Why data centre strategy is shifting from single sites to diverse portfolios

Data centre strategy in the UK is moving away from a focus on single, large-scale campuses towards varied portfolios of facilities. The shift isn't about reducing capacity. It is about planning capacity more intelligently.  

Artificial intelligence (AI) is driving higher compute demand. Meanwhile, grid access is limited, decarbonisation is changing power systems, and heat reuse is now commercially viable. As a result, operators and developers are reconsidering not only how much capacity to build, but also its location and format.

Energy demand is rising, but delivery is constrained

The International Energy Agency projects that data centre electricity consumption will more than double by 2030. AI will act as a primary driver. In the UK, this growth collides with grid connection queues, delivery sequencing challenges, and regional capacity limits.  

At the same time, the power system itself is changing. The UK grid is decarbonising rapidly, and renewable energy generation is increasing. However, the amount of energy produced can vary. Wind and solar do not produce power on demand, and data centres require stable, predictable loads to remain resilient.  

Closing this gap demands storage, flexibility, and sophisticated balancing mechanisms. These are improving, but they are not yet universally available at the scale AI workloads demand.  

This has consequences. Capacity is defined by where grid access can be secured and how quickly it can be delivered. We have seen sustainability and resilience become primary factors in design decisions, influencing site selection and phasing decisions from the outset.  

AI is changing

The most popular AI tools, including ChatGPT and Co-Pilot, rely on large language models to operate. They are trained with large quantities of data to operate as a ‘swiss-army’ knife for any of our problems. As such, much of the public debate around AI infrastructure treats them as synonymous with hyperscale facilities that enable these large language models. These models are powerful and compute-intensive, and it makes sense as we are training AI models to do nearly anything across language, reasoning, and data processing.  

However, there is a parallel trend towards smaller, task-specific models. These systems are trained to perform narrower functions more efficiently and often require less compute per task as they can be deployed in more targeted environments.  

The implication of this is structural. Instead of replicating hyperscale infrastructure everywhere, operators are beginning to think in layers: core campuses, metro facilities, and more distributed, aligned to specific use cases. With infrastructure taking years to plan and build, workloads will evolve more quickly during these processes, and so a diversified portfolio gives better optionality.  

Heat reuse is investable

Data centres consume significant power. As such, they also produce significant amounts of heat. Historically, much of this heat has been rejected from facilities as a waste product. In recent years, and thanks to successes seen in countries like Denmark and Poland, that is changing.  

UK Power Networks’ SHIELD programme pairs solar PV, battery storage and distributed data centres to support domestic heating. Themify’s HeatHub captures and stores excess heat for residential use. Octopus Energy Generation has invested £200 million in Deep Green to accelerate the deployment of heat reuse in swimming pools and district heating networks.  

These are no longer pilot programmes. These are capital-backed initiatives with clear commercial cases.  

There are engineering challenges to address. Heat output temperatures must comply with network requirements. If they do not, additional heat pumps or cooling systems are required, which reduces the efficiency of the systems. Whilst retrofitting is an option, it is complex and less efficient than designing systems in tandem from the outset, which will always be more effective. These smaller facilities, when located near heat demand, can unlock the value that large-scale, isolated campuses cannot.  

Data privacy becomes a design constraint

Data sovereignty is the principle that data is subject to the laws of the nation where it is collected, stored, or processed. Companies are becoming increasingly globalised and operating across borders, though this does increase complexity - particularly as fines for misalignment rise in cost. This factor is and will continue to shape procurement and data centre design.  

European and UK organisations increasingly require demonstrable control over data residency, governance, and access. Cloud providers have responded with sovereign offerings, but physical infrastructure must still sit within defined jurisdictions.  

For operators, this can create a challenge to placement. Certain workloads must remain onshore for their clients. This requirement supports smaller facilities catering to domestic markets rather than a limited number of large-scale hubs serving many regions.  

This factor, in combination with grid constraints and heat-integration opportunities, supports the case for a portfolio strategy.  

Efficiency is not a signal of declining demand

Hardware and software efficiency are improving. New chip platforms can deliver higher performance per watt. Software optimisation can reduce overhead. In addition, cooling systems are becoming more effective.  

Efficiency gains, though, do not automatically reduce the facility’s total energy consumption. The lower per-unit computation costs can, in fact, expand the demand. This rebound effect has been seen in other sectors.  

The implication is clear. System-level integration matters more than component-level improvement. Power, cooling, storage, heat reuse, and measurement frameworks have to be considered together, not in silos. Portfolio thinking allows this merging to be adapted to regional conditions rather than imposed uniformly across territories via reference designs.  

What this means for operators and developers

For operators and developers, these factors influence the following:  

• The new model is coexistence. Large-scale campuses remain essential for economies for scale and demanding AI workloads but will increasingly operate alongside smaller facilities.  
• Measurement and reporting will be more formalised with regulations about transparency and sustainability for comparing data centres.  
• Financial decisions and planning approvals will increasingly be influenced by sustainability factors, which will influence where investment flows.  
• Operators who integrate digital infrastructure with energy systems (micro-generation, on-site storage, hybrid grid strategies) will gain an advantage by reducing operational costs, lessening reliance on local power networks, and easing power demand.  
• Modularity will be more important. Facilities that scale and evolve with changing workload profiles will attract long-term investment.  
• Data centres will remain stable assets for investors if they can flexibly meet consumer requirements.  

The next phase of the UK’s data centre development will not be defined solely by scale. Placement and integration will be key factors, particularly for smaller-scale data centres. The strategic decision is no longer ‘cloud vs on-premises’. It will be which mix of facility types can deliver the required resilience, sustainability and performance for specific workloads. For operators and developers, a portfolio strategy is not a retreat from hyperscale ambition. It is a pragmatic response to the market's physical, regulatory, and energy realities.