reorganize around them.

The next phase of AI will be about building institutions capable of converting efficiency into agility. That requires a willingness to rethink roles and move talent across functions in ways many organizations have historically resisted.

Technology may create the opportunity for productivity gains. But without workforce flexibility and clear capacity governance, many of those gains risk remaining theoretical.

The organizations that recognize AI is not just a technology shift, but operational , will be the ones making those strides forward.

We've featured the best employee management 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

• US DOJ has seized nearly 400 domains
• The sites were being used to illegally stream World Cup games
• Users of the sites were exposed to malware, data theft, and other threats

Almost 400 domains have been seized as part of Operation Offsides - a coordinated global effort to take down sites illegally streaming the FIFA World Cup 2026.

The sites were seized by the US Justice Department's Criminal Division for violating copyright and intellectual property law.

The takedowns were coordinated by members of the International Computer Hacking and Intellectual Property (ICHIP) network.

US and friends enforce the offside rule

Many of the seized domains now display a banner explaining that the website was seized as part of Operation Offsides. “This action was taken to protect consumers and enforce intellectual property rights worldwide,” the banner states.

Back in May 2026, the FBI warned that thousands of domains were being registered ahead of the World Cup, with most set up with the intention to scam fans looking for cheap tickets, access to streaming services, and those looking for discounted merchandise. It appears that Operation Offside was focused on disrupting streaming sites in particular, rather than taking down the wider scam networks associated with these domains.

“We have seized hundreds of domains, used to illegally stream World Cup matches for profit, to disrupt the international networks that profit from the global popularity of the World Cup,” said Assistant Attorney General A. Tysen Duva of the Justice Department’s Criminal Division.

“This operation illustrates the Department’s respect for intellectual property rights and the responsibility of the United States as a host nation to protect the FIFA World Cup from criminals. The Criminal Division will continue to disrupt and, where appropriate, seek to prosecute these sites and the subjects responsible for this criminal activity.”

In many cases, the networks of fake domains offering cheap or free access to streaming services are run by cybercriminals deliberately operating at a loss in order to attract users to their services. In return for accessing the streaming site, the domain will use the user’s local network as an exit node for the cybercriminal network, obscuring their traffic and making it appear legitimate.

Unfortunately for the user, who may think they have just found free access to every World Cup game, their network and IP address could be used to distribute malware, cybercriminal communications, and illegal content such as stolen data and exploitative materials - including child sex abuse material.

8849 Tank 5: 30-second review

The Tank 5 represents the most complete version of a concept that 8849 has been refining for several generations. Where earlier models asked buyers to accept trade-offs between size, battery and projector quality, the Tank 5 attempts to resolve all three at once. The result is something that genuinely has no mainstream equivalent.

There is a very specific kind of person the 8849 Tank 5 is made for. They work far from mains power. They need a phone that survives punishment. And occasionally, they want to project something onto a wall, a tent, or the side of a cliff face. For that person, no mainstream smartphone comes close.

Where previous models asked buyers to accept mid-range processors and modest projection quality, the Tank 5 brings flagship silicon to the category for the first time. The MediaTek Dimensity 9400e is a genuine top-tier chip in 2026 terms, and finding it in a device this rugged at this price is a pleasant surprise.

The headline numbers are hard to ignore. A 17,600mAh battery. A 3K AMOLED display peaking at 3000 nits. A 2K TI DLP projector with 220-lumen output and laser autofocus to four metres. Triple 50-megapixel rear cameras. Android 16 out of the box. On paper, this reads like a wish list written by someone who spends more time on a mountainside than in a meeting room.

The catch, as ever, is the physical reality. At 715 grams and nearly 34 millimetres thick, the Tank 5 is not a device you slip into a jacket pocket. It is a tool. And like any serious tool, it rewards the user who actually needs what it offers.

Is this one of the best rugged phones? That entirely depends on what features you value. For those who don’t care about size and weight but want battery capacity, exotic features and colourful display, this is certainly a contender.

8849 Tank 5: price and availability

• How much does it cost? $900/£697/€809
• When is it out? Available now
• Where can you get it? Direct from the maker or via an online retailer

The Tank 5 launches at $999.99 in pre-order configuration with the standard retail price listed at $1,599.99. However, now that those numbers have come down somewhat, with the US pricing at $899.99. In the UK, it’s £696.87, and across Europe it is €808.52, direct from the 8849tech website.

To put that into perspective, the Tank X launched at $549.99 early bird against a $1,049.99 RRP. The Tank 5, therefore, represents a substantial step up in cost, but is justified by improvements in specification.

Currently, the Tank X costs $699.99 from Amazon.com and £599 via Amazon.co.uk, making the Tank 5 pricing less of an uplift for those in the UK.

When considering alternatives, it's worth noting that remarkably few phones have a projector, and none of them has one with the same specification as the DLP unit in this phone.

Other than the Tank X and Tank 4 Pro by 8849, rugged phones with projectors include the Ulefone Armour 34 Pro Plus and Oukitel WP100 Titan. From Amazon.com, the Tank 4 Pro is $789.99, and the Ulefone Armor 34 Pro Plus is $594.99. The Oukitel WP100 Titan isn’t available from Amazon, but can be bought directly from the makers for $869.99.

The cover these in a spec comparison, the Tank X offers a slower Dimensity 8200 SoC and a lower resolution 1080p projector. The Tank 4 Pro only has a Dimensity 8300 SoC, but only a 720p projector. And the Ulefone Armor 34 Pro has the slowest processor, with a Dimensity 7300, and a weird projector resolution of 854 x 480. And finally, the Oukitel WP100 Titan has the same processor as the 34 Pro, and the same odd resolution to its 100 lumen projector.

Most of these have more battery capacity than the Tank 5, but critically, their projectors don’t come close to being a match.

Based purely on value for money, the 8849 Tank 5 is a winner.

• Value score: 4.5/5

8849 Tank 5: Specs

8849 Tank 5: Design

• Thick and heavy
• Vents for projector on both sides

Pick up the Tank 5, and the sheer physical commitment of it registers immediately. At 177.1 x 87 x 33.8mm and 715 grams, this device sits somewhere between a phone and a piece of site equipment. The chassis combines brushed metal plates, rubberised TPU corner armour and polycarbonate panels, and the build quality throughout is excellent. Seams are tight. There is no flex anywhere. It feels, correctly, like something engineered to survive conditions that would end a normal phone's life in seconds.

The layout of the Tank 5 is dictated by one thing above all others: cooling. The projector needs active thermal management, and the ventilation grille that serves it runs across a significant portion of the left frame. That grille is not a cosmetic addition. It is a structural constraint, and it shapes where every other control can go. The volume controls and shortcut buttons sit above it, pushed into the upper corner.

Below it, separated by a visible gap of bare metal, sit red and silver programmable buttons. The vent physically cleaves the left side into two distinct groups. It is an honest piece of industrial design. Nothing is where it is by accident, but you need to be careful where you place your fingers when the projector is running.

The right side tells another story. A second vent grille occupies the upper portion, again serving the projector cooling system, and below it, the frame is noticeably cleaner with only a single large silver power key with its integrated fingerprint sensor. The visual contrast between the two sides is immediate. Where the left side is busy by necessity, the right side breathes.

The rear is where the Tank 5 makes its most direct statement of intent. The camping light bar dominates the centre of the chassis, a wide rectangular strip that illuminates with 1200 lumens and RGB warning modes. Above it sits the camera module, with two main lenses at the top right, the night vision lens below left and four infrared LEDs in a horizontal strip. The "TANK PROJECTOR INSIDE" text sits between the camera cluster and the lamp, which is either endearing or unnecessary depending on your tolerance for that kind of thing. The projector aperture itself is at the top edge, along with another flashlight and the SIM card slot.

Compared to the Tank X, the Tank 5 is a more resolved design. The Tank X was 91.8mm wide and 750 grams. The Tank 5 pulls back to 87mm and 715 grams. Those numbers sound modest, but the difference in hand is real. The Tank X sat at the outer limit of a comfortable two-handed span. The Tank 5 comes back just enough to feel deliberate.

What has not changed is the fundamental reality of the form factor. This is not going into any conventional pocket. A jacket chest pocket, a belt holster or a vehicle mount is the practical answer for daily carry. That is the honest consequence of fitting a projector, a 17,600mAh battery and an active cooling system into a single chassis, and the Tank 5 makes no apology for it.

Design score: 3.5/5

8849 Tank 5: Hardware

• MediaTek Dimensity 9400e
• AMOLED display
• 17600 mAh fast charge battery

The Dimensity 9400e is an important specification decision. Most rugged phones at this price use mid-range silicon from three to four years ago. The Tank X stepped up to the Dimensity 8200, and that was already a significant advance. The 9400e goes further still, using a 4nm process with Cortex-X4 architecture.

The 'e' suffix indicates a slightly binned or optimised version of the full 9400. In practice, real-world performance at this specification level is well beyond anything a rugged phone buyer has previously been able to access at this price point. The comparison with the Immortalis-G720 MC12 GPU is also notable for mobile gaming and sustained workloads.

The Tank X, launched in February 2026, was the first 8849 device to achieve 1080p DLP projection at 220 lumens. The Tank 5 carries that hardware forward and adds 2K output with laser autofocus rated to four metres. That four-metre figure is a meaningful practical upgrade, because how far away from the wall you can get dictates the size of the projection. Earlier models topped out at three metres, and the keystone correction has been expanded from two to four points. What this means in practice is a more practical outdoor cinema experience.

The 220-lumen figure also requires context, as it's extremely competitive within this product category, where some devices can only manage 100 lumen. The Ulefone Armor 34 Pro, a direct rival, managed 150 lumens at a lower resolution. However, any projector of this size still demands reasonable darkness for a clear image. The spec sheet implies more than the physics can deliver in daylight, as it needs to be twilight outside or in a shaded room for a good experience. But, this is more than any other brand is currently offering.

My only complaint about the projector is that not long after it is activated, the cooling fan starts up and can be a little noisy. This won’t be a problem if you are projecting an action blockbuster, but it isn’t ideal for anything with quiet audio and dialogue.

Earlier Tank models used AMOLED panels, but the Tank 5 moves to a 3200x1440 native resolution. The 3000 nits peak brightness is a headline figure borrowed from premium consumer phones, and the results are spectacular.

Considering how good the display is, that 8849 went with Panda Glass to protect it, and not stronger Gorilla Glass Victus is disappointing.

That choice looks entirely cost-driven, but conversely, the battery technology used in the Tank 5 doesn’t hold back.

Seventeen thousand six hundred milliamp hours is a substantial number, even by rugged phone standards. The Tank 4 used an 11,600mAh cell. The Tank X brought the 17,600mAh configuration back after the Tank 3 and Tank 3 Pro sat at around 23,800mAh. The Tank 5 pairs that 17,600mAh capacity with 120W charging, which can be recharged in just over 90 minutes, incredibly. It does this by using a dual-cell design, so it works like two 8800mAh batteries.

The addition of 25W reverse charging makes the Tank 5 a practical field power bank. At that capacity, it can top up a standard 5000mAh smartphone three times over. For field workers and outdoor users, this is a genuinely compelling capability, even if you lose some of the capacity in the transfer.

Another full-cream feature is Wi-Fi 7. Most rugged phones ship with Wi-Fi 6 at best. The inclusion of eSIM alongside dual nano SIM slots gives the Tank 5 flexibility that neither predecessor offered. For travellers, the combination of eSIM, dual-frequency GPS and extensive 5G band support is a strong package. The addition of eSIM also means you can ditch one of the physical Nano SIMs, fit a MicroSD card in the tray, and still have two network services.

With a few minor exceptions, the specifications of the Tank 5 place it in the premium rugged phone bracket.

• Hardware score: 4.5/5

8849 Tank 5: Cameras

• 50MP main + 50MP telephoto + 50MP night vision
• 32MP on the front
• Four cameras in total

The 8849 Tank 5 has four cameras:

Rear camera: 50MP Samsung ISOCELL S5KGN1SP, 50MP OmniVision OV50C40, 50MP OmniVision OV50D40
Front camera: 32MP OV32B40

The camera cluster on the Tank 5 is a significant rethink compared to what came before it. On the Tank X, the three sensors were arranged in a vertical strip, with the 8MP telephoto the obvious weak link in the lineup.

The Tank 5 design addresses this directly with all three rear sensors now 50 megapixels, and this is a much more balanced package.

