OpenAI is in talks to lease a 10-gigawatt data center in Ohio with Nvidia backing, The Information reported. The project could cost $500 billion and be built on federal land.
Key facts
- 10-gigawatt capacity, 100x typical hyperscale data center.
- Potential $500 billion cost, per Columbus Dispatch.
- OpenAI closed $6.6B round at $157B valuation on June 9.
- Nvidia backing signals co-investment in GPU demand infrastructure.
- PJM Interconnection would need to approve ~10% of peak load.
OpenAI is in talks to lease a 10-gigawatt data center campus in Ohio with backing from Nvidia, The Information reported. The project could cost as much as $500 billion and may be built on federal land, according to the Columbus Dispatch. The 10GW figure dwarfs the typical 50-100MW capacity of current hyperscale data centers. For context, a single gigawatt can power roughly 800,000 homes, making this campus potentially larger than the entire grid capacity of some U.S. states.
The Nvidia Angle
Nvidia's backing signals a deepening financial entanglement between the GPU maker and its largest customer. This is not a simple supplier-buyer relationship: Nvidia is effectively co-investing in the infrastructure that will consume its chips. The move mirrors Nvidia's broader strategy of securing demand for its Blackwell and next-generation GPUs by helping build the data centers that house them. Just this week, Nvidia announced NVFP4 4-bit precision for Blackwell GPUs, claiming 1.8x training speedups—the kind of hardware-software optimization that makes a 10GW campus more compute-efficient.
Infrastructure Arms Race
The Ohio campus would be the largest known AI data center project, exceeding even Microsoft's $100 billion Stargate plans. OpenAI closed a $6.6B round at a $157B valuation on June 9, and this infrastructure push suggests the company is betting that capital expenditure—not just model architecture—is the binding constraint on AGI timelines. The 10GW number implies a facility that could host on the order of 10 million Blackwell GPUs, assuming ~1kW per GPU power draw, though actual density depends on cooling and networking overhead.
Reuters independently confirmed the talks, noting the project's scale would require unprecedented coordination with regional grid operators. The PJM Interconnection, which manages Ohio's electricity grid, would need to approve a connection for a load equivalent to roughly 10% of its entire 185GW peak capacity.
What to watch
Watch for the Ohio EPA's air and water permit filings, which would reveal the campus's exact location, power delivery timeline, and whether Nvidia's backing includes direct equity or just GPU purchase commitments. Also track PJM interconnection queue filings for a 10GW load request.
Source: news.google.com
