Lead Analysis
Markets6 min

Nvidia Negotiates $250 Billion to Finance OpenAI's Data Center in Ohio, the Largest Project in its History

Paisagem rural de Ohio ao entardecer com a antiga usina de enriquecimento de urânio de Portsmouth ao fundo, cerca em primeiro plano e torre de transmissão no horizonte

According to the Wall Street Journal, Nvidia is negotiating to secure up to $250 billion in financing for OpenAI to lease a 10 GW project in Piketon, Ohio, built by SB Energy, a SoftBank company. Combined with the chip contract, the campus exceeds $500 billion.

Nvidia is negotiating to offer approximately $250 billion in financial guarantees to OpenAI for leasing a 10-gigawatt data center campus in Piketon, Ohio, located at the site of the former Portsmouth uranium enrichment plant. The original report was published by the Wall Street Journal this Sunday, with confirmation from Bloomberg and coverage by Reuters. The project's developer is SB Energy, a subsidiary of SoftBank.


The $250 billion covers the leasing of the campus and the debt used to build it. It does not include the chips. In a separate tranche, Nvidia is negotiating with OpenAI for an additional line of up to $350 billion to facilitate the purchase of the accelerators that will operate within the complex. Combining both vectors, Nvidia's commitment to a single customer exceeds $500 billion, marking the largest data center project ever announced.


The first phase, comprising about 800 megawatts, is scheduled for delivery in 2028. This positions the campus to operate at full capacity at some point in the early 2030s, when the training and inference curve of frontier models will multiply current requirements several times over.


The Strategic Reason is Clear, the Financial Instrument Not So Much


For OpenAI, the agreement is the first step toward controlling infrastructure rather than renting from Microsoft, Amazon, and Oracle. Every dollar saved on cloud contracts goes straight to the margin, and the company no longer has to negotiate GPU SLAs with a supplier that is also a competitor. For Nvidia, the benefit is even more straightforward: to secure multi-year demand for Blackwell and its successors in a contract that is nearly impossible to cancel. Nvidia's CFO, Colette Kress, has previously explained in calls that the company operates with four years of order visibility.


Michael Burry, the investor of the 2008 thesis, wrote that the design is 'around and around we go': Nvidia guarantees the debt that finances the data center, the data center buys Nvidia chips, and the customer pays usage royalties that return to the balance sheets of all involved. The same comment could have come from any analyst studying vendor financing since the ties between Lucent, Nortel, and telecom operators in the late 1990s.


What the Comparison with Lucent Shows and What It Ignores


The analogy with the telecom bust has merits. In that cycle, suppliers lent tens of billions for customers to purchase their equipment. When the demand for traffic did not materialize at the promised rate, the loans became write-offs and sank the balance sheets of creditors.


The material difference this time is the profile of the parties involved. Nvidia operates with an operating margin above 60% and cash exceeding $60 billion. SoftBank has already demonstrated the ability to roll over assets like Arm and is the second-largest shareholder of OpenAI after Microsoft. The site in Piketon has energy committed by the U.S. Department of Energy, which maintains exclusive access to the electricity grid of the former uranium plant, reducing the risk of the project stalling due to grid bottleneck, the issue currently delaying data centers in Virginia and Texas.


The fragility, however, remains on the cash flow side: OpenAI is still not generating a positive operating margin, and the company relies on successive funding rounds to sustain recurring expenses. If the subscription returns from ChatGPT Business and the APIs slow down before 2028, financing will depend on refinancing under conditions that no one can predict.


How the Effect Propagates Outside the U.S.


In Japan, SoftBank's exposure concentrates risk. Masayoshi Son has already committed $1 billion from his own balance sheet to SB Energy in Piketon, along with a $3 billion plan for a facility in Ohio for data center equipment. The stock's trajectory on the TSE in the upcoming quarters will reflect market perceptions of the solidity of this commitment.


In Europe, the effect is one of capital cost. If Nvidia ties up hundreds of billions with a single customer, the hardware available to smaller clients becomes scarce, and the unit price of GPUs in the spot market rises, impacting the architecture of sovereign clouds being constructed by companies like SAP and OVHcloud in Germany and France.


Piketon thus becomes both a symbol of the peak concentration of AI infrastructure and a stress test of how much counterparty risk can be accommodated on a balance sheet.

Lead Analysis