The two primary lenses sit at the top right of the module in a paired arrangement, noticeably larger than the sensors below. The night vision camera sits below and to the left, with the four infrared LEDs arranged in a horizontal strip beneath it. It is a purposeful, asymmetric layout that prioritises the main imaging hardware without pretending everything is equal.

The main sensor is a Samsung ISOCELL GN1 with phase detection autofocus. That’s a well-regarded chip with a strong track record on mainstream flagship devices. It is a genuine step up from the Sony sensor used on the Tank X, which, while capable, was hampered by the overall image processing pipeline more than the sensor itself. The GN1 brings improved low-light capture and better dynamic range handling. It might not be the 108MP sensor we’ve seen elsewhere, but the arrangement is decent and effective.

The caveat to this choice is that, while there is a PRO mode in the camera application, there are relatively few other special modes or HDR. This camera doesn’t do panoramas or slow motion, though it can do timelapse and super-resolution.

Video capture can be in resolutions up to 4K, but there is no control over the framerate, irrespective of the capture resolution.

The telephoto upgrade is the most important change for practical use. Going from 8MP to 50MP with optical image stabilisation transforms what the camera can do at a distance. The Tank X telephoto was essentially a crop tool dressed up as a lens. The Tank 5 telephoto is a genuine optical system with 3x zoom, and OIS means it remains usable in the kind of low-stability field conditions this phone is designed for. Hiking, construction sites, and vehicle-mounted use. All of these benefit from stabilisation in ways that a rugged phone buyer will actually notice.

The night vision camera is the one area where the upgrade is less straightforward. The Tank X used a 64MP OmniVision sensor for its night vision work, paired with four IR LEDs. The Tank 5 drops to a 50MP sensor for that role, but it's just as effective. The megapixel reduction is not inherently a problem, as night vision performance depends far more on sensor size, IR LED power, and processing than on raw resolution.

The front camera remains unchanged from Tank X, a 32MP with an OmniVision OV32B40 sensor and an f/1.7 aperture. A competent selfie and video-call solution by definition.

What the cluster cannot do is match a photography-focused flagship. The Samsung Galaxy S25 Ultra or Pixel 9 Pro operate in a different league for computational photography, colour science and video processing.

That comparison is not unfair, but it is also largely irrelevant. The Tank 5 camera system is built for documenting field conditions, capturing evidence, seeing in the dark and shooting at range from an unstable platform. On those terms, it is well-equipped, and the results are better than those of some phones with more megapixels to play with.

One disappointment is that with a camera that can capture 4K video, and play that back nicely on the AMOLED screen, you won’t be using that display to see high-quality streams, because 8849 wouldn’t pay for Widevine L1 encryption.

8849 Tank 5 Camera samples

• Camera score: 4/5

8849 Tank 5: Performance

• 4nm SoC
• Premium performance

With so few competitors in this niche, it seemed logical to compare this 8849 to its predecessor, the Tank X.

And what we quickly learn here is that the Dimensity 9400e is a beast compared to the Dimensity 8200, outperforming it in every respect. The most sobering score is the 3DMark Nomad Lite performance, where it is nearly 300% better at this demanding benchmark.

But the other takeaway here is that you can have a more powerful SoC, without it impacting the power consumption dramatically. With the same battery capacity, the running time is remarkably similar to that of the Tank X, even though it does more over that time.

But where the Tank 5 really shows the most dramatic improvement is in the recharging of the battery, which took hours on the Tank X, and can be done in close to 90 minutes on the Tank 5. In just thirty minutes, being able to recover 46% of the battery is excellent.

Overall, the Tank 5 is a top-tier performer for whatever apps you wish to install.

• Performance score: 4.5/5

8849 Tank 5: Final verdict

The 8849 Tank 5 is the most technically accomplished rugged projector phone yet made. It takes the Tank X's headline concept and delivers a genuine performance upgrade at every level. The Dimensity 9400e is fast. The 3K AMOLED display is bright. The 17,600mAh battery is enormous. The projector is now 2K capable with laser autofocus at four metres.

None of that comes without compromise.

The device is very heavy, and the projector still benefits from darkness, whatever the spec sheet implies. And the lack of a published software update policy is a concern worth flagging for any buyer planning to keep this device for several years.

For the specific buyer this device is designed for, the Tank 5 is a compelling proposition. It does things no other phone can match in a single chassis. That remains, as it has been throughout the Tank series, both its greatest strength and its clearest self-selection filter.

Yes, it might have the ideal characteristics to make a decent boat anchor, but when you pack this much technology into a phone, it was never going to be lightweight and slim.

Should I buy a 8849 Tank 5?

Buy it if...

Don't buy it if...

Also Consider

For more ruggedized devices, we've reviewed the best rugged tablets, the best rugged laptops, and the best rugged hard drives

Most large enterprises have already experimented with AI in some form.

They have tested copilots, automated workflows, analytics platforms, content generation tools and customer service assistants, and initial reactions are often positive. Demonstrations create excitement, leadership teams engage quickly and investment follows.

Yet many organizations still struggle to move beyond isolated successes. Adoption slows, usage becomes inconsistent and AI initiatives gradually lose visibility inside day-to-day operations.

What begins as a strategic priority often becomes another innovation program that never fully reshapes the business.

At that point, organizations frequently conclude that the technology is not mature enough.

In reality, the bigger obstacle is often structural rather than technical. AI adoption rarely fails because the tools are incapable.

More commonly, organizations fail to adapt their operating models, incentives and decision-making structures to support meaningful change.

Fragmented ownership weakens adoption

One of the biggest barriers to enterprise AI adoption is unclear accountability: technology teams manage IT infrastructure, governance, security and vendor relationships; innovation teams run pilots; individual business units experiment independently; and senior executives communicate ambition and strategic direction. Yet in many organizations, nobody owns AI adoption from end to end.

But without clear ownership tied to operational outcomes, AI initiatives often become disconnected from how work actually happens. Teams are encouraged to experiment but lack the authority to redesign processes or redefine how decisions are made. Pilots move forward without long-term accountability and successful experiments fail to scale beyond individual departments.

As a result, AI can exist inside the organisation without becoming embedded into its operating model. The technology itself may function well, but adoption stalls because no one is responsible for turning experimentation into lasting behavioral change.

Why technology teams cannot solve this alone

This challenge becomes particularly visible inside platform, data and IT functions. These teams are frequently tasked with enabling enterprise AI adoption by assessing vendors, integrating systems, securing data environments and establishing governance frameworks. At the same time, they are expected to minimize operational risk and ensure compliance requirements are met.

However, they rarely control how individual departments actually work. Technology teams cannot independently redesign sales processes, restructure customer support operations or redefine HR workflows. They can provide tools and infrastructure, but they are not usually empowered to drive organizational change across business functions.

That imbalance creates predictable tension. If AI deployments introduce operational or security risks, technology teams are held accountable. But if adoption slows because departments resist changing established processes, responsibility becomes far less clear.

Over time, this dynamic naturally encourages caution. Teams carrying significant risk without the authority to control it often become more conservative in how aggressively they push transformation initiatives forward.

Incentives matter more than strategy documents

Many organizations also underestimate how strongly incentives shape adoption behavior. A customer service team may be encouraged to use AI tools at the same time as being measured primarily on ticket throughput and response speed. Marketing teams may be asked to experiment with AI-generated content while facing scrutiny over even minor inconsistencies in tone or branding.

Compliance teams could be expected to support innovation even though they’re evaluated almost entirely on risk reduction. In each case, employees respond rationally to the incentives in front of them.

Meaningful AI integration almost always creates short-term disruption. Teams need time to test workflows, adjust processes and learn how humans and AI systems operate together effectively. Productivity can temporarily decline before long-term gains become visible.

If organizations continue rewarding operational stability above all else, employees will avoid experimentation regardless of how ambitious leadership messaging may be.

This is one reason many “AI-first” strategies struggle to move beyond isolated use cases. Declaring strategic intent is relatively easy. Adjusting performance frameworks, redefining accountability and creating room for experimentation is far more difficult.

Unclear governance creates hesitation

Another major obstacle to adoption is uncertainty around governance and operational boundaries. Many organizations still have not clearly defined what AI represents within their broader operating model. Is it an individual productivity layer? A centrally governed capability? A feature embedded into existing enterprise platforms? Or a specialist function managed by dedicated teams? When those questions remain unanswered, ambiguity spreads quickly.

Employees become unsure what usage is permitted, while managers struggle to establish consistent expectations. Technology, legal and compliance teams disagree on where accountability begins and ends - and in practice, this uncertainty often slows adoption more than technical limitations do.

Clear governance does not need to eliminate experimentation. In fact, successful organizations usually balance flexibility with oversight. Employees are far more likely to engage confidently with AI systems when they understand where experimentation is encouraged and where stricter controls apply. Without that clarity, even capable tools can remain underused.

AI transformation is an operational challenge

For CIOs and senior technology leaders, this requires an important shift in perspective. AI transformation is often framed primarily as a technology modernization effort focused on infrastructure, integration and data readiness. Those foundations remain essential. Without them, large-scale deployment is impossible. However, technical readiness alone does not determine adoption outcomes.

The organizations making meaningful progress with AI tend to treat it as an operational redesign challenge rather than simply a software rollout. They integrate AI into existing workflows, align ownership with accountability and adapt governance structures to support new ways of working.

This also explains why many AI programs gradually shift from transformational ambitions into smaller experimental efforts. Experimentation is organizationally safer because it avoids forcing structural change. Unfortunately, it also limits long-term impact.

Successful organizations tend to share several characteristics. They establish clear executive accountability for measurable outcomes linked to AI adoption. Rather than prioritizing short-term stability, they align incentives with workflow evolution. By integrating AI directly into operational systems, they aren’t left to rely on disconnected standalone tools. And they define governance boundaries clearly enough that employees understand how AI should be used.

Notably, none of these are primarily technical decisions. They are organizational and leadership choices.

The real question organizations need to answer

When AI initiatives underperform, organizations often focus first on the technology itself. Vendors are reassessed, models are compared and infrastructure decisions are revisited. Sometimes, those issues do genuinely matter. Often, however, the technology is functioning adequately while the organisation surrounding it has not evolved enough to support adoption at scale.

That distinction is crucial because organizational barriers are solvable. Accountability can be clarified. Incentives can be redesigned. Governance structures can be simplified. Operational ownership can be aligned more effectively with responsibility.

Ultimately, adopting AI means changing how work gets done across the business. And that means the question facing enterprises today is no longer whether AI technology is capable enough to deliver value. It is whether they are prepared to redesign themselves around it.

We list the best employee management 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

• Box finds 2 in 3 ITDMs expect headcount to grow over the next three years as AI evolves
• New roles – like agent operators and automation specialists – will emerge
• Nearly all ITDMs admit stronger governance would help

New research from Box has claimed two in three (65%) UK IT decision-makers expect overall headcount to grow over the next three years, not shrink.

In fact, only 14% believe headcount will decrease, with the consensus generally pointing toward AI creating new jobs, not replacing human workers.

This shift in narrative comes at an important point for AI, as it evolves from experimentation to actual deployment.

AI is creating new roles, not eliminating them altogether

Box now believes we're entering the era of the 'agentic enterprise' where AI agents are embedded into everyday workflows and have organizion-wide knowledge and context, but even as they become more sophisticated and capable, the impacts on human workers are expected to be positive.

Of all the companies using or testing AI agents, only 8% say they've led to job losses today.

Among the roles more likely to be created are AI agent operators within IT teams – nearly half (48%) of all organizations are hiring for this role. Workflow automation specialists (32%), security, risk and compliance professionals (31%), change management and AI enablement roles (31%) and AI ethics and governance specialists (26%) are also crucial opportunities for human workers.

"As organisations embrace the next phase of AI, the focus is increasingly shifting from individual productivity gains to transforming how work gets done," EMEA President Samantha Wessels commented.

Though only around one in three companies are hiring for security professionals, the opportunities could grow further as nearly one in two (45%) experience an AI-related data exposure incident.

Rather than framing the opportunities as being capped, with either humans or AI helping to meet goals, Box proposes a higher ceiling where AI and humans can work together to push goals even further. Most ITDMs (94%) now believe better governance would help them adopt agentic AI faster.

Any new technology tool is a double-edged sword for a startup founder.

It can be a powerful force multiplier, accelerating processes, cutting back on costs, streamlining communication, and empowering time-saving automations.

But it can also become an attention trap that shifts a founder’s focus away from business goals and towards tool management.

Claude Code is a tech tool that fits that description.

