Running out of power: should data centres go off-grid?

We aren’t delivering energy fast enough to meet the UK’s artificial intelligence (AI) growth aspirations. London, Europe’s largest data centre market, has an electricity grid already at capacity, partly due to high power needs from the data centres already connected. In 2022, housing developers faced up to thirteen years’ delay on building in West London because there was not enough capacity left. The London authorities stepped in to negotiate a compromise with National Grid, but as of 2025 the grid remains congested. For the UK as a whole, grid connections are slow and electricity prices are among the highest in Europe. Even opportunities to repurpose power infrastructure from the coal era, for example Blackstone’s £10 billion data centre development or Aberthaw’s redevelopment in Wales, can experience grid upgrade delays. Now might be the time to consider flexible data centres, including building off the grid entirely.

Photo by İsmail Enes Ayhan on Unsplash

Data centres are energy intensive. They house the computing power needed for cloud services, high frequency trading, and generally making the internet work. All these computers generate heat, which needs significant electricity and water to cool down. Data centre customers have also historically required high levels of reliability, just 26 minutes of permissible outages per year at the highest tier[1]. They typically need a grid connection with access to diversified power sources and use back-up systems like diesel generators to minimise disruptions. AI is increasing demand for data centres even further because of the computational power needed to train models. Data centre capacity in the UK, currently estimated at 1.8 gigawatts (GW), could triple by 2030 according to McKinsey estimates later published by the UK Government.

While the UK is a global financial hub and the world’s sixth largest economy, the viability for data centres is constrained by grid access and energy prices. In the UK, grid connections have historically been delayed by five years on average beyond the original planned connection date. Alleviating grid constraints are critical to realising McKinsey’s British data centre growth projections. The firm reckons even higher growth is possible if energy costs come down. British industrial electricity prices are some of the highest in Europe at over £0.22 per kilowatt-hour (kWh)[2] against the US’s £0.07 per kWh[3]. Timing uncertainty and delays can increase project costs and potentially lead to companies building in more reliable markets elsewhere in Europe.

However, AI-driven data centre demand is not always as time or location-sensitive as other data services. Data centres typically need to be located near demand for their services to deliver connectivity as fast as possible. But because many AI applications are training models rather than serving real-time needs, their proximity to customers matters less. In theory, they can tolerate more flexible operating conditions and could relax constraints around operating at the highest levels of reliability.

Flexible data centres could create opportunities for UK developers to lower their energy costs. Co-locating with solar and wind plants could reduce generation costs to £0.07 per kWh based on projects coming online between 2026 and 2028 under the government’s fixed pricing scheme (Contracts for Difference)[4]. Siting near renewable projects that cannot reliably dispatch their energy to the grid could be even cheaper. The system operator incurred £0.03 per kWh in compensation costs for curtailing 4.6 billion kWh of renewable energy in the first half of 2025, the equivalent of over 1 GW of data centre load[5]. Beyond using renewable energy as their power source, flexible grid-connected projects can also offset some of their energy costs by being paid to switch off at peak demand times.

Data centres could even go off-grid entirely, at least in the initial stages of operation while they wait for grid connections. This avoids the cost and delay of a grid connection and associated grid charges. It also increases the availability of viable development areas for projects, potentially lowering land costs. Even with the UK’s low solar irradiation and the additional cost of batteries, this might still be cost-effective relative to a grid-connected site if a project does not require 24/7 availability. To match grid levels of availability, Loughborough University analysis suggests that energy costs could be less than half of a grid connected site using an off-grid blend of gas, batteries, solar, and wind. Their analysis suggests that even on-site gas generation would be cheaper than grid connected power if gas connections can be delivered faster than electricity, though a multi-year delay for gas turbines may make this option unviable.

However, flexible and off-grid data centre business models are still nascent. In the US, the economics for off-grid data centres theoretically work, especially with large land availability and high solar irradiation. Crusoe uses flared gas for its off-grid systems but has developed a renewable off-grid solution using recycled electric vehicle (EV) batteries, but it is still small scale. Intersect Power, TPG Rise, and Google are developing data centre sites near generation to put less pressure on the grid, but will still ultimately require a connection. Adjusting to a more flexible off-grid model would be a major pivot for the data centre giants’ strategies, and could potentially lead to underutilisation of the data centre infrastructure that may not be offset by the reduction in costs and connection times. It would also require a shift in their customers’ expectations and procurement, although several major tech firms are trialling flexibility in grid-connected data centres already. Planning processes likely also need updating to make it easier to co-develop data centre and energy projects off-grid.

Data centres are a major opportunity for the UK’s economy, but only if we can power them. Unless the UK delivers energy fast, and ideally at low cost, it risks losing AI growth to other countries. The government should continue to focus on building transmission, accelerating grid connections, and bringing bills down. In the meantime, piloting flexible and off-grid data centres for AI applications could unlock more projects in the short-term and keep capacity available for new homes and industry too.

---

[1] Tier IV reliability 99.995%

[2] June – July 2024 prices for very large customers. Data centres are likely to be very large customers if their demand is higher than 8MW.

[3] July 2025 industrial prices, nation-wide average, adjusted at $1.34 per £

[4] Prices are usually provided relative to 2012, 1.44x is the appropriate scalar for prices in 2025.

[5] This estimate is using the 100% uptime load of data centres and is only intended to be an illustrative estimate of what would be possible with the curtailed power.