For the past few years, the data centre story has been simple: build as much capacity as possible, as quickly as possible, and demand will follow. That logic held when the market was driven primarily by hyperscale expansion and the race to grow cloud storage capacity worldwide. Demand was concentrated, relatively predictable and largely confined to established hubs.

But the growth of AI, the development of large language models, rising inference demand and the emergence of NeoCloud providers are changing that equation. Today, the required compute profile often needs to be understood before a data centre can even be designed, making speculative development outside the traditional cloud market far less likely – and increasing the risk of capex-heavy stranded assets.

The risk now is not that we stop building, but that we continue building for the wrong type of demand. In fact, there is a growing argument that we are not overbuilding – we just aren’t necessarily building what the market will need next.

From a supply race to a demand question

The current pipeline reflects a very specific moment in time: the AI training boom.

That has favoured:

  • Large, centralised facilities
  • Significant power density
  • Locations with existing hyperscale clusters

Developers and investors have moved at pace to meet that demand, particularly in core European markets. But training is only one phase. It is intensive, but it is not representative of how data infrastructure will be used at scale over time. The next phase – inference – presents a different challenge altogether.

Inference changes where and how we build

As AI moves from development into deployment, demand becomes more distributed and more sensitive to location.

Inference requires infrastructure that is:

  • Closer to end users or operational assets
  • Capable of supporting real-time processing
  • Embedded within specific sectors, from manufacturing to logistics

Put simply, we are moving from a scale problem to a placement problem. This shift is already visible in client conversations across the UK and Europe, where proximity, responsiveness and resilience are becoming just as important as sheer capacity. For the market, that means demand is fragmenting – and becoming more selective.

The risk of misaligned supply

While demand is evolving, much of the current development pipeline is still anchored in a centralised, hyperscale model. That creates a disconnect. Building more capacity in established hubs does not necessarily align with where future demand will emerge. In some cases, it risks reinforcing infrastructure in locations that are already constrained – particularly in terms of power and planning.

This is not simply a question of oversupply. It is a question of relevance. Investors are increasingly asking not just whether demand exists, but whether it exists for a particular asset, in a specific location, and for a defined use case. That is a more complex – and more important – test.

The UK: strong demand, growing constraints

The UK illustrates both the opportunity and the challenge. It remains one of Europe’s most important digital infrastructure markets, underpinned by strong cloud demand and a mature connectivity ecosystem.

However, development is increasingly shaped by:

  • Power availability, which is limiting where projects can progress
  • Planning complexity, particularly in established clusters
  • The need to consider alternative and secondary locations
  • Comparatively high energy costs, which can weaken the UK’s cost competitiveness relative to other European markets

At the same time, the UK is well positioned to benefit from inference-led demand, given its concentration of data-intensive industries and urban centres. The implication is clear: the market does not need less infrastructure, but it does need a different type of infrastructure – more distributed, more flexible and better aligned to end use.

A shift in how value is created

This change is also shifting where value sits in the development process. Success is no longer defined purely by speed or scale. It depends on how well projects are aligned to:

  • Actual demand
  • Power and infrastructure constraints
  • Long-term changes in how data is generated and used

That requires more front-end thinking – particularly around site selection, energy strategy and end-user requirements – before construction even begins. In this environment, “build it and they will come” is no longer a strategy. It is a risk.

The bottom line

The data centre market is not slowing down, but it is becoming more complex and more selective. Demand remains strong, but it is changing shape – driven by new technologies, new use cases and new constraints. If the industry continues to develop for a centralised, training-led model, it risks falling out of step with where growth is actually happening. Because increasingly, the issue is not how much capacity we deliver. It is whether we are delivering the right capacity, in the right place, for the next phase of demand.