Power: The UK’s next digital infrastructure bottleneck

With most enterprises still heavily dependent on connectivity, power availability may increasingly influence where and how enterprise connectivity infrastructure is built. While artificial intelligence (AI) needs massive electricity, grid connection queues are growing longer, with more datacentres battling for limited power capacity. This leads to a key question: what happens when power availability ends up limiting the UK’s digital infrastructure ambitions? And what could this mean for businesses? “Power has flipped from an assumed utility into a binding constraint. AI training and inference, on top of normal cloud growth, makes hyperscale sites so energy-hungry that grid capacity, – not fibre or land – now decides where compute can physically sit,” says Egiziago Cioffi, CEO of software company SynSphere . “Connectivity planning increasingly has to follow the power map rather than the other way around. A grid connection, or the ability to bring your own power, is now a primary site-selection criterion, sometimes ahead of latency.” Why power has become digital infrastructure’s next bottleneck Digital infrastructure is increasingly dependent on power infrastructure, due to its significant electricity requirements. According to Our World In Data , datacentres use approximately 1.5% of global electricity, or 485 terawatt-hours (TWh), which is the annual electricity generation of Germany. Apart from AI model training and deployment, these datacentres also house servers, storage systems and networking equipment, making up the core infrastructure required for the vast majority of digital services, such as email, online banking, Google Maps and messaging platforms. Currently, there are around 477 datacentres in the UK, but almost 100 more are due to be built in the next five years, according to planning documents analysed by construction researchers Barbour ABI. Modern AI-driven datacentres also require far more power per square foot than before, which has led to significant changes in traditional datacentre design. By contrast, grid capacity has failed to keep up with this soaring demand. According to data from UK energy regulator Ofgem, UK developers have already requested around 73GW of power, equivalent to around 315 datacentres, which is nearly 30GW more than the UK’s peak energy demand of 45GW. There are also considerable equipment backlogs for grid capacity increases, as crucial high-voltage electrical hardware such as transformers and switchgear can have manufacturing lead times of up to five years. Waiting for a high-capacity grid-tie connection can take up to a decade in congested regional markets. As such, speed mismatch remains a key challenge as digital infrastructure providers usually scale at software speed, whereas physical power plans and grid upgrades are often subject to lengthy regulatory and construction periods. “At this point, planning for power and connectivity is no longer separate. The massive compute capability required by artificial intelligence systems, as well as datacentres and cloud services , requires both reliable electricity and large amounts of computing ability,” says Tomas Novosad, founder of Full Fibre Checker . “While an area may have great fibre connectivity, that connectivity has limited value without sufficient power to operate and grow. In some areas of the UK, grid connection delays have resulted in electricity availability being the limiting factor to where new digital infrastructure can be built.” How power constraints could reshape enterprise connectivity One of the biggest ways limited electricity is affecting enterprise connectivity is by forcing organisations to shift away from massive, centralised data hubs. By decentralising and prioritising edge proximity, enterprises are increasingly routing data to rural or secondary markets with better access to power. This helps to break up large energy loads into smaller footprints. However, not all workloads can simply follow the cheapest available power. “Training is a schedulable, batchable load you can site next to cheap power and run overnight. Agentic inference is the opposite: continuous, bursty and user-facing, so it has to sit where the users and the data are. That is what turns a power question into a connectivity question,” says Chase W. Hughes, founder of generative AI SaaS platform ProAI. Organisations may also need to build extra backup communication pathways to support multi-path routing. This is often done by using separate physical fibre routes laid in completely different geographic trenches. Connectivity planning increasingly has to follow the power map rather than the other way around Egiziago Cioffi, SynSphere However, as compute becomes more geographically distributed, enterprises need more high-capacity connections between locations. Similarly, faster data transfer between remote clusters has become crucial, increasing the need for stronger backbone routing and fibre infrastructure. Enterprises may also need to rely far more on robust telecom and WAN due to the remoteness of these compute nodes. Organisations must invest more in latency management too, to compensate for longer physical distances between the user, main corporate cloud and secondary datacentre. In regions with constricted power, enterprises could face higher costs for infrastructure expansion and connecting geographically distributed facilities, as well as longer wait times for new grid capacity. This all needs to be accounted for at a higher management level. Additionally, organisations need to prepare for a divergence between regions where digital infrastructure can grow quickly and places where it cannot too, while planning long-term strategy. If the expansion of underlying infrastructure is delayed by these power limitations, enterprise could ultimately face slower access to new connectivity or cloud capacity in certain regions. As such, these constraints are likely to reshape the very fundamentals of site selection. Enterprises are now less likely to choose sites based mainly on low latency networks or proximity to financial hubs, instead prioritising guaranteed access to regional megawatt capacity much more highly. The hidden challenge: balancing energy and enterprise connectivity planning Although power considerations are increasingly becoming more important for organisations, it is still considerably challenging to integrate them into enterprise connectivity planning. One of the biggest reasons for this is that upper management, such as CIOs, rarely has visibility into local power grid constraints. Although this is now changing somewhat due to rising power bottlenecks, AI-driven energy demands and multi-year connection queues, it remains a considerable hurdle while considering long-term enterprise connectivity planning. Another issue is that energy planning and IT planning teams are often located in different organisational sections or silos. This means that coordination can sometimes be fragmented, with approvals needing to go through several layers. “Most organisations never modelled power as a variable, and facilities, network and procurement teams sit in separate silos,” Cioffi says. “Grid