DOE Genesis Mission Picks 278 AI Science Projects as Data Center Deals Advance on Federal Land

The Department of Energy runs seventeen national laboratories and some of the fastest computers in the world. Since November 2025 it has also been running the federal government’s main bet on AI for science.

Executive Order 14363, signed on 24 November 2025, launched the Genesis Mission. It told the Energy Secretary to list at least twenty national science and technology challenges within 60 days, to find computing resources and data within 90 to 120 days, and to show an initial working version of a shared AI platform on at least one challenge within 270 days, which fell in late August 2026.

Here’s where it stands.

Projects and partners

In December 2025 DOE signed memoranda of understanding with 24 organisations. The list reads like the AI industry’s top table: Anthropic, OpenAI, Google, Microsoft, AWS, Nvidia, AMD, Intel, IBM, Oracle, xAI, Palantir and others. Genesis director Darío Gil said in February that work would be organised around 26 challenges identified by DOE’s own offices. DOE hasn’t disclosed a total budget.

On 22 July 2026 the department selected 278 projects for award negotiations from a $293 million funding call. National labs lead 87 of them, universities 168, companies 19 and non-profits four, with 342 institutions taking part. The biggest is a three-year, $60 million nuclear energy project. DOE is careful to say a selection isn’t yet a funding commitment.

A first public demonstration of the Genesis platform, with model access, simulation agents, lab datasets and tools for connecting AI agents to them, was shown at a DOE technology summit in late July.

The machines

The computing comes from deals struck in October 2025. With Nvidia and Oracle, Argonne National Laboratory is getting Solstice, built on 100,000 Nvidia Blackwell GPUs, and Equinox, with 10,000, due in 2026. Los Alamos is getting two systems on Nvidia’s next-generation Vera Rubin platform with HPE, the first expected in 2027. Oak Ridge struck a $1 billion partnership with AMD: Lux, built with HPE and Oracle and due online about six months after the deal, and Discovery, due for delivery in 2028.

The model is public-private. The labs get machines they couldn’t buy on their own budgets. The companies get placement in the national research system and a share of the capacity.

Data centres on federal land

The second track uses DOE’s land. In July 2025 the department picked four sites for private AI data centres paired with new power: Idaho National Laboratory, the Oak Ridge Reservation in Tennessee, the Paducah Gaseous Diffusion Plant in Kentucky and the Savannah River Site in South Carolina.

Two now have partners. At Paducah, Brookfield plans a 1.8-gigawatt AI campus, with NextEra building about 2 gigawatts of gas generation and up to 2.6 gigawatts of batteries. The companies talk of more than $100 billion in private investment and completion in 2031. At Savannah River, the National Nuclear Security Administration selected Amentum to negotiate a phased lease for a 1-gigawatt data centre with about 2 gigawatts of on-site generation, gas first and nuclear later. Neither is a final lease. As of this week, Oak Ridge and Idaho had no named developer.

The power plans matter as much as the computing. A gigawatt-scale data centre draws about as much power as a mid-sized city, and DOE is already using emergency orders to keep old plants online. Building generation on site keeps that new load off the public grid, at least on paper.

How other governments do it

Other governments are building public AI computing too, with a different emphasis. The EU’s approach runs through its EuroHPC supercomputers and planned AI factories, owned jointly by the EU and member states and opened to researchers and start-ups. Britain has backed national AI research computing at universities. Those are mostly public machines, publicly owned.

DOE’s version leans on private capital far more. The government brings land, labs, data and scientific problems. Industry brings chips, money and, at Paducah and Savannah River, its own power plants. That gets scale fast. It also means the national AI-for-science effort depends on commercial partners’ timetables, and on deals that are still being negotiated.

The 270-day deadline for a working platform has passed. Whether DOE formally met it isn’t clear from public records. The next real test is whether the 278 selected projects turn into signed awards, and how much money follows.