Become a TechRadar Insider
Every enterprise AI roadmap I've reviewed in the last eighteen months assumes connectivity is a solved problem. It isn't. It is the single biggest reason these programs stall in year two, and it is the one the board never asks about.
CIO at Expereo.
The board signs off. Eighteen months later, workloads are running in the wrong regions, latency is killing the user experience of a tool nobody benchmarked, and the team is troubleshooting an observability gap they didn't know existed. At which point someone writes a second-stage proposal to fix the network, and the cycle starts over.
This is not a hypothetical. In a recent survey of over 800 global technology leaders, 38% said network performance was a direct factor holding back their AI or digital transformation programs. Only 8% said their networks were ready for what is coming next.
Over half reported revenue loss tied directly to network failures in multi-cloud environments. These are not fringe cases. They are the median experience of enterprises currently trying to ship AI.
The three questions boards don't ask
Boards ask about model accuracy, cloud spend, and time-to-value. They rarely ask the three questions that now decide whether any of the above is possible.
Where is the data allowed to sit?
A growing list of jurisdictions, not only in Europe but across the Middle East, LATAM, and APAC, now require enterprises to keep certain data inside national borders. If those rules are not part of the architecture from day one, they become retrofit projects the quarter before a product launch.
Sovereignty is no longer a compliance footnote. It is an architectural constraint that shapes where workloads can run, which providers can be used, and how much the program will actually cost.
Can anyone see what is happening?
Most IT teams monitor applications closely and only look at the network after something has already broken. In a multi-cloud environment, that is too late. The failure mode is rarely a clean outage.
It is a slow degradation somewhere between a data center in Frankfurt and a SaaS tenant in Virginia, and the first person to notice is a customer. Continuous visibility at the connectivity layer is the difference between catching a problem and explaining it to the board after it has hit the P&L.
Does the business keep running when a link fails?
A site in London or Lagos should not go dark because a single fiber cut took out the only path into the city when diverse connectivity, across fiber, wireless, and satellite, can take over in seconds.
Business continuity is not a back-office concern or an annual tabletop exercise. It is the difference between a network that enables a transformation and a network that becomes the reason the transformation failed.
What to do before the next AI budget gets signed
I'm going to say the unfashionable thing. You don't need another transformation program. You need three decisions, and you need them before you sign your next round of AI funding.
First, make sovereignty a gating question on every architecture review. If the team cannot answer "where is this data allowed to live" in under a minute, you are already retrofitting.
Second, demand real-time observability at the network layer as a non-negotiable requirement from whoever runs your connectivity. Not a monthly report. Not a ticket-based escalation path. A live view that someone on your team looks at before a customer does.
Third, stop treating redundancy as insurance and start treating it as capacity. Diverse connectivity is not a luxury line. It is the condition under which the rest of the AI strategy functions at all.
Why this will not fix itself
Connectivity has been sold to boards for a decade as a commoditized utility. That pitch was convenient when the workloads were email and file sharing. It breaks the moment a production AI system sits on top of it.
Trade tensions, conflicts, sovereignty laws, and multi-cloud complexity are not going to slow down to accommodate a procurement process that still treats the network as a line item.
The enterprises that will ship AI at scale in the next two years are the ones that have already made these three decisions. The rest will keep funding models on a foundation their own board has not looked at in a decade, and they will keep wondering why the ROI never arrives.
We've featured the best AI 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