As a platform that can help you do more with less, Claude Code can be a godsend for a startup founder. But its amazing capabilities can also trigger addictive behavior that ultimately leads to startup burnout.

As startup founders seek to harness the power of Claude Code, it’s important to keep the proper perspective. While there is a lot the technology can do for a startup, it can also very easily distract you from the work needed to build a solid and scalable business.

Claude Code can be empowering for founders, but also exhausting

My journey with Claude Code addiction was driven by a desire for efficiency, cost savings, and independence. When kept in balance, those three elements can be extremely beneficial for a startup founder. But Claude Code led me to a place where I lost that balance.

It started when I was introduced to OpenClaw. I knew my tech team was using vibe coding tools for their work, but I had never considered how it might help me with the tasks I was facing as a founder. My research revealed it could be very helpful, simplifying my day through task management. I started using it and was hooked by the way it increased my efficiency.

But there was a problem: OpenClaw was expensive. In the world of startups, cash flow is often an issue in the early phases of development, which is a key reason tech tools are so valuable. A Vibe coding platform can help a startup make progress without hiring a CTO or principal engineer, but it needs to be affordable to make sense.

When I started getting concerned about the expense of OpenClaw, I began looking for alternatives and found Claude Code. At that point, the platform had become more agentic, providing almost all of the same capabilities as OpenClaw. When I learned it was more affordable, I shifted from paying on the go with OpenClaw to a monthly Claude Code subscription.

With Claude Code, I had an efficient and cost-effective tech tool that unlocked a lot of tasks I had previously depended on others for. I could run my own sales outreach campaigns by just connecting Claude to my Google Workspace. I could draft email copies and update the marketing website. I could even code any software project I wanted to without getting programmers involved.

Claude Code was empowering, helping me carry out tasks that opened new doors. But it soon became exhausting.

Claude Code gives founders independence, but also gets them off track

Delegating is a critical skill for startup founders, giving them the bandwidth they need to focus on big-picture work. Without staying committed to delegating, founders can get caught up in day-to-day tasks, which not only slows growth but also leads to founder burnout.

Claude Code made it easier for me to do things myself, so I stopped delegating. For example, with Claude Code, I could manage changes to my website, aligning certain aspects to match the course I was plotting in other areas. While it was great to have the capability, leveraging it required me to invest a lot of time going back and forth with an AI agent to get the site exactly the way I wanted it to be.

As I saw myself spending hours at my desk using Claude Code to adjust the website, I realized I was losing the time I needed to accomplish much bigger tasks. That was when I knew I needed to limit my time with Claude Code and get back to delegating.

A dream come true - or not as the case may be

Getting hooked on the independence you can gain from Claude Code is completely understandable. For startup founders trying to bootstrap their way to success, the platform is a dream come true. Using it means you don’t need to make your budget even tighter by hiring a programmer or outsourcing a project to an expensive tech expert, and there’s no limit to the number of agents it can unleash.

The reality, however, is that Claude Code will never be an experienced senior programmer who understands your context, anticipates issues, and positions you to scale successfully. At best, it is a junior programmer, which means it will need a lot of help from you to get it right. For a busy startup founder, the extra attention Claude Code needs could be the very thing that pushes you into burnout.

Claude Code has triggered a generational moment in the business world, giving startups the capacity to gain valuable skills on a tighter budget. I see it changing entrepreneurship for the better, leading to an increase in business launches.

But founders who want to see long-term success need to know Claude Code isn’t a long-term solution. Using it to get off the ground will give you a beneficial boost. Using it indefinitely will rob you of the time and energy you need to take your startup from launch to growth and beyond.

We've reviewed the best laptop for programming: Top picks for professional programmers, coders, software engineers, and developers.

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

• Microsoft Teams is getting better bot protection
• Humans will now need to clear any bots or agents attending a meeting
• Developers will be able to register and pre-clear their agents

Microsoft is cracking down on bots infiltrating Teams meeting, bringing in a new technology which will let humans check all participants in a call are who they say they are.

Much like a nightclub bouncer, the new tool will require a human user check the identity of bots in the call's lobby, before the meeting commences.

The company says it has used a combination of "behavioral and infrastructure signals to identify bots with a higher degree of accuracy” to be able to boost Microsoft Teams' ability "to distinguish between bots and human participants as they join a meeting.”

Bots in Teams

Rolling out now, the launch comes as transcription and note-taking bots and agents are becoming an increasingly common sight in meetings - ostensibly to help participants recap and recall details, but these unwanted guests could also pose a security and privacy risk.

“Bots have begun joining meetings that participants never intended them to attend,” wrote Microsoft product marketing manager Meera Ajam wrote in a company blog post. “For example, after connecting a third-party service to a meeting, some users have found that its bot continues joining future meetings automatically.”

“Admitting a bot should be a deliberate decision, not something that happens by mistake,” Ajam added, noting that multiple clicks from a human will now be required for a bot to be allowed in.

If this sounds like unwanted extra hassle, then never fear - Microsoft says it has added a way for users to pre-check agents or bots - notably, “a registration path for independent software vendors (ISVs) that build meeting experiences for Microsoft Teams.”

“When Teams recognizes that marker, it can identify the bot as a known participant,” Ajam wrote.

This means developers will be able to register with Microsoft to make sure their tools are cleared for use in Teams, with Ajam noting the company is working with "a limited set of ISVs to preview this capability and validate the experience before broader availability.”

• OpenClaw has launched iOS and Android companion apps
• Users can remotely control their self-hosted AI agents
• Initial reviews say the apps could do with a bit of refinement

OpenClaw has released its first official mobile apps, giving users an alternative to earlier reliance on messaging platforms like Telegram and WhatsApp.

Importantly, the iOS and Android apps act as companions to an existing OpenClaw installation rather than standalone assistants, much like the Codex controller OpenAI has embedded into the ChatGPT app.

The apps connect to a self-hosted OpenClaw Gateway running on their own hardware, rather than in OpenClaw's cloud, which has been key to the startup's success.

OpenClaw now available for mobile

OpenClaw described three core functionalities for the mobile apps: "native mobile apps, finally; agents in your pocket; and channels, tasks, replies on the go."

Users will be able to receive and approve actions requested by the AI remotely, and get status updates for ongoing workflows. Remote action triggering and managing are also available.

The AI agent gained popularity due to its local-first stance, allowing users to control their own Gateway and ultimately reduce their reliance on Big Tech and insecure cloud.

But while the tool itself proved a hit, selling out Mac minis across multiple regions after users scrambled to buy cost-effective hardware to run the agent, its mobile app has been a bit more of a flop, at least to begin with.

Its Android version in particular has attracted criticism over its poor interface and usability – cosmetic niggles that don't detract from its core functionality.

9to5Google reports many initial reviews are mostly negative, with the app currently on a 2.2 star rating, amid multiple reports of the app being buggy, with users calling it “unusable,” unable to pair, and “the worst app I’ve ever used in my entire life.”

The apps were notably published by the OpenClaw Foundation – an open-source project that OpenAI now supports following its hiring of OpenClaw founder Peter Steinberger, who stressed the project would remain "open and independent."

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

• AMD Zen 5-based EPYC 9965 is on sale on eBay for just under $6,000
• With 192 cores and 384 threads in tow, AMD's offering can handle as much as 6TB of RAM per processor for AI-centric workloads
• The CPU not only offers the highest core count in all x86 CPUs to date, but also holds its own against the competition

When AMD unveiled its new EPYC CPUs in June 2024, one particular chip caught everyone's eye, given how outrageously powerful it was compared to the competition.

The AMD EPYC 9965 CPU was also the company's most ambitious server-grade offering to date, and one that has held its own against the competition with relative ease thanks to its yet-unmatched 192-core, 384-thread x86 configuration.

At a suggested MSRP of $14,813 (now $11,988) at launch, it is also one of AMD's most expensive products across the board, with only a few of its server-grade GPUs pushing noticeably higher.

Why the eBay listings are somewhat hard to digest

Despite being unveiled over two years ago, the EPYC 9965 CPU remains largely unchallenged by the competition, thrashing most benchmarks with relative ease. It remains the flagship server chip for AMD-based systems worldwide, with Team Red shipping it globally, including China.

The chip itself is a genuine monster. The EPYC 9965 packs 192 Zen 5c cores and 384 threads onto a single SP5 package, with a 2.25 GHz base, a 3.35 GHz all-core, and a 3.7 GHz maximum boost, 384 MB of L3 cache, twelve channels of DDR5, 128 lanes of PCIe 5.0, and a 500 W default TDP, AMD's highest yet. It is the densest part in AMD's 9005 stack, built for cloud providers aiming to cram the maximum number of virtual machines into a rack.

All in all, it is designed primarily to serve hyperscalers and cloud providers, with a focus on low-power, efficient cores that dominate such setups.

Even at its reduced reference price of just under $12,000, eBay listings selling the same CPU for just under $6,000 seem inexplicable, offering a sub-50% discount over current advertised prices and over 60% off its original launch price.

This is even lower than the last time we tracked this CPU in 2025, but there might be an explanation of sorts for what would otherwise be seen as an anomaly. Most server-grade CPUs trade at prices below MSRP, with OEM and negotiated discounts that quickly add up to significant savings for large-scale deployments.

The other explanation that often applies to older server-grade CPUs that often get cycled out or decommissioned by data centers does not apply here, however; the EPYC 9965 does not have a successor yet, and it is hard to imagine a currently available CPU replacing AMD's most capable chip yet in a data center just yet

Other reasons prices are low could include excessive inventory, canceled orders, or simply someone liquidating hardware in an industry that is increasingly GPU- and memory-focused to meet its AI needs.

Most industry players prefer a direct OEM solution, especially given AMD's support for PSB-locking (secure boot) on their CPUs for certain vendors, which makes them unwilling to purchase from 3rd-party sellers. This also raises the risk profile for buyers, who could unwittingly end up purchasing a vendor-locked CPU, making for a rather annoying testing and return process.

With a shift in spending towards GPUs and memory becoming prohibitively expensive, hardly helping matters, while enterprise consumers are still buying up the EPYC 9965, prices might reflect a market reality: they are unlikely to be scouring eBay for their next 192-core monster CPU purchase, even assuming they have a compatible motherboard and memory modules to match.

• Disney Imagineering is planning to use Adobe's Firefly AI services to build the next generation of theme park rides based on its own IP
• Disney says the move builds further on decades of collaboration between the two companies
• The move is being seen as an important win for Adobe as it navigates increasingly pessimistic investor sentiment with its own AI offerings

Adobe and Walt Disney Imagineering have revealed the studio's research-and-development arm will leverage Adobe's Firefly Foundry's tech to design future parks, hotels, cruises, and attractions.

Adobe's Firefly Foundry AI offering is not just another one-size-fits-all service but one custom-tuned to handle Disney's unique needs. The customized AI is trained on Disney's own IP offerings, allowing for what the companies call a 'responsibly built' solution to handle the following workflows to start:

- A sketch-to-image model that transforms rough hand-drawn concepts into fully rendered 2D concept art.

- A custom image model trained on Disney's own IP that generates franchise-accurate creative assets across Mickey & Friends, Frozen, Moana, Lilo & Stitch, and Cars.

- A 3D modeling capability that takes 2D renderings and transforms them into detailed prototypes for planning builds and coordinating with engineering teams.

How does AI help build the next generation of Disney rides?

Kyle Laughlin, SVP, Walt Disney Imagineering Research & Development, stated that the company is 'long-time users' of Adobe's Creative Cloud tools, with a relationship spanning decades, and were also among the early adopters of its Firefly offerings.

“We were looking to find a collaborator that could help us do that responsibly and do it in a way that ultimately respected the fact that we are a creator-driven company," Laughlin noted. "We are a talent-driven company, and respecting that the creative process has a human as a part of what we do.”

Disney says it does not plan to replace the human element in the process but will use solutions provided by Firefly to compress workloads and speed things up considerably, even as "outputs remain consistent with the company’s storytelling heritage and visual language."

"As the teams at Imagineering build new experiences for fans around the world, our tools and workflows will provide a creative foundation to explore bolder ideas and make the best ones a reality," noted Hannah Elsakr, vice president, GenAI New Business Ventures at Adobe.

Collaborations with companies such as Disney are the need of the hour for Adobe, which has investors often questioning the viability of its business model in an AI-enabled future where everyone has access to generative AI tools, even as it pitches itself as a frontrunner in a fast-evolving industry.

The move is being seen as a big win for Adobe, with Walt Disney Imagineering Research & Development arguably being one of the biggest clients it could court for its Firefly Foundry service.