headroom by region is not transparent, and lead times for new power can run to years, far longer than a typical connectivity project. Power has to enter the same design conversation as bandwidth and latency, and today it usually does not.” The current real estate and connectivity constraints seen in many parts of the UK also means that infrastructure decisions are often made years in advance. This can make it significantly difficult to adapt to changing grid availability, especially if the land in question for a project has already been secured and cannot be offloaded easily. This is especially because multi-year construction projects and evolving enterprise demand has made it more difficult to estimate what power will be available when an enterprise actually needs it. Longer wait times around connection timelines and difficulty estimating future AI and cloud workloads has added another layer of complexity to this as well. In many cases, enterprise priorities still revolve around bandwidth, resilience and cost, with electricity availability rarely being seen as a major concern. This is especially as for years – if not decades – electricity availability was considered a given, making it an added struggle to now change this approach. As such, few procurement frameworks account for energy resilience either, which is also likely to take more time to change. Similarly, many suppliers do not always communicate power-related risks. This has increasingly led to CIOs realising that questions around grid resilience and ease of incorporating electricity availability into infrastructure planning are quickly becoming part of long-term strategy. Organisations are also recognising the rising need to designate a particular role or team to address these issues. How businesses can reduce their exposure to power constraints There are several ways organisations can shield themselves against the impact of power constraints at an enterprise level. One of the key things to prioritise is more energy-aware infrastructure planning. This means factoring power availability into crucial decisions about datacentre, connectivity and cloud requirements. “Right-placing workloads instead of piling everything into one mega-region; hybrid and multi-region designs so a power-constrained region is not a hard dependency; choosing cloud regions by energy profile and cost, not just latency; and, for on-premise, higher-density and liquid cooling plus on-site renewables and storage,” Cioffi says. “Some enterprises are repatriating steady-state workloads to controlled sites specifically to escape volatile hosting economics.” Geographic diversification is vital as well to avoid overreliance on infrastructure in a single constrained location. Workload flexibility and routing is another way to reduce the effects of limited power. This involves treating electricity availability such as bandwidth and routing tasks dynamically between locations or times of lower demand. Similarly, it may become necessary to spend much more on multicloud backups, as enterprise workloads need to be divided between different cloud providers to ensure power fluctuations or outages at one company does not stall operations across the board. On-site power generation and battery storage has become crucial, with more enterprises now choosing to bypass congested public grid queues through microgrids, renewable energy sources or dedicated local power deals. Organisations are focusing more on energy-efficient infrastructure, such as more efficient servers, networking equipment, cooling and datacentre design. This helps to reduce overall energy demand in the long run so that power constraints are not felt quite as keenly. In many cases, a change is required at the procurement level as well. This means asking cloud, colocation and connectivity providers earlier and more clearly about their available power capacity and plans for future expansion. However, enterprises cannot solve the power constraint alone, as certain changes require closer coordination with regional authorities. In cases where public grids are absolutely necessary, faster and more strategic grid investment is necessary, through joint regional planning by both grid operators and infrastructure developers. Not only does this help to locate digital infrastructure closer to where grid capacity is available, but it also supports scaling flexibility. Stronger coordination between grid operators and digital infrastructure developers is also needed, for earlier visibility into what capacity is realistically available in a region and when, and better alignment of timelines and construction schedules. This helps timelines to align better, with enterprises being able to share their expected power requirements and construction schedules earlier, so that grid upgrades can be planned alongside the project. Similarly, more transparent connection information is required, such as clearer information about queues and likely connection dates. Better and more accurate demand forecasting metrics around expected AI and datacentre electricity demand in the medium and long-term are increasingly necessary now to help decrease power bottlenecks. Towards power-first digital infrastructure For years, the UK’s digital infrastructure strategy mainly focused on expanding connectivity, with power being seen as a given. However, as AI, cloud and connectivity demand steadily rises, energy infrastructure and grid capacity are increasingly determining where and how fast the UK’s digital economy develops. This has led to several enterprises slowly recognising the need to treat power availability as a key strategic and infrastructure priority to be embedded in long-term enterprise connectivity strategy. This means planning around grid capacity and accounting for connection delays, before making any expensive, long-term site decisions. “In the next five to 10 years, power availability stops being a facilities concern and becomes an architecture decision that engineers make – the same way latency did,” Chase says. As such, organisations who prepare accordingly now could be in a much stronger position to adapt to evolving electricity demand. Read more about network infrastructure demands Cisco: 36 months to modernise networks before AI overwhelms capacity : Research finds capacity and performance the top network challenge for UK organisations, with 81% of respondents saying their network does not have room to house evolving AI demands. ‘Soaring’ AI demand accelerates Lumen’s multibillion-dollar network expansion : Progress report from metro and long-haul data connectivity firm shows pace quickens on network expansion to meet massively increasing workloads including new fibre miles, significant added network capacity. The demands of AI on network as a service : An expert panel at ONUG’s AI Networking Summit Spring 2025 gave insight into the future of AI in networking, particularly in NaaS environments. Rocketing AI demand sees Lumen light up custom networks division : Connectivity provider shakes up corporate structure in response to artificial intelligence economy changing business operations and companies recognising need for powerful network to manage unprecedented data flow.