AIANALYSISPODCAST
NVIDIA's $105 billion contingent guarantee turns a chip company into a data center financier, and the bet now sits behind 8 gigawatts of grid and demand that have not yet been delivered.
OpenAI's planned 8-gigawatt AI campus in Pike County, Ohio, is the largest single-site data center project announced in the United States. NVIDIA is guaranteeing up to $105 billion of the revenue that is supposed to flow from it. What matters more than the size of the campus is who is on the hook for the power bill.
Eight gigawatts is roughly the peak demand of a mid-size US metropolitan area, or close to the output of several large nuclear reactors running flat out. It is a power plant that happens to compute, and it is being built on a site that until recently grew corn. OpenAI's announcement frames the campus as the next phase of US AI build-out, with the company committing $164 million to local Ohio projects as part of the deal.
Wire coverage has reported the deal as NVIDIA writing a $105 billion check. The SEC filing describes something different. The guarantee is contingent: a promise that if a buyer of NVIDIA's chips cannot pay, NVIDIA will, up to a cap. The cap is $105 billion. The exposure is real, but it sits behind a stack of conditions. The data center has to be built. The grid has to deliver the power. Demand for the AI compute has to materialize to fill the racks. Only then does the contingent structure turn into NVIDIA revenue; only if it does not, does it turn into NVIDIA liability.
The project is PORTS-Pike. The project site and the full SEC text describe a build-out phased in stages, tied to power and capacity milestones. Project early-stage estimates have put total capex at roughly $500 billion, a figure that the Chinese-language analysis podcast Eric观察 episode 080 relayed from preliminary planning and that has not been independently confirmed as a final number. Even as a planning estimate, it puts the campus in the same capital range as the largest US infrastructure projects on record.
The first critical question is the grid. Connecting 8 gigawatts of new load requires a multi-year interconnection process, a large-transformer supply chain that is already booked out, and a generation build-out that has not been ordered. Ohio local coverage reports the project is moving through state-level siting steps, but a binding interconnection agreement on the projected timeline has not been disclosed. If the grid slips, the chips sit idle. NVIDIA's guarantee does not trigger, but neither does the revenue.
The second question is demand. A contingent guarantee only pays if a customer cannot. If AI compute demand grows as projected, the chips sell at full margin and NVIDIA keeps the upside. If demand plateaus, NVIDIA is the residual claimant on a stranded power asset. The deal hedges OpenAI's balance sheet against NVIDIA's. That is not how chip companies have historically been paid, and it is one reason trade press has flagged the structure as unusual for a vendor financing arrangement.
The shift is not that AI got more expensive. It is that the bottleneck moved. The 2024 and 2025 story was who could buy the most GPUs. The 2026 story is who can deliver the gigawatts that make those GPUs useful. The same project, read through OpenAI's announcement on one hand and the SEC filing on the other, tells two different stories: a public statement of partnership, and a private structure in which a chip company is now a data center financier.
Three things to watch. Whether the first phase of PORTS-Pike reaches a binding interconnection agreement on the timeline currently reported. Whether the $500 billion build-out estimate holds as the project moves from planning to procurement. And whether NVIDIA's quarterly filings begin to disclose the contingent guarantee as a balance-sheet item rather than a footnote. The campus is a promise. The guarantee is the test of whether the promise is funded.