The move comes at a time when Disney is a plaintiff in multiple IP-centric lawsuits against AI firms, including image- and video-generation firm Midjourney, that have allegedly plagiarized and therefore abused its copyrights.

• IBM pushes transistor density below the long-feared one-nanometer barrier
• NanoStack abandons flat chip layouts in favour of vertical transistor stacking
• The prototype delivered 50% more performance during IBM laboratory testing phases

IBM has unveiled what it describes as the world's first sub-1 nm chip technology, carrying nearly 100 billion transistors on a fingernail-sized surface.

The breakthrough revolves around a new 3D NanoStack architecture that moves transistor scaling into the 0.7 nm or 7 angstrom era.

For context, today's most advanced commercial chips typically sit around the 2nm mark, making this a substantial leap in density.

Building upwards to keep Moore's Law alive

The semiconductor industry has spent decades squeezing more transistors onto increasingly smaller pieces of silicon to improve computing performance.

That process has become progressively harder as transistor dimensions approach the scale of only a few atoms across modern processors.

IBM's approach avoids further horizontal compression by stacking transistor layers vertically through a three-dimensional nanosheet architecture instead.

The design packs nearly twice the transistor density of IBM's 2 nm chip technology introduced back in 2021.

According to the company, the architecture also delivers approximately 40% greater SRAM scaling to support increasingly demanding AI workloads.

This vertical method allows engineers to separate n-type and p-type transistors into distinct layers, which, according to IBM, permits independent optimization of materials for each.

, compared it with building a big block of flats rather than houses in a city.

"IBM's NanoStack is like proposing a 100-storey skyscraper," said Professor Alan Woodward, a computer scientist at Surrey University.

Using this analogy, IBM’s closest competitors, like Intel and Samsung, are somewhere around a 30 to 50-storey building, a far cry from IBM.

In testing, the company reported a 50% performance improvement and 70% greater energy efficiency compared with its existing 2nm chips, alongside a 40% gain in on-chip memory scaling.

Despite the quoted performance improvements, the technology remains years from commercial use, with IBM estimating production could begin within five years at the earliest.

"With our new NanoStack architecture, we're not just making smaller transistors, we're reinventing how chips are built to deliver dramatically more power and energy efficiency," said Jay Gambetta, Director of IBM Research and IBM Fellow.

The trade-offs behind the density gains

Vertical stacking introduces complications mostly around heat dissipation, since transistors generate heat that becomes harder to manage when layered closely together.

This same tight spacing also raises the stakes for wafer alignment, since layers must be bonded with extreme precision to avoid malfunction.

Researchers acknowledge that when gaps between layers grow too thin, transistors can fail to switch off correctly, undermining the very density gains NanoStack is meant to deliver.

These engineering trade-offs are symptoms of a deeper problem facing the entire chip industry.

For decades, manufacturers have relied on Moore's Law, the pattern of transistor counts doubling roughly every two years.

But that pace has grown harder to sustain as designs approach the physical limits of individual atoms.

Whether NanoStack genuinely extends that trajectory by another decade, as IBM projects, depends on whether these unresolved manufacturing challenges can be solved at scale.

It is partly for this reason that IBM has drawn in partners including ASML, Lam Research, and Tokyo Electron, signalling an industry-wide effort behind this push toward angstrom-level scaling.

Even so, similar bold claims accompanied IBM's 2nm chip unveiling in 2021, but turning lab success into mass production historically takes longer than initial announcements.

Via IBM

The rise of robotics has long threatened human employment, with machines in manufacturing settings especially seemingly performing the work that people used to do. Take this to its logical conclusion, and it's not outside the realm of possibility to have a massive unemployment problem on your hands. So how do we deal with it? Microsoft co-founder Bill Gates has a contentious idea.

Death, taxes and robots

There's an active (and ongoing) debate about how the rise of apparent job-stealing technology could be handled by policymakers.

This is a challenge to which Gates offered a solution nearly a decade ago, suggesting in an interview with QZ that a robot that performs a human's job should be taxed like its fleshy counterpart.

This, Gates argued at the time, could slow down the rapid speed of automation and be used to instigate exploration into other kinds of employment.

This is, of course, not a policy that everyone (especially much of the business world) agrees with. The International Federation of Robotics (IFR), for instance, decried the move and instead suggested further tax on profits to offset the impact of automation.

Tax-paying robots

Despite that, there are supporters of similar measures, including Anthropic's CEO Dario Amodei, who called for an AI tax proposal to build a multi-million-dollar worker fund. Andrew Yang, too, has called for a tax on automation and AI instead of a tax on humans.

Since 2017, the conversation has certainly moved on – with much of the conversation around automation based on software and digital tools like large language models (LLMs) rather than physical embodied AI robotics systems. That said, the robotics industry is growing fast. While AI is a key focus among policymakers, the conversation may well shift back to machines that we can see and feel.

• Meta recycled retired DDR4 memory instead of purchasing expensive new DRAM
• CXL technology turned discarded server memory into useful computing capacity
• Meta reported 25% fewer servers for machine learning inference workloads

Memory shortages, rising DRAM prices, and extended delivery schedules have pushed hyperscalers toward alternatives that seemed impractical only recently, but Meta has developed a way to reuse old DDR4 memory pulled from decommissioned servers rather than discarding it.

The approach allows companies to expand server memory capacity without purchasing new DRAM, a cost researchers describe as the so-called RAM tax.

This expansion is made possible through Compute Express Link technology (CXL), which connects the older DDR4 modules alongside newer DDR5 memory pools on the same machine.

Reusing old memory instead of buying new memory

Meta describes the approach as delivering near-zero-cost memory expansion while also reducing electronic waste and infrastructure emissions considerably.

The strategy arrives at a time when memory supply constraints continue affecting server deployment schedules across cloud computing environments worldwide.

According to Meta researchers, existing CXL implementations struggled because expanded memory delivered nearly ten times lower bandwidth than local memory.

The company also reported approximately 60% higher latency levels compared with directly attached memory residing alongside processor sockets inside servers.

Another limitation involved commercial CXL products bundling controllers with DRAM modules, preventing practical reuse of existing DDR4 inventories at scale.

Meta responded by developing an internal ASIC known as Vistara, specifically engineered around low latency, power efficiency and recycled memory usage.

The accompanying software stack automatically determines suitable memory ratios for individual workloads while disabling expansion where delays become unacceptable operational compromises.

"We address these challenges via hardware–software co-design. On the hardware side, we design an in-house CXL ASIC, Vistara, optimized for DRAM reuse, power efficiency, and low latency,” Meta said.

“On the software side, we build an optimized solution based on TPP (Transparent Page Placement), determine the appropriate local-to-expanded memory ratio for each workload, and automate per-workload configuration, including disabling expanded memory for workloads that cannot tolerate the increased latency."

Meta claims the architecture demonstrates sufficient practical value to justify deployment in production environments that handle diverse computational requirements daily.

Why Meta believes the economics finally work

Meta reported that disaggregated machine learning inference workloads achieved server count reductions reaching as high as 25% through implementation.

Distributed cache systems reportedly recorded average latency reductions of about 29%, despite relying partly upon slower recycled memory resources underneath.

The findings suggest that additional capacity sometimes outweighs raw memory speed when applications struggle more with shortages than response times.

Interestingly, the same interconnect technology attracting Meta's attention is also drawing interest from semiconductor firms developing large accelerator fabrics globally.

The broader ecosystem includes work from companies pursuing alternatives to proprietary interconnect technologies such as Nvidia's widely adopted NVLink systems.

Among these is Ultra Accelerator Link, or UAL, a separate initiative backed by AMD, AWS, Google, Microsoft and Meta to connect accelerators across different hardware vendors.

Within Meta's own testing, disaggregated machine learning inference systems and distributed caching infrastructure were the two workloads examined directly by researchers.

Both recorded measurable improvements from the recycled memory approach, with inference systems requiring fewer servers and caches experiencing lower average latency.

Whether recycled DDR4 through CXL becomes standard practice will likely depend on performance trade-offs remaining acceptable outside hyperscale environments.

Via Blocksandfiles

With the FIFA World Cup 2026 now well underway, the excitement is building as the tournament reaches its most crucial stages.

But this edition of the world's biggest sporting event is also notable for being the most technologically-friendly so far, with a raft of new AI-powered tools, services and broadcast innovations.

As the official technology partner of the World Cup and FIFA itself, Lenovo has been at the heart of much of this innovation, and the company invited me to Miami to see it all in action.

Behind the curtain

Now - a word of caution - sadly this article isn't going to delve too much into specifics, or include photos of all the amazing work the FIFA team is doing to keep this World Cup safe and running smoothly.

That's because the organization is understandably pretty focused on making sure the tournament (the biggest and most spread-out yet, with 48 teams competing at 16 stadiums across 3 countries) continues to do so, and for that reason we weren't allowed to take any video or photos during our visit - for obvious reasons.

However the Technology Command Center (TCC), situated in the sleepy Miami suburb of Coral Gables, is a dream come true for tech nerds like me - banks of screens showing detail on everything from network infrastructure strength to any possible cyberattacks to the live match feeds themselves.

We're visiting on a day when six matches are scheduled to take place - a huge strain on the FIFA networks, Nacho Fresco, Director of Technology at FIFA explains.

The TCC is able to monitor in real-time across the entire tournament's tech stack, which includes 30 proprietary apps, spotting any issues before they happen, and ensuring fans around the world continue to get access to the action.

Inside the TCC is a huge team of analysts, managers and other tech experts making sure all the services are functioning properly, with direct links to the wider FIFA organization along with remote teams at all 16 stadiums.

This includes Lenovo engineers, who are present at stadiums but also the central International Broadcast Center (IBC) in Dallas overseeing the global distribution of the action to the estimated six billion people watching across the globe.

We saw a screen showing the various connections between the 16 venues and the IBC, all including fallback options in case any issues should occur, making sure the broadcast continues.

In order to help boost this, Lenovo has deployed servers at the IBC to provide the power needed for the most expansive broadcast operation in FIFA World Cup history, with a total of over 17,000 Lenovo and Motorola devices  also deployed across venues and Team Base Camp training sites.

Under threat

Probably the most intriguing screen was the one showing cybersecurity threats targeting FIFA's systems and services at the World Cup.

In real-time, the team can see exactly how many attempts are being made to hack or disrupt the platforms, including the location of the criminals, and even what kind of bots are being used.

It's fascinating to see the sheer scale of attacks - Fresco says they currently face around a staggering 300-500 million cyberattacks a day, a number that seems crazy, but shows just how attractive a target FIFA is.

Fresco notes that attacks have been slowly ramping up over the last few months, and expects the total number to top that seen in the 2022 World Cup in Qatar, which saw around 11 billion attempted attacks.

I ask Fresco what would constitute success for FIFA when it comes to the technology stack at the tournament - is it making sure everything runs smoothly, or just keeping the lights on?

Unsurprisingly, he replies that delivering a great viewing experience for the fans is the priority, as the organization looks to continue with its Lenovo partnership and take it further in the future.

"It's a game changer," he says, "it's what we need for this tournament."

So - no pressure then. But as Fresco pointed out, the mere fact that all the talk around the tournament has been about the football itself, not any issues with the technology, is the best outcome FIFA and Lenovo can hope for.

• Alogic Folio Duo offers dual-QHD portable monitor setup that works with both MacBooks and Windows-based laptops
• It is one of the first portable displays to offer full gesture control and 10-point multitouch on both Mac and Windows
• The Folio Duo also offers users the ability to charge their laptops on the go with 45W of USB-C passthrough

Portable monitors are hardly a new concept, with most products offering a sizeable upgrade in screen size, resolution, or both compared to the desktops, laptops, or MacBooks they connect to.

However some, like the Alogic Folio Duo, kick things up a notch while doubling the screen real estate, and its foldable design, coupled with an integrated cover stand, makes for a much easier-to-deploy and carry solution than many of its peers.

With a dual-QHD configuration coupled with a 400-nit display, the Folio Duo outdoes many of the devices it will inevitably be attached to, and considerably so, while weighing approximately 1.2kg.

A premium portable dual-monitor offering

The Alogic Folio Duo first appeared at InfoComm 2026, pitched as a premium dual-monitor offering for professionals who simply need more screen space without adding as much of a volumetric footprint as some of its peers.

Alogic's offering features dual IPS panels rated for 100% sRGB color accuracy, with a native resolution of 2560x2880 (or 2560x1440 per display). Prospective gamers, however, might be disappointed with the relatively tame 60Hz refresh rate, but one could argue that most would not be looking for a portable dual-monitor setup to begin with.

