Summary

• Ofgem is exploring whether datacentres should reduce demand during periods of electricity system stress.
• Voluntary flexibility arrangements could give some projects earlier grid access in exchange for load reduction capability.
• Datacentres are becoming part of energy system planning as AI and cloud growth increase demand for high density compute.

Ofgem is considering whether datacentres should be required to reduce electricity consumption during periods of grid stress, bringing the UK’s cloud and AI infrastructure buildout into direct contact with power system reform.

The regulator is exploring mandatory curtailment as a backstop measure for future datacentre demand. It is also considering voluntary flexibility arrangements under which datacentre operators could reduce consumption in exchange for earlier grid access.

The proposals form part of a wider overhaul of the UK’s grid connection process. Connection dates for large energy users can stretch years into the future, with some projects facing waits into the next decade. The queue has become a constraint on industrial electrification, housing, renewables, transport, and digital infrastructure.

Datacentres are a visible pressure point because AI and cloud demand are increasing the need for high density compute. Operators need large, reliable power supplies, often in locations where grid capacity is already constrained. Policymakers also want the UK to attract AI infrastructure, cloud investment, and digital capacity as part of a broader economic strategy.

The tension is difficult to avoid. Datacentres are critical to modern digital services, but their power demand can compete with other strategic needs. If every large user waits passively for a grid connection, projects stall. If every project connects without flexibility, system stress increases. Curtailment and demand flexibility try to make large users part of balancing the system rather than treating them as fixed demand.

For operators, mandatory curtailment would create a serious operational question. Customers buy compute, storage, and cloud services on the assumption of resilience, uptime, and predictable performance. A datacentre that may be asked to reduce load during grid stress needs technical and contractual mechanisms to manage workloads, backup power, service levels, and customer expectations.

Not all workloads are equal. Batch processing, training, backup, testing, and non urgent compute tasks may be easier to shift or throttle. Latency sensitive services, financial systems, public sector platforms, healthcare applications, and critical cloud workloads are harder to interrupt. Flexibility therefore requires workload orchestration, customer segmentation, resilience planning, and commercial negotiation.

The grid debate also affects backup generation. If projects cannot secure timely electricity connections, developers may look to gas networks, onsite generation, or hybrid energy arrangements. Such approaches can help individual projects proceed, although they may create tension with decarbonisation goals unless emissions and system impact are managed carefully.

The UK does need datacentres. Businesses, public services, AI systems, and cloud platforms depend on them. The policy challenge is how digital infrastructure should connect to an energy system already being rebuilt for renewables, heat pumps, electric vehicles, industrial electrification, and new transmission demand.

Datacentres are likely to behave more like flexible grid participants over time. That may include demand response, battery storage, onsite generation, workload shifting, and tighter integration with grid planning. Operators able to provide credible flexibility may gain faster access or stronger planning arguments. Those unable to do so may face longer waits or tighter controls.

The commercial consequences will reach cloud procurement. Customers may ask whether providers can maintain service levels under grid flexibility arrangements, how workloads are prioritised, and whether energy constraints affect availability. Sustainability claims will also have to account for grid impact, not only renewable power purchase agreements.

The physical limits behind AI growth are becoming clearer. Chips and models attract more attention, but power, grid capacity, planning, and flexible demand may determine where compute intensive infrastructure can actually be built.