OpenAI is leasing a $500 billion data center from parties who do not share its upside.
The 10-gigawatt project in southern Ohio is being developed by SoftBank’s energy subsidiary at a projected cost exceeding $500 billion. Nvidia is providing a roughly $250 billion guarantee on the lease — a backstop that puts OpenAI’s counterparty exposure on Nvidia’s balance sheet, not OpenAI’s. The electricity that powers the data centers is controlled by the U.S. government and funded separately by Japan under a recent trade deal. The power allocation, the project development, the capital infusion, and the guarantee all come from outside OpenAI. The model is OpenAI’s. The return on the model, if it materializes, is OpenAI’s.
Here is the cap table on the upside.
OpenAI’s equity is held by its founders, employees, and a small group of institutional and strategic investors. Microsoft holds approximately 49% under the 2024 restructuring. The named individual beneficiaries of any equity appreciation are Sam Altman, the OpenAI employee profit-participation pool, and the early venture investors who retained diluted shares through the restructure. SoftBank, through its energy subsidiary, holds project-level equity in the developer of the data centers — not a primary equity stake in OpenAI itself. Masayoshi Son, SoftBank’s chairman, captures the developer-level return; OpenAI’s enterprise value is outside his position. Nvidia holds the guarantee, not the equity. Jensen Huang’s compensation is tied to Nvidia equity, which would benefit from any AI capex cycle the guarantee helps sustain, but Nvidia does not hold an OpenAI equity claim. The Japanese trade-deal capital is structured as a power-purchase funding arrangement, not an OpenAI equity position. The upside, if any, accrues to OpenAI’s holders.
Here is the cap table on the downside.
The guarantee structure works like this. Nvidia’s $250 billion backstop puts a contingent liability of that size on Nvidia’s balance sheet. If OpenAI’s revenue does not materialize at the scale required to service the lease, the loss flows through Nvidia’s income statement and into Nvidia’s market capitalization. Nvidia’s shareholders bear it.
Nvidia’s shareholders are not, on net, a small group of wealthy speculators. The company is held across thousands of institutional accounts, with the largest positions held by Vanguard, BlackRock, and State Street — each operating as custodian for index funds, mutual funds, retirement accounts, and sovereign-wealth vehicles. The California Public Employees’ Retirement System, the Texas Teachers Retirement System, the Florida State Board of Administration, and the New York State Common Retirement Fund each hold meaningful positions. A 30% write-down in Nvidia’s market cap propagates through these vehicles into public-pension funding ratios and into the retirement savings of every U.S. household with a 401(k) holding an S&P 500 index fund.
The power channel is the second. A 10-gigawatt continuous demand is roughly the output of ten large nuclear reactors or two Hoover Dams, and roughly the household demand of 7.5 million U.S. residences. The federal exposure runs through the power allocation itself. The electricity is controlled by the U.S. government and routed through utilities serving southern Ohio — primarily American Electric Power and Duke Energy Ohio. Allocating that capacity to a single private project — without a revenue claim against OpenAI in the public record — is a federal resource commitment priced at wholesale PJM electricity rates.
The state-level exposure is the rate-base subsidy. The transmission lines, substations, and balancing generation required at this scale do not exist. They will be built, and their cost will be recovered through state utility-commission-approved rate base. Ohio households and businesses served by AEP and Duke Energy Ohio will pay for the buildout through monthly bills over the depreciable life of the assets — typically thirty to forty years. If the project is later written down or restructured, the rate base is not. The ratepayer continues to service the financing.
The capital channel is the third. Japan’s contribution is denominated in U.S. dollars and priced against the U.S. Treasury yield curve. If the Fed raises rates, the yen-funded dollar carry cost of that capital rises; if the Fed cuts rates, the dollar weakens against the yen and the dollar-denominated return expectation erodes. Japan’s taxpayers bear the exchange-rate exposure. The return expectations are negotiated in the trade-deal framework, not in the open market.
The labor channel is the fourth. The construction of the data-center complex and its associated power infrastructure employs — at peak — tens of thousands of building-trades workers in southern Ohio. The semiconductor supply chain feeding Nvidia’s GPU production runs through TSMC Arizona, Samsung Austin, and packaging and testing facilities in the Pacific. Cyclical contractions in AI capex lay off construction workers and process technicians in that order. Ohio’s prevailing-wage workforce and the semiconductor labor force in Arizona and Texas absorb the contraction when the financing environment tightens.
Now to the rate decision.
The Fed’s role here is not symbolic. The OpenAI-Nvidia-SoftBank financing structure is denominated in U.S. dollars. The lease carries interest expense priced off the risk-free rate. The Nvidia guarantee is marked to market against Nvidia’s equity, which is priced off discount rates that move with the Fed funds rate. The Japanese trade-deal capital carries a yen-dollar carry cost that moves with the U.S.-Japan rate differential. The Ohio utility rate base carries a cost-of-capital component set by state public utility commissions in reference to Treasury yields. Every component of this structure has a financing cost indexed to a rate the Fed sets or influences.
The dual mandate — maximum employment and stable prices — is the framework inside which the Fed makes that decision. A tightening path that lowers inflation toward target raises the cost of capital across every layer of the structure. An accommodative path that supports employment in the cyclical sectors the project touches — construction, semiconductors, power generation — keeps the structure affordable. The Fed is not asked to subsidize this project. It is asked to set a rate environment that determines whether the structure remains viable. Wednesday’s decision is one input into that determination.
This is a contingent liability on Nvidia’s balance sheet, a federal resource commitment allocated without a revenue claim, a rate-base subsidy paid by Ohio ratepayers, a trade-deal-funded capital injection whose return expectations are politically negotiated, and a labor-cycle exposure absorbed by building-trades and semiconductor workers. The equity capturing any upside is held by OpenAI’s founders, its employee pool, Microsoft as the 49% holder, and the early venture investors who retained diluted equity through the 2024 restructuring.
The Fed sets the rate on Wednesday. The structure was designed for a financing environment where money was essentially free. If that environment no longer exists, the cost migrates to Nvidia shareholders, Ohio ratepayers, Japanese trade-deal capital, and the building-trades workforce — in the order the contracts permit.