Like most of its peers, the Folio Duo offers USB-C connectivity but kicks things up a notch by adding USB-PD input for an optional power source and a USB-C port for the device it connects to, allowing it to charge a MacBook or Windows laptop up to 45W via USB-C passthrough.

Alogic offers a generous 2-year warranty for the unit and lists device compatibility that includes Apple, Chromebook, and Microsoft devices, with the package including a monitor cover in the box, while skipping the pressure-sensitive Active Style the company sells, even as it integrates wireless charging into the package.

The fact that it is one of the first portable dual-display offerings to support full gesture control and 10-point multitouch on both Mac and Windows might underscore the ambitious price the portable monitor currently commands: a whopping $1,299, which makes it rival the recently repriced 13.6-inch Apple MacBook Air (M5), a full-fledged computer.

For those who can afford it, however, the Folio Duo seems to be a premium portable dual-monitor offering with significantly higher resolution (and PPI), color, and build quality than the UPerfect Delta Max Touch we recently reviewed, albeit at more than twice the asking price.

• AMD has 'relaunched' three processors that use legacy cores to address demand at the lowest end of its mobile and desktop CPU offerings
• The CPUs range from 2 to 8 cores and leverage its older Zen+ and Zen 2 core technology as the chip designer grapples with memory and storage shortages to serve an increasingly cash-strapped budget segment
• All three processors support Microsoft Windows 11 and leverage cheaper DDR4 memory as manufacturers aims to 'hold the fort' in the face of rising AI demand creating challenges for entry-level PC buyers

AMD's latest Ryzen CPU offering might not be one that most of its performance-centric consumers would appreciate in 2026.

The company is looking to reintroduce three budget-centric mobile CPUs that were last seen over half a decade ago, using legacy Zen+ and Zen 2 cores.

The move comes at a time when assemblers and consumers alike are grappling with increasing chip, memory and storage costs, even as lower-end options continue to grow increasingly limited for price-conscious users.

A Dual-core CPU in 2026?

AMD is reintroducing the Ryzen 3 3100U, Ryzen 5 3501U, and Ryzen 7 4700LE to consumer markets, according to reports. The CPUs offer dual, quad, and octa-core options.

While the Zen 2-based desktop-class Ryzen 7 4700LE has a listed launch date of 25th March 2026, the other two SKUs were launched more recently, in June 2026, according to AMD's own listings.

The Ryzen 3100U and the 3501U both use AMD's legacy Zen+ architecture on a 12nm process node, making them less power-efficient than the 7nm-based Ryzen 7 4700LE.

In an emailed statement to PC World, AMD stated, "The Ryzen 3100U and Ryzen 3501U are additional SKUs based on AMD’s existing Picasso architecture that were developed to support specific OEM requirements in the value segment. These processors are intended to address targeted customer demand for lower-cost solutions and will be available in limited volumes through select OEMs".

As OEM components are made directly available to laptop assemblers, AMD does not disclose prices for these models, even as rival chipmaker Intel seems to be taking a similar route.

What might rankle some users in particular is not the fact that the 3100U and the 3501U are rebadged CPUs that will still lag behind pretty much even the most basic laptop CPUs available in the last few years, but the fact that AMD might be thinking it is acceptable to offer a 2-core, 4-thread CPU in 2026 as a baseline for new, entry-level laptops.

The Ryzen 3 3100U is getting considerable ire, as Redditors are already drawing unfavorable comparisons to current-generation Athlon CPUs. While it offers the same iGPU (Vega 8), TDP (15W), and memory configuration (DDR4-2400), along with the same socket (FP5) as its quad-core sibling, it is noticeably weaker than most, if not all, recently launched mobile CPUs.

Unsurprisingly, it does make the cut for Windows 11 by offering support for fTPM, allowing for secure boot, and just pulling ahead of Microsoft's stated requirements for its OS (1 gigahertz (GHz) or faster with 2 or more cores), even as most users will struggle to do anything but the bare minimum on a laptop sporting a CPU essentially rebadged from 2019.

AMD's move, however, is predicated on the fact that rising RAM, SSD, and cutting-edge silicon prices all create a void that must be addressed for the foreseeable future, even as its target audience seems to be getting an increasingly short end of the stick as part of the solution.

• Russian Intelligence are targeting Signal accounts of officials based in Ukraine
• They pose as Signal support services and ask users to submit their Backup Recovery Keys
• Using these keys, the hackers can hijack the users account and any other accounts created using the same mobile phone number

The FBI has warned Russian Intelligence Services are posing as commercial messaging application support services in order to steal Backup Recovery Keys belonging to targets of high value in the military and government of the US, Europe, and Ukraine.

In a joint warning alongside the CISA and the Security Service of Ukraine (SSU), the FBI outlined the new phishing campaign which seeks to access messaging accounts in order to perform intelligence gathering of secret information.

Specifically, the FBI provided sample phishing lures targeting users of the Signal messaging app. If the hackers successfully lure a victim into sharing their Backup Recovery Key, they can access the account's message history, private and group messages, and fully take over the victim's account.

Russian Intelligence pose as Signal support services

In the FBI warning, the phishing techniques are further detailed. The Russian Federal Security Service (FSB) are targeting government officials, military personnel, political figures, journalists, and key officials from the US and Europe located in Ukraine.

The attackers send emails that appear to be automated messages from Signal, asking users to turn on their message backup using their Backup Recovery Key. Victims are provided with false instructions that instead send the Backup Recovery Key to the attacker, who can then use the key to take over the victim’s account.

In order to establish urgency and trust that the message is legitimate, the attackers posed the phishing message as a protection against recent hacking attempts from “Iran and post-Soviet countries.” In another sample message, the attacker's message says that the victim’s account data “is at risk of permanent loss due to a sync issue.”

If a victim shares their unique Backup Recovery Key, it allows the attacker to hijack their current Signal account alongside any subsequent accounts made with the same phone number.

For users who may fear their Backup Recovery Key has been compromised, users are instructed to use Signal settings to create a new Backup Recovery Key. This new key will invalidate all previous Backup Recovery Keys and prevent account takeover if the previous key was leaked.

In order to avoid falling victim to phishing messages, there are several ways to stay safe:

• Support services will generally only communicate with users via an official company email address. Always carefully check communications from the legitimate email address.
• Customer support will never request that you supply your Backup Recovery Key via the application
• You will never be asked to verify or restore your account via an automated customer support message

In order to further protect your Signal account, or other accounts, against phishing, users should consider the following:

• Use a passkey wherever possible. This will use your device’s built in biometric verification methods to authenticate your login.
• Use phishing resistant multi-factor authentication where possible
• Always double check messages and emails are legitimate, and are using an official company email
• Never supply your Backup Recovery Keys unless you are actively attempting to regain access to your account via a legitimate service

• Tens of millions of credentials may have been leaked following an attack on one of Japan's largest ISPs
• The attack leveraged a vulnerability in a third-party software used by KDDI
• Five other ISPs were also affected in the attack

A data breach that has potentially exposed the email and password combinations for over 14 million customers across six internet service providers (ISPs) has been disclosed by Japanese telecoms provider KDDI Corporation.

According to the company, hackers exploited a vulnerability in a third-party software to access the database of credentials. KDDI said that it immediately blocked the hackers' access after discovering the intrusion on June 17, 2026.

“Although technical defensive measures have already been implemented for the system, there remains a possibility that customers' email addresses and passwords were obtained by unauthorized third parties as a result of the incident,” the company said in a statement.

Millions of credentials exposed

Unfortunately, the breach was not confined to just KDDI. The email services of five other ISPs were also affected by the breach:

• STNet, Inc.
• JCOM Co., Ltd.
• Chubu Telecommunications C., Inc.
• NIFTY Corporation
• BIGLOBE Inc.

KDDI is yet to finish a formal investigation into the attack, but said that the hacker may have gained access to the emails addresses and passwords for 14.22 million current and former customers. The company also said that some of the passwords were stored in an encrypted format, and so will be inaccessible for the hackers, but the company did not say how many were stored in this manner.

Since discovering the breach, KDDI has also been working alongside the affected ISPs to secure systems and put in place mitigation measures to counter the abuse of exposed account credentials.

In order to stay protected, customers have been advised to change their account passwords and implement two-factor authentication.

Breaches such as these are particularly dangerous because they expose email and password combinations. As most people will have either one or two email addresses across their accounts, it increases the likelihood that hackers can attempt to use the exposed email and password combinations to try and access other accounts created with the same email.

This is especially true if the same password (or a variant thereof) is used across multiple accounts. Hackers can use brute force techniques to try hundreds of password combinations in a very short amount of time in order to crack weak or reused passwords.

When creating or updating a password for any account, no matter how infrequently it is used, always create a strong unique password. Password managers can create and suggest strong passwords, securely store them, and automatically fill login forms to take the hassle out of remembering passwords.

Alternatively, some services offer the ability to login using a passkey, which utilizes the built-in biometric authentication mechanisms of your device such as a facial scan or fingerprint. These login methods not only remove the need to type in passwords, but also reduce the possibility of hackers accessing your account through phishing attacks.

Via BleepingComputer

• Real-world impact is more important than benchmark scores, Meta AI VP says
• Song emphasizes security, trust and real-world benefits
• Meta's latest model prioritizes people, she says

Meta's newly appointed AI research chief, Dawn Song, is betting on agentic AI for the future of artificial intelligence, emphasizing they should augment humans rather than replace people altogether.

Song sees agents performing "economically valuable" tasks like repetitive and time-consuming work, ultimately freeing up humans to do more creative work.

While companies struggle to quantify AI's impacts and deliver a meaningful ROI, Song believes the focus should be real-world impact rather than benchmark scores.

AI agents should augment – not replace – humans

"The goal is not to replace humans," Song told the South China Morning Post. She confirmed she would be joining Meta Superintelligence Labs in a LinkedIn post, together with other team members from Virtue AI.

"[AI] must be secure, trustworthy, and beneficial," she added.

Song is also a professor in computer science at the University of California, Berkeley – a university that recently introduced Agents' Last Exam (ALE), a new type of benchmark that assesses whether AI agents can complete more than 1,500 economically valuable tasks across 55 different industries.

Meta itself launched its first new model, Muse, in April, which it says is designed to "prioritize people."

As MSL's Vice President of AI Research, Song will focus on AI safety, security and research, and will likely continue to emphasize the role of humans in an AI-first era.

Model capabilities are no longer a drawback for AI developers, with human, socioeconomic and geopolitical impacts now emerging as a major focus. Anthropic was recently forced by the White House to pull its latest frontier models over jailbreaking concerns.

Enterprise AI has entered a new operational phase, moving rapidly from experimentation into production systems integrated into customer experiences, workflows, and software delivery pipelines.

However, as organizations operationalize AI, they are also introducing new complexity around infrastructure, governance, debugging, capacity planning, and cost control.

This complexity introduces new operational risks.

AI systems continuously evolve as prompts change, models are updated, agents become more autonomous, and infrastructure dependencies shift over time.

Without end-to-end visibility across the full AI stack, issues related to reliability, latency, output quality, or cost efficiency can gradually slip into production unnoticed: resulting in what many teams refer to as “invisible drift.”

As AI adoption scales, observability is becoming essential for helping engineering teams maintain operational control, reliability, and resilience in rapidly changing environments.

Multi-provider AI brings a new wave of platform engineering challenges

Organizations are increasingly adopting multi-model AI strategies rather than relying on a single provider. Recent research shows that more than 70 per cent of organizations now use three or more models in their production environments. This reflects a broader shift toward diversified model libraries, with teams are selecting models based on specific workload requirements such as latency, reasoning ability, operational risk, and cost efficiency.

This shift is creating a new generation of platform engineering challenges. AI environments now span evolving ecosystems of models, agents, orchestration frameworks, APIs, vector databases and infrastructure layers. As coding agents accelerate development, organizations are generating more code, dependencies, and operational overhead than teams can realistically manage manually.

At the same time, enterprises are accumulating significant LLM technical debt as they rapidly integrate new tools and frameworks. Tool sprawl, fragmented visibility, and constantly evolving AI architectures are making systems harder to govern, troubleshoot, optimize and secure. This makes AI observability essential, providing centralized visibility into model behavior, prompts, latency, hallucinations, token usage, infrastructure performance, and operational bottlenecks across complex multi-model environments.

