Drax's £20bn AI datacentre plan, and the nuclear question it leaves open

Drax owns the biomass plant in North Yorkshire and sits in the FTSE 250. On 3 November it said it had struck a deal with the infrastructure investor Brookfield and the AI infrastructure platform Radiant, according to The Guardian. The partners will develop what Drax calls the country's largest datacentre, with an eventual capacity above 1 gigawatt.

Phase one is modest by comparison: up to 100MW of compute capacity, operational from the first quarter of 2028. Drax will supply land and access to the national grid through existing connections on the site. Direct power from the plant is a later ambition, pencilled in for the early 2030s. It depends on regulatory approval the company hopes to secure before 2031.

The money is the headline. Drax says the development could attract more than £20bn ($26.8bn) in investment. Its market value rose almost 9% on the London Stock Exchange to almost £3bn on the day, The Guardian reported. Chief executive Will Gardiner framed it as a way to "help secure the long-term future of the site and protect hundreds of jobs" and to support "the north's reindustrialisation and strengthening sovereign AI capability".

Why a biomass plant is in an energy story

The context matters for anyone tracking the nuclear build-out, because the same grid and the same political capital are being contested. Drax burns woody biomass pellets and produces almost 5% of UK electricity. That makes the site the single largest emitter of carbon dioxide, according to analysts cited by The Guardian. The government does not count those emissions because Drax claims equivalent carbon is absorbed by regrowing trees. Ministers have promised the plant a "much more limited role" after concerns over sustainability claims. From 2027 to 2031 it will have access to only half its current £2m a day in subsidies.

That subsidy cliff is the commercial logic of the datacentre deal. A power station losing half its support needs a new revenue line. Land plus a grid connection is exactly what a hyperscale campus needs. The deal landed a week after the government published a draft national policy statement on datacentres. That document described them as "as fundamental to our economy as the railways and electricity networks on which we have long depended".

The unresolved question is supply. The UK's grid is congested, and its legally binding climate targets limit how much new fossil generation can be bolted on. The Guardian notes that this has raised questions over how a dash to build datacentres can be powered without leaning on expensive and polluting off-grid gas. Nuclear, in this framing, is the slow answer to a fast question.

The battery argument, and the gas default

Across the Atlantic the same debate is running with more numbers attached. On 3 November Canary Media reported on a new study from Georgetown and Duke researchers. The study argues that batteries, not gas plants, should be the first choice for datacentre demand. U.S. utilities are planning more than 60 gigawatts of new gas plants for datacentres, the report says. Around 90 GW of behind-the-meter gas generators have been proposed nationwide.

Ian Hitchcock, a report co-author at the Duke Initiative for Science and Society, told Canary Media that "there's a lot of inertia and status quo pressure for much of the behind-the-meter resources being natural gas". The stranded-asset risk is the argument that travels: if datacentre proposals fall through, other utility customers pick up the bill for overbuilt gas. Batteries, the researchers argue, are modular enough to be scaled back. Jackson Ewing of Duke's Nicholas Institute said batteries are "already showing serious momentum in terms of their deployment trajectory", unlike advanced geothermal or next-generation nuclear.

That last point is the awkward one for nuclear advocates. A BloombergNEF forecast cited by Canary Media sees U.S. energy storage deployment surpassing 11,000 gigawatt-hours by 2036. That is a 30% increase on its forecast from six months earlier, with datacentre demand driving much of the growth. Storage is being ordered now. Reactors are not.

Where nuclear money is actually going

Public nuclear spending is aimed at fuel, not at new plants. On 2 November HPCwire reported that Sandia National Laboratories has been selected for $3.38 million in Department of Energy funding. The money will build analytical tools for next-generation reactor fuels.

