Smarter, not faster: AI infra’s sobering effect on fiber strategy

While the AI data center boom is undoubtedly an opportunity for the fiber industry, it also presents something of a challenge.

Sure, investment in network infrastructure is rarely a bad thing. I can’t help, however, getting flashbacks to the early 2000’s - operators pouring billions into fiber, without proven demand.

This resulted in stranded assets, empty capacity, and ultimately, a bust that hampered telecoms for nearly a decade.

This time around, as serious AI momentum continues to grow, with megawatt campuses, GPU clusters, hyperscaler capex announced seemingly every week, the temptation to ‘follow the boom’ is rearing its head again.

I’m optimistic, however, that the industry has learnt its lesson. Not just to avoid overspend, but to be far more targeted about where fiber actually gets built, what kind of fiber is needed, and which partnerships can make new routes commercially viable.

AI demand is real, but discipline will be crucial

There’s no denying AI demand is there, but it's important operators don’t get caught up in the hype and assume it will be everywhere.

It may sound basic, but the industry needs to read the tea leaves carefully. Data centers have never had the luxury of a ‘build it, and they will come’ approach. The strongest sites tend to sit where demand, power, fiber and customer ecosystems can all be brought together. Things are changing however; data centers are on the move.

Fiber investment therefore, should focus on solid signals, from AI Growth Zones and established data center corridors to cloud on-ramps, internet exchanges and emerging compute clusters. While the latter two have traditionally centered around London (particularly the London Docklands and Slough), the Government’s proposed AI Growth Zones point to a broader regional shift.

Sites in Oxfordshire, the North East, North Wales, South Wales and Lanarkshire in Scotland aim to spread AI infrastructure development away from the capital, bringing data center demand closer to new regional power, land and connectivity ecosystems.

Beyond the Government initiatives investors are also placing bets on other development opportunities outside of the traditional areas.

AI hyperscalers want platforms, not just physical fiber

But even if you do invest in fiber build in all the right places, that’s no longer a guarantee. While location is one aspect (and admittedly, not a ground-breaking one, pardon the pun), there’s a new, arguably even bigger factor that network operators can’t ignore - consumability.

Hyperscalers, neoclouds and AI infrastructure businesses are not always buying connectivity in the same way as traditional telecoms customers. They expect capacity to be identifiable, orderable and scalable at speed. In other words, they want the fiber market to feel more like a platform than a slow procurement exercise.

AI workloads are also pushing future rack-density designs towards 1MW, which places extreme demands not just on power and cooling, but on fiber density and the speed at which capacity can be provisioned.

As these companies scale, providers that can offer API-driven ordering, rapid provisioning, clear visibility of available routes and flexible rerouting will be in the strongest position when capacity decisions are made.

Having fiber in the right place may secure a place in the conversation, but making that fiber easier to consume will increasingly determine who wins the work.

Compute is useless without connectivity

There’s another major factor, however, one that is often overlooked and missed in the media conversation around UK AI infrastructure - backbone connectivity.

As a call back to the opening of this article, AI infrastructure is often viewed as a hardware arms race. With much of the focus falling on megawatts, GPU clusters and hyperscaler capex, with data center capacity treated as a shorthand for infrastructure readiness.

This view missed the wider systems those facilities depend on. Without the right infrastructure backing up development, the UK risks repeating old DC patterns, where new sites sit underutilized due to a lack of customers, or are delayed entirely due to missing fundamental backbone infrastructure. This concern is already starting to bite. According to recent research, 82% of UK DC operators have already delayed new facilities or expansions because of limited access to high-capacity fiber infrastructure.

Take the UK government’s compute ambitions. The UK has said it will need at least 6GW of AI-capable data center capacity by 2030 to support the growing demands of AI. That is a serious statement of intent, but the gap between ambition and deliverability is still understated. The UK’s AI ambitions are often framed in terms of compute capacity, rather than the connectivity and required to make those investments viable.

None of this makes the UK’s AI ambitions unrealistic. But it does mean fiber can’t be treated as an afterthought. Data-center operators, hyperscalers and network providers need to plan connectivity earlier. Developing a clearer view of where capacity exists and where new resilient routes are needed, particularly in more regional locations outside of London or where the landscape is more challenging for development. Alongside how quickly they can be made commercially available.

Interestingly, in recent months we have seen a change in the approach to fiber emerging. Much earlier engagement from the DC developer/operators on fiber availability and timeline to deliver, along with a significant increase in the capacity required from day one, with a much steeper ramp in uptake.

This tells us two things, the forecast in uptake and data movement has shifted, as has the move to further de-risk a development by getting assurance on fiber/connectivity at a much earlier stage.

Check out our guide to 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