Scaling AI safely, reliably and at speed requires control

As organizations race to scale their AI initiatives, operational failures are becoming more visible. Recent analysis shows that two per cent of all LLM calls returned errors, with rate limit issues accounting for almost a third of these (equating to approximately 8.4 million rate limit errors in total). This highlights the operational strain on systems as AI adoption accelerates.

At the same time, pressure to remain competitive is pushing organizations to move projects into production before operational controls have fully matured. Scaling too quickly introduces significant reliability, resilience, and governance risks. Real-time observability across the AI stack gives engineering teams the visibility needed to move quickly while maintaining high performance standards.

AI agents are adding yet another layer of complexity. Adoption of agent frameworks has doubled in the past year, leading to increased “agent sprawl”. These agents autonomously interact with multiple tools, systems, APIs, and datasets, making it harder for organizations to monitor behavior, diagnose faults, manage security risks, and maintain governance controls without deeper telemetry.

To manage this complexity, organizations need enterprise-grade observability that delivers end-to-end visibility across the AI stack (from development through to production). This includes visibility into prompts, model interactions, inference pipelines, infrastructure performance, latency, failures, and downstream dependencies. With comprehensive telemetry in place, teams can accelerate AI innovation while improving reliability, security, and operational controls at scale.

Four ways observability helps organizations scale AI more reliably

Organizations moving AI into production are increasingly treating observability as a foundational operational discipline, rather than simply a monitoring capability. Four practices are becoming particularly important as enterprises scale multi-model AI environments:

1.Managing multi-model environments more effectively

Teams are implementing gateways, routing layers, and evaluation frameworks that enhance their ability to select, assess, and manage multi-model environments effectively. These systems enable organizations to compare model behaviors, evaluate outputs, optimize workload placement, and enforce governance policies across various providers. AI observability provides the real-time data needed to support these decisions.

2.Reducing operational overhead and tech debt

Centralized visibility across prompts, models, inference pipelines, and infrastructure helps teams manage increasingly distributed environments. Observability reduces operational overhead and limits the accumulation of LLM technical debt as tools and frameworks evolve.

3.Improving agent reliability and preventing infrastructure failures

AI observability improves agent reliability and helps organizations eliminate failures caused by capacity constraints and infrastructure bottlenecks. Real-time monitoring of GPU utilization, throughput, latency, request failures, and workload behavior enables engineering teams to identify emerging scaling limitations before they impact production systems or user experiences.

4.Diagnosing faults and understanding agent behavior

Detailed tracing across prompts, workflows, APIs, orchestration layers, and infrastructure dependencies provides the operational context needed to investigate anomalies and identify root causes. This is critical for understanding how AI agents behave in real-world production environments.

Moving to a state of production-ready AI

Enterprise AI is now entering its operational era. As organizations move from experimentation to production, observability becomes the backbone for managing the growing complexity of multi-model architectures, autonomous agents, and distributed AI systems.

Without deep visibility into how these systems operate in production, organizations risk increasing operational failures, accumulating technical debt, and allowing invisible drift to undermine performance, reliability and governance over time.

AI observability provides the control needed to scale AI safely and effectively. Visibility across models, prompts, infrastructure, agents, and workflows helps teams build more governable, resilient and cost-effective AI systems.

Success in the next phase of AI adoption will depend on transforming experimental AI systems into disciplined production platforms that can be continuously evaluated, improved and trusted at scale.

We've featured the best data migration 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

• Samsung wants NAND layer counts to reach four digits this decade
• Future M.2 SSDs could expand from 8TB to 32TB capacities
• Cell Multi-Bonding may replace traditional approaches to NAND density growth

Samsung has mapped out a NAND strategy stretching toward 1000-layer memory designs as demand for denser storage continues accelerating across industries.

The roadmap, disclosed during the IEEE/JSAP VLSI Symposium 2026, extends current vertical scaling plans far beyond existing commercial products.

At the centre of those ambitions sits a future generation of storage capable of pushing familiar SSD capacities into unfamiliar territory.

The route from today's drives to tomorrow's 32TB M.2 SSDs

Samsung expects its NAND technology to reach roughly 420 layers by 2029 before advancing beyond 560 layers during 2030.

Beyond that point, the company intends to explore architectures carrying between 900 and 1000 layers within future generations of flash memory.

Rather than constructing one towering NAND structure, Samsung plans to combine multiple stacks through its Cell Multi-Bonding technology approach.

The method joins separate NAND structures together inside one package to achieve densities that resemble a single 1000-layer device.

Samsung specifically discussed pairing two approximately 450-layer structures to approach the effective density associated with future four-digit layer counts.

According to company projections, this arrangement could increase storage density by as much as four times current generation solutions.

One example outlined during the presentation involved an 8TB QLC M.2 drive eventually scaling upward to reach 32TB capacities.

That scenario would allow substantially larger SSDs without increasing physical dimensions, preserving the compact M.2 format many users remember well.

The same scaling approach could eventually contribute toward enterprise drives exceeding 100TB while bringing Petabyte SSD discussions closer to reality.

The engineering problems standing between Samsung and a petabyte SSD

Samsung acknowledged that increasing layer counts introduces manufacturing complications that become progressively harder as structures grow vertically taller.

One major concern involves wafer warpage, where taller structures can deform during production and reduce manufacturing consistency or yields.

Another challenge comes from maintaining alignment accuracy across hundreds of stacked layers, requiring extremely precise overlay control throughout fabrication processes.

Even relatively minor deviations between layers can affect long-term reliability, performance, and manufacturing efficiency across finished storage products.

To limit deformation effects, Samsung plans to introduce an Upper Chuck Design intended to stabilise increasingly complex wafer structures.

The company also discussed Overlay Correction technologies intended to improve alignment precision as future NAND structures continue growing upward.

These developments arrive as conventional process shrinking becomes increasingly difficult, forcing memory manufacturers toward more elaborate vertical architectures.

Samsung has separately discussed a record-breaking 400-layer NAND generation that could help push AI hyperscaler SSD capacities beyond the 200TB barrier.

The company has also linked 430-layer NAND technology with a future where 100TB SSDs become increasingly mainstream across enterprise deployments.

Samsung will work with hafnia ferroelectrics in order to extend the practical layer counts beyond 1000 layers.

However, whether these designs eventually deliver mainstream 32TB M.2 drives or petabyte SSDs remains dependent upon manufacturing realities rather than roadmaps.

Samsung's latest roadmap nevertheless suggests the race toward NAND supremacy depends upon stacking ingenuity rather than shrinking dimensions.

Via The Guru of 3D

• Oracle admitted it needed a new approach back in February
• Now, the company has a new community-focused approach to MySQL
• Community roles and steering committee will ensure maximum quality

Oracle's VP of External Standards & Community Engagement, Heather VanCura, has revealed in a blog post the company's "next phase" of MySQL community engagement as it seeks to the open-source database project more transparent and collaborative.

The update comes after community criticism over Oracle's leadership of the project, and the company has finally responded with new governance frameworks.

In February 2026, Community Manager Frederic Descamps admitted the project was letting down community members. Descamps promised a "decisive" and "reinvigorated" approach.

Oracle announces "next phase" of MySQL

VanCura clearly acknowledged the role that community members have played in MySQL over the past three decades, referencing the need for visibility and deeper participation.

Under the new plans, Contributors can participate through code submissions, bug testing, code reviews, technical discussions and more. Experiences Contributors can become Committers, taking on responsibilities like code quality maintenance, and they're set to be guided by Mentors and Project Leads.

"Trust is built through transparent processes, clear decision-making, and meaningful opportunities for participation," VanCura added.

Oracle also launched a new Vulnerability Group and Technical Steering Committee. AWS, Google Cloud and Oracle will have representatives on the committee, but the company also noted the need for "additional perspectives from users of MySQL."

Percona co-founder Peter Zaitsev praised Oracle for its clear willingness to want to make changes (via The Register): "This is a step and in the right direction... Oracle has shown a desire to show more openness to the community in terms of sharing and including the wider community in the decision-making process."

Looking ahead, VanCura promises to share metrics with the community, engage in more discussions and uphold a high quality of contributions.

At SAP Sapphire in Orlando, Christian Klein put it plainly: “For the mission-critical processes of our customers, almost right just isn’t good enough.”

It was the line that crystallized the Autonomous Enterprise vision, and it is also the line that should reframe how every operations leader thinks about AI for the next eighteen months.

Sapphire made one thing unambiguous. AI is becoming visible at the top of the stack.

Joule (or your chosen equivalent) is being positioned as the new front door to enterprise software, with more than two hundred agents and over fifty assistants spanning finance, supply chain, procurement, HCM, and customer experience. Users will increasingly describe an outcome and let agents orchestrate the work across SAP and non-SAP systems.

That is the visible layer. There is also an invisible one, and it is the one that determines whether any of this actually delivers.

Agents only behave as well as the operational substrate they run on. They need systems that are healthy, observable, and consistent enough to act on safely. They need clean process telemetry, automated remediation when things break, and governance that extends across the hybrid landscape most large enterprises actually run. Without that foundation, agentic AI does not reduce operational risk. It multiplies it.

This is what future-proofing now means. Less about adopting the latest model, more about building the operational layer underneath it so that agents become a source of measurable outcomes rather than a source of new incidents.

The opportunity is significant for organizations that get this right. Two groups are forming. Those who have built the operational readiness to let agents execute, and those who will spend the next two years discovering they have not.

Pragmatism in ERP transformation

Enterprises are navigating significant transitions in their core systems, and the 2027 SAP ECC end-of-mainstream-maintenance deadline is the most visible forcing function. But the SAPinsider 2026 research surfaces a more interesting signal underneath it. AI readiness is now cited by 43% of organizations as the primary driver of their transformation investment, ranking above the deadline itself. The deadline creates urgency. AI readiness creates direction.

For many large enterprises, the preferred approach is not wholesale reinvention but incremental change. Brownfield migration has become a common starting point. It allows organizations to move existing systems to modern platforms while preserving established processes and minimizing disruption. In complex landscapes with extensive integrations and dependencies, that level of continuity is non-negotiable.

A brownfield approach also provides a structured path forward. It enables organizations to stabilize their core systems before introducing further innovation, including agentic AI. The transition to cloud ERP software plays a central role here. Managed, scalable environments establish the platform that supports both current operations and future capabilities, with continuous updates and easier integration of new services.

This foundation matters particularly for AI. As intelligent features become embedded within enterprise applications, cloud platforms provide the IT infrastructure needed to support them at scale. From advanced analytics to autonomous execution, AI capabilities are increasingly delivered as part of the platform rather than as separate tools.

During these transitions, most organizations operate in hybrid environments that combine on-premises and cloud systems. This state can persist for years, introducing complexity in governance, monitoring, and integration. Managing hybrid operations effectively requires clear definitions of roles and responsibilities, and an operational substrate that is observable, automatable, and consistent across the entire landscape.

As legacy solutions reach the end of life, organizations are reassessing how they support operations in this mixed environment, and the bar is rising.

AI as invisible infrastructure, AI as visible interaction

The Sapphire announcements make clear that AI is now operating at two layers, and both have to work.

At the interaction layer, AI is becoming the front door. Joule Work, the Autonomous Suite, and the broader agentic stack are designed to let users interact with enterprise systems through conversation and outcomes rather than screens and clicks. This is the visible AI, and it is what most of the industry will spend the next year talking about.

At the execution layer, AI is also becoming part of the underlying infrastructure. It will show up in observability, in automated remediation, in capacity and performance management, in the operational disciplines that have always determined whether mission-critical systems actually behave. This is the invisible AI, and it is what determines whether the visible layer delivers.

Lacking context is the number one reason enterprise AI projects fail to deliver value. Operational data, process telemetry, and the live state of the landscape are a critical part of that context. Agents that act on stale, incomplete, or unobservable systems will produce confident answers that quietly create new failure modes. Agents that act on a well-instrumented, well-automated estate will deliver the outcomes Sapphire promised.

This is why operational readiness is emerging as the real differentiator. Two groups are forming. Those who have built the foundation that lets agents execute reliably, and those who have never closed the gap between AI ambition and operational reality. The divide is not driven by access to technology. AI capabilities are increasingly available across major platforms. The divide is driven by whether the operational layer is ready to absorb them.

Positioning for long-term resilience

For enterprise and technology leaders, the convergence of cloud transformation and agentic AI presents a clearer opportunity than at any previous point in the SAP cycle. The path forward is not defined by rapid disruption but by deliberate, sustained evolution.

