AI Datacentre Energy Meets Telecoms Grid in UK 2026

Why the UK connection queue is forcing a rethink

Ofgem's own numbers tell the story: the interconnection queue swelled from 41GW in November 2024 to 125GW by June 2025, with at least 73GW of that tied directly to data centres. The regulator has responded with stricter readiness tests, deposits, and demands for planning and financing evidence before projects get a place in the queue.

NESO's parallel overhaul has already reshaped the pipeline, with more than 100GW of large-scale transmission demand remaining queued: around 13GW of firm transmission demand is expected to connect before 2030, with a further 86GW between 2030 and 2035. Average lead times of five to ten years, with some offers stretching into the 2030s, mean buyers can no longer treat a grid connection as a given at project outset.

AI load is migrating into telecom and edge infrastructure

The IEA's projections make the scale of the shift explicit: total datacentre energy consumption is set to double between 2025 and 2030 to 950 TWh, three percent of global electricity demand, while AI datacentre consumption alone will triple over the same period. Telecom networks are rising in parallel — the IEA put telecom electricity consumption at 260 to 360 TWh annually as far back as 2022, up to 1.5 percent of global electricity use, with mobile networks accounting for two thirds of that total.

This isn't theoretical. NVIDIA has described telecom operators building distributed 'AI grids' across roughly 100,000 network data centre sites worldwide, representing more than 100GW of potential new AI capacity over time. For UK buyers, that means what edge computing entails for businesses is shifting from a latency conversation to a power-sourcing one.

Idle backup assets are becoming revenue-generating infrastructure

Telecom and datacentre sites have long carried backup generators, batteries and, increasingly, photovoltaic and wind capacity — largely dormant until an outage strikes. The commercial case for changing that is now well established: full-stack storage systems combining battery cells, battery management, energy management and power conversion into one integrated platform, alongside AI-driven optimisation, can turn that idle capacity into an active grid asset.

One vendor example illustrates the model. ZTE's approach pairs a 261 kWh dual liquid-cooling cabinet battery energy storage system for edge and core sites with a containerized system that scales from tens to hundreds of megawatts for AI datacentres, zero-carbon industrial parks and grid-side storage. The company reports overall system efficiency above 90 percent, sufficient in some scenarios to replace diesel backup entirely. Revenue comes from peak-valley price arbitrage, demand response and frequency regulation subsidies, green energy certificate and carbon credit trading, and long-term asset leasing.

The UK's own hybrid experiments point the same way

UK developers are already testing this playbook. Innova's proposal for ten projects totalling 4.4GW includes five schemes pairing data centres directly with battery energy storage, aimed at improving connection prospects — though even these hybrid sites are looking at connection dates between 2029 and 2032. National Grid's upgrade work at Didcot substation is designed to support data centre growth while also enabling 650MW of new battery storage, a direct link between rising demand and storage build-out at the transmission level.

Regionally, the economics vary. As industry analysis of AI-driven energy storage deployment across multiple sectors has noted, Southern Europe favours integrated PV-storage deployments scaled up once returns are proven, while Northern Europe's volatile power prices and mature frequency regulation market push operators toward ancillary-service revenue instead. The UK's price volatility and frequency response market fit closer to the Northern European pattern, which is one reason battery storage projects now account for 83GW of the capacity retained after NESO's queue culling.

What this means for procurement and site strategy

For buyers evaluating new capacity, the practical implications are concrete. Sites with existing grid connections at 33kV, 66kV or 132kV are increasingly contested for hyperscale and AI-factory-class loads, so pairing any new build with on-site or near-site storage is becoming a negotiating tool with network operators rather than a nice-to-have. Government-backed AI Growth Zones, alongside a reported £14bn package, are explicitly designed to steer development toward better-powered locations and faster planning routes.

Buyers should also treat edge and telecom-adjacent sites as genuine alternatives to central campuses, not just latency plays. Before committing to a design, it's worth using power and cooling tools and calculate GPU requirements for AI workloads to model realistic load profiles against available grid headroom, then reviewing research on AI servers and vendor configurators such as Dell, HPE or Lenovo against the power envelope the site can actually support.

• •Model storage-plus-compute proposals as a queue-priority strategy, not just a resilience measure
• •Assess whether telecom edge or exchange sites can host workloads faster than a fresh greenfield connection
• •Factor peak-valley arbitrage and frequency response revenue into total cost of ownership calculations
• •Check AI Growth Zone eligibility before finalising a site shortlist