Sixty years ago, Pike County, Ohio was quietly enriching uranium for America’s nuclear arsenal. Now Nvidia is dropping $1.5 billion into the same 640 acres — and backing OpenAI’s 20-year lease there with up to $105 billion in credit and residual-value guarantees, per SEC filings disclosed August 17. This is the moment AI stopped being a software story. It became a heavy infrastructure story, and the bill is staggering.
From Uranium to Gigawatts: What’s Actually Being Built
The scale here isn’t just impressive — it’s the kind of numbers that make you re-read the sentence twice.
The campus starts at 4.25 GW of compute capacity and scales toward 8 GW, with broader site plans targeting up to 10 GW. For context, a typical U.S. nuclear reactor generates roughly 1 GW. Powering all of that is a 9.2 GW natural gas plant — often described by project backers as one of the largest dedicated to a single campus anywhere — valued at approximately $33 billion. The whole operation sits on 640 acres of DOE-owned land at the former Portsmouth Gaseous Diffusion Plant in Pike County. AEP Ohio is leading roughly $4.2 billion in transmission upgrades to connect it all to the grid.
Nvidia Isn’t Just Selling Chips. It’s Acting Like a Bank.
The $105 billion guarantee structure is less a business deal than a declaration of total commitment.
Those guarantees backstop OpenAI’s lease commitments and reduce lender risk. If OpenAI defaults, Nvidia covers the shortfall — up to that $105 billion ceiling. In return, Nvidia becomes the exclusive compute supplier for the entire campus. That’s vertical integration taken to a nearly absurd extreme — imagine your laptop manufacturer also co-signing your mortgage and owning the power plant keeping your lights on.
The energy math underneath all this is uncomfortable. According to BloombergNEF, “costs to build natural gas-fired power plants in the US increased 66% between 2023 and 2025,” driven by turbine prices, stressed supply chains, and hyperscale demand. Average capital costs for combined-cycle plants climbed from under $1,500 per kilowatt to roughly $2,157. Analysts warn that as AI campuses compete directly with LNG export markets for gas supply, regional prices could spike significantly.
The Uncomfortable Trade-off Nobody’s Pretending Isn’t There
You can hold both the economic optimism and the environmental concern at once — this project demands it.
The DOE frames PORTS as Cold War infrastructure reborn — a public-private triumph funded largely by Japanese capital under a U.S.-Japan strategic framework. SoftBank, which sold $5.8 billion in Nvidia stock in late 2025 to redeploy toward AI infrastructure, is now an investor in SB Energy alongside OpenAI. The capital circle here is remarkably tight.
Local activist groups aren’t applauding. They flag legitimate concerns:
- air quality
- water impacts
- the climate math of constructing one of the world’s largest dedicated gas plants to train AI models
Both narratives are valid simultaneously, which is precisely what makes this project worth watching closely.
AI infrastructure is now indistinguishable from energy infrastructure. Chipmakers are commissioning power plants. Former nuclear sites are becoming GPU farms. The next serious debate about AI won’t happen at a developer conference — it’ll happen at a utility commission hearing.