Future-proofing now means building the foundation that lets continuous improvement happen safely. It involves modernising core systems, embracing incremental change, and ensuring that emerging capabilities, especially agentic ones, can be integrated into operations without expanding the risk surface.

As AI becomes embedded across both the interaction layer and the execution layer, success will depend on how well organisations have prepared for both. The goal is intelligent operations that deliver tangible business outcomes, with AI serving as the enabler at every level of the stack. Resilience, adaptability, and operational discipline are the disciplines that will define long-term competitiveness in the autonomous enterprise era.

We list 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

As AI increasingly underpins competitive advantage, UK organizations face mounting pressure to build a workforce capable of using these technologies effectively and confidently.

Universities are central to this talent pipeline. However, many employers see a growing disconnect between what students learn academically and the skills required in today’s workplace.

UK job postings mentioning AI skills are now 127% above pre‑pandemic levels, highlighting how quickly employer expectations are evolving.

To close this gap, particularly amid the ongoing AI skills shortage, higher education institutions need to place greater emphasis on practical, applied training.

That means ensuring students are not only familiar with AI tools, but capable of using them confidently, critically, and responsibly in real-world settings.

Core competencies such as prompt engineering, AI-enabled data analysis, and translating AI insights into business outcomes are essential.

Without these skills, graduates risk entering the workforce underprepared.

Employer expectations are rising faster than graduate readiness

The standard for what constitutes an ‘AI-ready' employee is increasing. As AI shifts from experimental to embedded in everyday workflows, organizations expect new hires to use these tools effectively and with minimal onboarding.

At the same time, many employers lack the internal capacity to build these capabilities themselves. A significant proportion report limited hands-on AI use across their workforce, and relatively few have invested in structured training programs. As a result, businesses are increasingly looking to early-career talent to help close this gap.

However, many graduates still enter the workplace without meaningful exposure to AI in a professional context or an understanding of how it informs decision-making. Without stronger integration of practical AI experience into higher education, the gap between employer expectations and graduate readiness is likely to widen.

Equipping educators is critical to building AI-ready talent

Universities are on the front line when it comes to preparing students for the responsible use of AI, both during their studies and beyond. But this depends on educators themselves feeling confident in how these tools work in practice.

This includes understanding which tools are appropriate for different use cases, how to incorporate them into learning in a way that builds critical thinking rather than shortcuts, and how to demonstrate responsible, real-world usage. Yet many educators still lack confidence in this area, particularly when it comes to applying AI in daily teaching or administrative tasks.

The challenge is that when educators are not fully comfortable with AI, it becomes harder to equip students with the practical skills they need. To address this, institutions need to move beyond basic awareness and invest in meaningful professional development.

Educators should be supported to understand not just how to use AI, but where it adds value, where it falls short, and how to evaluate its outputs critically. With this foundation, they can then introduce low-risk, practical opportunities for students to engage with AI. For example, using it to break down complex concepts while strengthening independent reasoning.

Embedding AI throughout the student journey

Despite AI becoming increasingly prevalent in the workplace, many students still remain hesitant to use it during their studies. This is often due to uncertainty around what is permitted. Clear and practical guidance from universities is essential to remove this ambiguity, reduce anxiety, and support responsible adoption.

Once expectations are clearly defined, institutions can create learning environments that help students understand not only how to use AI, but when and why. Greater transparency fosters confidence and builds the kind of ethical judgement that employers increasingly value.

In practice, this means modelling responsible use of AI while encouraging reflection and critical thinking. Assessment methods may also need to evolve by shifting focus from the final output to the full process of learning, including research, drafting and iteration.

Regular feedback and check-ins also play an important role. Tools that provide insight into how work develops over time can help educators offer more targeted guidance and create space for open conversations around AI use. Over time, this approach helps students build transferable skills, such as evaluating their own work, applying sound judgement, and clearly articulating the reasoning behind their decisions.

Final thought

The gap between what universities teach and what employers expect, particularly around AI, continues to grow. But with clear frameworks, better-supported educators, and a more deliberate integration of AI across curricula, there is a clear opportunity to realign education with workplace needs.

By investing in AI capability at every level (from faculty development to student experience), universities can ensure graduates leave with more than theoretical knowledge. Instead, they enter the workforce equipped to use AI with confidence. In doing so, institutions play a vital role in strengthening the talent pipeline that organizations increasingly depend on.

We list the best recruitment 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

• Excel's AI updates give it reusable skills and real-time data integration
• Microsoft knows finance workers are an untapped area of the market
• User demand is responsible for which tools Microsoft adds next

Microsoft has announced a new update to Copilot in Excel aimed specifically at finance workers, positioning the spreadsheet software as an AI-powered tool for financial modelling, forecasting and reporting.

OpenAI and Anthropic are already proving the field could be a lucrative business model, targeting both finance and legal departments with their own domain-specific offerings, and Microsoft has become the latest.

The company proudly proclaimed that Excel has been the go-to tool for finance workers for decades, hoping that this AI-powered makeover will concrete its position for decade to come.

Excel is getting new finance-specific AI tools

Rather than pushing useless and unrelated AI on users, Microsoft stressed that its users "shape it as much as they use it, telling us where it falls short."

Skills are the focus of this latest update – pre-built, repeatable workflows like closing the boots and refreshing a monthly report so that users can generate accurate insights without having to set it up each and every time. Customers can build their own skills or pick from a library of common skills, and soon, partners will also be able to offer their own skills through Excel like LSEG, Ramp, Rogo, Samaya AI, Velixo and Vena.

And for companies looking to tap into external data, direct integrations with third parties like CB Insights, Daloopa, FactSet, Morningstar, PitchBoot and S&P Global pull in real-time information, eliminating the need for time-consuming manual data pulls.

Importantly, Microsoft continues to position its AI tools as an AI colleague, not a human replacer, comparing these particular Excel upgrades to a "trusted analyst" that makes "transparent and reviewable" changes.

But despite offering native AI integration, Claude and ChatGPT already have their own Excel add-ons.

The latest Copilot AI updates are now available for M365 Copilot users, with continued rollout of custom skills and partner-built skills over the course of the next few months.

For years, software subscriptions followed the same pattern.

Businesses signed up for new tools, teams tested them for a while, interest faded, budgets came under pressure, and subscriptions were cut.

Many companies ended up with overlapping platforms that few people fully used or even understood.

The result was fragmented systems, rising costs, and growing frustration.

SMEs continue to adopt AI.

Among UK small and midsize businesses that paid for AI tools in 2024, nearly eight in ten were still paying for them a year later.

That matters because businesses do not keep spending money on business software that does not prove its value.

What feels different this time is how naturally AI fits into the software businesses already use every day. For many people, it is simply becoming part of how work gets done.

Why SMEs adopt technology differently

Small businesses are usually the clearest test of whether technology genuinely works. SMEs operate with tighter margins, smaller teams, and far less room for wasted spending. If something does not help, it goes.

The businesses seeing the most value from AI are often using it in practical ways. They are reducing admin, speeding up repetitive tasks, organizing information faster, supporting customer service, and helping small teams get through more work in less time.

In the UK, businesses report using AI most heavily in admin, data processing, and customer service work. Those are areas where the work is structured, outcomes are clear, and oversight is straightforward.

AI is a productivity multiplier. It does not make businesses less busy, it removes friction, which means they progress faster than it previously thought possible.

That idea of removing friction may explain why this software cycle feels different from previous ones. Businesses are improving work that already exists instead of layering on more disconnected tools.

The most useful AI tools are often the least noticeable.

Relying on AI everyday

Many businesses are now using AI inside software they already rely on every day — not as a separate product to log into, but as a capability running quietly inside existing systems. For smaller businesses especially, that distinction matters. The value is not in adopting something new. It is in existing work becoming faster, more accurate, and less dependent on manual effort.

That shift changes the nature of the relationship between a business and its technology. When AI is a standalone tool, it remains optional. A business can trial it, decide it is not worth the cost, and remove it without consequence.

The calculation changes entirely when AI is integrated into the systems businesses already use to manage invoicing, customer records, financial reporting and daily communications. The tool is no longer separate from the work. It has become part of how the work happens.

That is why retention rates look the way they do. Businesses are not renewing AI subscriptions out of habit or inertia. They are renewing because the workflows they have built around these capabilities would have to be rebuilt without them.

AI allows a small team to operate with the maturity, governance, and delivery capability of a much larger organisation.

For SMBs that have spent years competing against larger businesses with bigger teams and deeper resources, that is not a small claim. It is the argument for AI in a single sentence.

Businesses still want people at the center of important decisions

Businesses remain cautious about where AI should be used — and that caution is well founded.

Research into how businesses use AI suggests hesitation is closely tied to trust. Businesses are comfortable using AI where the work is routine and easy to review. Areas involving legal judgments, people management, and financial decisions still depend heavily on human oversight.

The businesses getting the most from AI are usually clear about those boundaries. They use it to handle routine work more efficiently while keeping people closely involved in the decisions that carry real responsibility.

A few years ago, many companies were adding software faster than they could properly use it. Today, smaller businesses are becoming far more selective. Tools stay when they save time, reduce pressure, or help teams work more effectively. They disappear when they do not.

That may explain why AI is sticking. For a growing number of UK businesses, AI is no longer experimental. It is becoming part of the operational basics.

Checkout our list of the best IT automation 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

• BT and Verizon new joint venture could target $4 billion in annual revenue
• The companies are targeting 3,000 global enterprises across 180+ countries
• AI and cloud are the two primary selling points for this new joint venture

BT and Verizon have come together to form a 50:50 joint venture looking to serve around 3,000 multinational customers across more than 180 countries.

The official announcement details the creation of a new "scaled international connectivity platform" designed for a cloud-first, AI-first world.

Martijn Blanken has been appointed as CEO of the new joint venture, the name of which currently remains unconfirmed.

BT and Verizon target multinational enterprises in new JV

By combining their respective expertise, the two hope to provide enterprise customers more resilient global networking, more secure connectivity and infrastructure designed specifically for local laws, regulations and data sovereignty demands.

"Customers will benefit from new, secure and resilient connectivity platforms, which are designed for the age of AI and sovereign where it matters," BT CEO Allison Kirkby noted.

"Our international customers require secure, flexible connectivity that works seamlessly across borders and cloud environments," Verizon CEO Dan Schulman added.

The announcement also implies that, by spinning off a new JV for global enterprises, it would allow the two companies to improve focus on their domestic markets.

BT boasted about its "resilient communication and network services," and Verizon said it would bring secure connectivity via its enterprise wireline arm.

Though it's unclear which type of enterprises the JV will be targeting, it will likely bid on contracts for some of the biggest enterprises given the target of around 3,000 customers. Private enterprise networks, cloud connectivity, cybersecurity and cross-border connectivity will form the basis of the company.

According to the firms, the join venture could target around $4 billion in combined annual revenue.

For the past two years, the generative AI conversation has been dominated by one piece of hardware: the GPU.

GPUs supplied the parallel compute needed to train large language models, and their scarcity quickly became a proxy for AI readiness.

But that shorthand is now incomplete.

The next phase of enterprise AI will not be defined by accelerators alone.

It will be shaped by CPUs, memory bandwidth, cloud capacity, networking, and the workflow systems that allow AI to move from casual experimentation into daily business operations.

AI’s true economic impact will not come from model access; it will come from whether businesses can turn AI into reliable, cost-efficient operational capacity.

AI is Becoming an Infrastructure Problem

The first wave of generative AI adoption was largely experimental. Employees used standalone tools to draft emails, summarize documents, or write code. These ad-hoc use cases were useful, but they did not require companies to redesign how work actually gets done.

The next wave is different. As AI moves deeper into enterprise workflows, IT infrastructure requirements become exponentially more complex.

A customer service tool that drafts a response is simple. An AI system that reads account history, checks policy, updates a CRM, logs the interaction, and triggers a follow-up task is an entirely different beast. This system does not just need a powerful model; it requires compute orchestration, secure data access, software integrations, permissions, audit trails, and fallback logic.

This is where the GPU-centric view fails. While GPUs remain critical for heavy inference, CPUs coordinate how these workloads interact with databases, APIs, security layers, and operating systems. As a result, memory bandwidth, latency, and power availability are becoming the true strategic constraints.

The High Cost of Unstructured AI Usage

The early enterprise playbook was simple: give employees access to powerful tools and see what happens. While this accelerated learning, it also exposed a massive financial vulnerability. Individual, unstructured prompting is expensive, difficult to measure, and hard to tie to tangible business outcomes.