The money sits inside HORNIG, a program worth up to $50 million, run by ARPA-E and focused on fuels made from materials recovered from used nuclear fuel. Sandia's project, TRU-PATH, will assess safety, safeguards and the economics of scaling production. Announcing it on Friday, DOE Under Secretary for Nuclear Security Brandon Williams said the investment would "build on Sandia's capabilities to help fuel developers overcome the barriers to commercial-scale manufacturing". Deborah Frincke, Sandia's deputy laboratories director for integrated security missions, said the project is "purpose-driven to look at what brings the nation the safest and most secure energy solutions".

Read the numbers together and the picture is clear. A single datacentre deal in Yorkshire is worth £20bn. The U.S. government's newest nuclear fuel analysis program is capped at $50m, and Sandia's share is $3.38m. Fuel-cycle research is necessary groundwork, but it is not a construction schedule.

Who is actually building generation

The fastest capacity being added is renewable. Also on 3 November, pv magazine reported an Ember analysis finding that hybrid wind, solar and battery plants could supply up to 90% of Kazakhstan's typical demand profile. The delivered cost would be below $60/MWh. Across 15 sites and 1,875 hourly simulations, Ember puts the average delivered cost at $56/MWh against $70-170/MWh for new coal.

Kazakhstan's coal units are on average over 50 years old, and Ember counted 891 incidents of various severity between January and July this year. Demand rose 16% in five years. Kostantsa Rangelova, a senior battery analyst at Ember, told pv magazine that solar is "really cheap and very fast to build here". The country's wind is strongest in the winter months when demand peaks. That is the shape of the competition nuclear faces in fast-growing markets: not another baseload plant, but a hybrid that can be built in slices.

Manufacturing is moving the same way. On 3 November pv magazine reported that Chinese cell maker Yingfa Ruineng has unveiled a 7.5 GW n-type TOPCon plant in Indonesia, its first overseas base. The company says its four sites total 51.5 GW of effective capacity, including 45.5 GW of TOPCon. InfoLink ranked it second for global cell shipments in the first half of 2026 behind Tongwei.

The demand side is not guaranteed

Every projection above assumes datacentre demand keeps climbing. Boom Supersonic is a useful stress test of that assumption in another sector. Ars Technica reported on 30 October that United Airlines chief executive Scott Kirby doubts Boom's 2029 delivery target for the Overture airliner. "'29 is their official timeline," Kirby said on the Office Hours: Business Edition podcast on 28 October. "It's not going to happen in '29, it's going to take longer."

Boom's pivot is the relevant part. On 22 October the company unveiled a factory in Centennial, Colorado, for a 42-megawatt natural gas turbine aimed at AI datacentres. A supersonic hopeful now sells gas turbines to data centres. That is where the money is going, and it is gas, not fission.

Smaller examples of the same energy squeeze keep appearing. NASA reported on 30 October that astronauts Anil Menon and Joshua Kutryk completed a 7-hour, 11-minute spacewalk. They installed the seventh roll-out solar array on the International Space Station. The upgrade will lift available power from 160 kilowatts to roughly 215 kilowatts, a rise of as much as 30%, with the eighth array due on 5 November, ScienceDaily reported. Even in orbit, the answer to a power shortfall is more panels.

The regulatory queue is the real constraint

None of this makes nuclear irrelevant. It makes it slow. The Drax project needs regulatory approvals before 2031 to supply the datacentre directly, and the government's draft policy statement on datacentres is still a draft. The gap between a grid connection today and a reactor online in the 2030s is where the gas plants, batteries and hybrid parks are being built.

There is a wider governance argument in the mix. Writing in the South China Morning Post on 2 November, former editor-in-chief Wang Xiangwei argued that China and the United States should treat the malicious use of AI, and the long-term risk of autonomous systems, as a shared adversary. He wrote that if the two cannot agree rules, a global body of respected experts should be set up. Energy supply is the physical substrate of that argument.

For now the ledger is lopsided. One biomass site in North Yorkshire is being repackaged as AI infrastructure with £20bn of promised investment. A U.S. storage market is forecast above 11,000 GWh by 2036. The newest federal nuclear money is a $3.38m analysis grant. If nuclear construction is the topic, the honest answer from these sources is that the construction is happening elsewhere.