A U.S. Department of Energy official recently quipped that the Piketon, Ohio site — once known as “A Plant”, short for uranium plant — will soon be called the “AI Plant.” The wordplay points to something concrete: the former Portsmouth Gaseous Diffusion Plant, which enriched uranium from 1954 until 2001, is now at the center of the largest datacenter project ever announced anywhere in the world.
Nvidia is in advanced negotiations to provide a financial guarantee of roughly $250 billion to OpenAI, enabling the company to lease a 10-gigawatt campus that SB Energy, SoftBank’s energy subsidiary, is developing in Piketon, approximately 50 miles south of Columbus. The project’s total cost — including chips — could exceed $500 billion, according to the Wall Street Journal. In a separate but parallel deal, Nvidia is also discussing financing OpenAI’s chip purchases, in a transaction that could reach $350 billion.
The need for an external guarantee is straightforward: OpenAI is not yet profitable and cannot independently secure an investment-grade credit rating. Nvidia’s backing would reassure lenders of the project’s financial soundness. This is not the first time CEO Jensen Huang has deployed this approach: he has previously supported AI infrastructure operators such as CoreWeave, Nebius, and Naver in South Korea, underwriting large-scale builds as a way to lock in future Nvidia chip orders.
The site spans 189 acres in its first development phase and is expected to create around 10,000 construction jobs and more than 2,000 permanent positions. The initial 800-megawatt phase is slated for completion by 2028. The site sits on federal land, with energy managed by the U.S. government: Japan has invested $33 billion in a natural gas project in the area as part of a bilateral trade agreement, with electricity revenues split between the two countries. Commerce Secretary Howard Lutnick is personally involved in allocating energy allocations, while other major tech players — including Microsoft and Google — are already competing for access to the same infrastructure.
Ten gigawatts is a figure worth putting in perspective: it is roughly equivalent to the output of ten large nuclear reactors, and would dwarf any existing datacenter, including a 3.2-gigawatt facility in Georgia valued at $20 billion. The fact that the physical infrastructure for next-generation artificial intelligence is being built literally on the foundations of America’s nuclear industry speaks to a deeper continuity between two energy eras — and to where the country’s strategic resources are increasingly being directed.