We are seeing a major corrective shift play out among tech giants. Microsoft recently began pulling back internal licenses for Anthropic's Claude Code—which was costing between $500 and $2,000 per engineer monthly due to high token consumption—and is forcing its Experiences and Devices division to transition to GitHub Copilot CLI ahead of its June 30 fiscal year-end.

Similarly, Uber completely exhausted its entire AI coding tools budget in just four months. The ride-hailing giant deployed Claude Code to roughly 5,000 engineers and aggressively stoked adoption using internal leaderboards. The experiment was incredibly effective—assisted systems generated nearly 70% of committed code—but token usage scaled faster than anyone anticipated, forcing Uber's leadership to publicly question the net ROI.

Consequently, the future of enterprise AI will move away from fragmented prompting toward a central intelligence model. Rather than thousands of disconnected interactions, companies will rely on shared intelligence layers—centralized systems that understand corporate data, apply consistent business rules, route tasks across applications, and track performance.

This model is inherently more efficient because the same intelligence is reused across workflows rather than recreated from scratch by individual users.

From Answers to Workflows

The most critical shift in enterprise tech is the transition from tools that answer questions to systems that perform work.

Traditional software is deterministic: a user clicks a button, and a system performs a known action. AI workflows are more dynamic. An agentic workflow can retrieve real-time data, reason through a multi-step process, interact with third-party software, and loop in a human for approval.

This puts immense pressure on the full technology stack. To unlock actual productivity gains, businesses need clean data infrastructure, disciplined governance, and robust integrations. Advanced models are useless if layered on top of fragmented, disconnected corporate systems.

Unprecedented Change Management and the "AI-Native" Workforce

As these agentic systems mature, the impact on global employment will trigger a corporate change management crisis on a scale never before seen. AI will fundamentally alter hiring patterns and role requirements long before it eliminates headcount at scale.

Historically, headcount was the default lever to scale capacity; more customers required more support staff. AI breaks that linear relationship. Instead of asking how many people are needed to handle an influx of volume, leaders will increasingly ask how much of a process can be handled by automated systems.

This environment will aggressively reward adaptability. Professionals who stay ahead of the technology curve, learn to design AI-enabled workflows, and manage systemic exceptions will disproportionately benefit.

Conversely, the risk of displacement is starkest for those relying purely on legacy industry experience. Traditional technical and managerial paradigms are being disrupted by a new cohort of AI-native developers, product managers, and team members. These professionals do not just use AI as an assistant; they build, manage, and think in terms of automated, model-driven systems.

Those who fail to transition from traditional operators to AI-native orchestrators risk being replaced by those who do.

AI Infrastructure is Economic Infrastructure

The broader economic impact of AI will be determined by how deeply it can be embedded into the core systems that run global businesses.

GPUs, CPUs, networking, and data centers form the physical foundation. Agent orchestration, security, and observability form the operational foundation. Together, they dictate whether AI remains a novelty or becomes a scalable business capability.

The GPU race was merely the opening chapter of the AI boom. The next chapter will be defined by the holistic compute, data, and workflow systems that allow AI to do real work at scale. That is the moment AI stops being a tool and truly becomes infrastructure.

Check out our list of the best IT automation 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

Generative AI is capable of some astonishing feats, with image and video generators, in particular, capable of creating stunning visual scenes based on text-based instructions. Plenty of the images you can generate, however, steal the distinctive styles of iconic artists – and many of these artists aren't best pleased.

Imitation is the best form of flattery

Filmmaker Tim Burton, who created heavily stylistic films like Edward Scissorhands and The Nightmare Before Christmas, was among several artists whose iconic styles were showcased in AI-generated images.

BuzzFeed used AI to show us what iconic Disney movies would look like if they were directed by various filmmakers, with a series of garish and creepy results.

Burton, responding to the feature, told the Independent that seeing his own artistry imitated was akin to some cultures believing that capturing your image with a camera takes something away from your soul.

These comments also tap into the debate as to whether media generated by AI can ever be considered 'original' or 'creative' – or if they're simply crude averages.

The rise of AI in the creative industries

Burton's viewpoint mimics the thoughts of the iconic Japanese filmmaker Hayao Miyazaki, the mind behind Studio Ghibli hits such as My Neighbour Totoro.

"You can make horrible things if you want but I want nothing to do with it. It's an awful insult to life," Miyazaki said in response to learning an illustrator was using AI to aid with his work on Boro the Caterpillar.

But AI has increasingly infiltrated the creative industries, with several major studios finding ways to use AI in the production process, but also on-screen elements, such as in the case of Lionsgate's partnership with Runway AI.

Despite the warning from filmmakers, the trend continues, with the likes of studio A24 also partnering with Google to build AI-powered filmmaking tools.

• Despite Tesla trademark application, the Megapod concept already exists
• Megapod's trademark is owned by someone else
• Nvidia and others already dominate this market

Tesla has filed a trademark application for ‘Megapod’ as the company begins to expand beyond electric cars, batteries and solar energy. Already with fingers in the autonomous transportation and humanoid robotics pies with Robotaxi and Optimus, the company is now looking to build modular AI data center infrastructure.

Though the filing is based on an intent-to-use application, meaning that no commercial product is available yet, it describes a self-contained AI computing platform that includes servers, AI hardware, networking equipment, power distribution units, cooling and software.

However, the project and associated trademark application has already hit three big walls – the concept already exists, the ‘Megapod’ trademark is already owned by somebody else, and the market itself is highly crowded with Nvidia, Huawei and others already more established.

Megapod could be related to Megapack

Tesla already uses the ‘Mega’ naming strategy, as evidenced by its Megapack. A battery system that offers a similar commercial proposition to the proposed Megapod, consisting of factory-built complete modules that can be deployed quickly with minimal on-site assembly or construction.

Rather than customers assembling servers, networking, cooling and other infrastructure themselves on-premises, Megapod could arrive as a plug-and-play AI data center, expandable by its modular design.

The news comes around a year after Musk’s company reportedly wound down its Dojo AI training computer project, indicating that it’s no longer gunning for the AI chip market. It now looks like Tesla could be going after more complete physical infrastructure using existing chips, instead.

Mitsubishi already has its own MegaPod, and it possesses a trademark already. And it’s not the first time Musk has faced complications over trademarks, failing to acquire a Robotaxi trademark over it being too generic and facing Cybercab trademark delays after another applicant got in first.

Submer even sells its own MegaPod, described as a data center in a box, adding to the naming complications.

Is there room for Tesla to join the market?

Trademarks aside, if Tesla were to launch a Megapod-type product, it would face stiff competition from established rivals. Nvidia’s DGX and HGX platforms are already commonplace in enterprise deployments, and Huawei has also developed its own solutions based around its Ascend accelerators. Server manufacturers like Dell and HPE also have their own hardware.

However, Tesla could bring its broader experience to the market to entice some customers into its ecosystem. Integration with its Megapack could, for example, give it uninterrupted power supply – xAI has already purchased $1 billion worth of Megapacks. Purchasing power and other internal efficiencies could also keep costs low.

While the company doesn’t really have an existing enterprise customer base, AI startups could be sold on its tech.

Integrated cooling could also be a major selling point, with thermal efficiency now nearly as important as chip performance itself. Nvidia, for example, has already introduced next-generation liquid cooling specifically for its Rubin systems.

As for what’s next, given Musk’s track record and those of his companies, we’re probably more likely to learn about any potential Megapod developments via X posts or surprise launches, rather than blog posts and official announcements.

• Fake tax notices are becoming delivery vehicles for sophisticated remote access malware
• Attackers hide malicious code behind convincing government branding and legal references
• The malware quietly establishes encrypted communication with servers outside the country

A new phishing campaign is using fake income tax assessment notices to deliver dangerous malware to unsuspecting victims across India.

Researchers at CYFIRMA identified the operation, which relies on a fraudulent website built to resemble official communication from the Indian Income Tax Department closely.

The fake portal, hosted on a recently registered domain, presents a convincing assessment order complete with legal references, financial penalties, and urgent compliance language designed to pressure recipients into acting quickly.

How the infection unfolds

Victims who interact with the fake notice are prompted to download a ZIP archive disguised as official assessment documentation and supporting calculations.

Once extracted, that archive reveals a disk image file functioning as a container for the actual malicious payload.

Inside sits a loader program that quietly triggers a second component, a DLL file disguised to resemble a legitimate Windows service.

Researchers found that this loader uses reflection-based techniques specifically built to make automated detection and analysis considerably more difficult.

Both files were obfuscated using a known protection tool, further complicating efforts by security teams to inspect the code.

Once active, the payload behaves like a Remote Access Trojan, granting attackers persistent, encrypted access to the infected machine.

It can collect system details, monitor user activity, check which security software is installed, and silently load additional malicious components on command.

Communication with the attacker's server happens over an encrypted channel, using a hardcoded address traced to infrastructure based in Hong Kong.

These capabilities point toward a financially motivated operation, rather than one focused on immediate damage or disruption, and they closely resemble traits associated with known commodity RAT families such as XWorm.

However, researchers note that conclusive attribution to a specific threat actor remains unconfirmed at this stage.

Why this campaign matters

This is not an isolated phishing attempt but part of a broader pattern of attackers exploiting tax season anxiety to bypass user caution entirely.

CYFIRMA's findings show the same loader-and-payload architecture has previously been linked to ransomware operators, suggesting this infrastructure may serve more than one type of attack depending on the victim.

Up-to-date antivirus software with behavioral detection remains one practical defence against this kind of staged, multi-component malware delivery.

Security researchers recommend that individuals verify any tax-related correspondence directly through official government channels rather than clicking embedded links.

Organizations are advised to restrict the execution of unknown files arriving through archives or disk images, since this campaign relies heavily on that exact delivery method to succeed.

• Fake GTA VI beta keys are already draining cryptocurrency wallets worldwide
• AI-generated scam websites now imitate Rockstar branding with alarming accuracy
• Malware hidden inside fake game downloads can expose banking credentials instantly

Grand Theft Auto VI is not due on consoles until November 19 2026, but official preorders open soon, and cybersecurity researchers have warned criminals are already exploiting the wait with a coordinated wave of fraudulent websites.

Malwarebytes and NordVPN have both flagged sites promising "VIP early access" or exclusive beta keys to one of gaming's most anticipated releases.

The schemes ask victims to hand over money, personal information, or both, often before any real product changes hands.

How the scam works

Some fraudulent sites ask players to pay a few hundred dollars in cryptocurrency for a so-called VIP beta key. This method makes refunds or fraud reports practically impossible once the payment clears.

According to Stefan Dasic of Malwarebytes, GTA VI is "the perfect bait" that can be used by cybercriminals.

The franchise sold hundreds of millions of copies and went 13 years without a new entry — conditions that make hype, and therefore impatience, unusually intense.

Gerald Kasulis of NordVPN said scammers now use AI to mimic Rockstar's official branding so convincingly that polished emails and websites slip past a gamer's usual scepticism.

Some pages invoke the phrase "help us build Vice City," a reference to the game's fictional setting, to create a false sense of insider access.

Victims are sometimes directed to download software branded as an early build, including one fake file called GTA Mobile 6.

According to researchers, this file contains malware capable of letting fraudsters remotely access the victim's device, often bypassing antivirus software.

NordVPN has separately traced some of these fraudulent domains to a wider network with a documented history of spreading banking trojans, infostealers, and ransomware.

Other variants simply harvest names, addresses, dates of birth, or existing GTA login credentials, data that can then be resold.

Several of these scam sites even target PC and Android users, despite Rockstar never confirming that those versions exist yet.

Who is being targeted?

The typical victim tends to be someone too young, too eager, or simply underinformed, and primarily driven by a desire to be first in line for the game.

However, Malwarebytes' assessment of the scam wave reveals that the trick itself is rarely sophisticated, yet it consistently fools people regardless of age.

The character of those falling for these scams goes beyond simple naivety, since urgency and curiosity are what scammers are really exploiting across these campaigns.

Younger players and newcomers to online gaming appear especially exposed, given their relative unfamiliarity with how official preorder and beta access processes normally function.

Neither company has data on exactly how many people have visited these sites or lost money so far.

Rockstar Games has not responded to requests for comment on the ongoing scam wave or its impact on players.

Security researchers are urging anyone tempted by claims of early GTA VI access to pause and verify the source before entering any personal or financial details.

Players who have already entered credentials or payment information are advised to change their passwords immediately.

They should also contact their bank without delay, since cryptocurrency payments in particular cannot be reversed once sent.

Via PCGAMER