The number is $105 billion. That is not a valuation. It is a guarantee. Nvidia, the chipmaker, agreed to backstop OpenAI's lease obligations for a data center campus in Pike County, Ohio. The filing is a financial instrument, not a press release. It reveals a structure that smells of desperation dressed as partnership.
Let me be clear: Nvidia is not lending. It is guaranteeing. That distinction matters. A loan is a direct claim on cash flows. A guarantee is a contingent liability, triggered only when the primary obligor fails. Nvidia's obligation kicks in if OpenAI goes insolvent or stops paying rent. Then Nvidia covers the shortfall between the guaranteed minimum lease value and whatever SB Energy recovers by reletting or selling the space. The guarantee terminates once OpenAI achieves a satisfactory credit rating. That clause is the tell.
OpenAI's own credit is not strong enough to support $105 billion in leases. So Nvidia steps in. The chipmaker's balance sheet becomes the collateral. In exchange, Nvidia secures exclusive compute provider status and a $1.5 billion equity investment in SB Energy. The structure is a credit derivative wrapped in a supply chain agreement.
Context: The Geometry of the Deal
The PORTS-Pike Technology Campus will be built by SB Energy, a SoftBank-backed developer. The initial phase covers 4.25 gigawatts of information technology load, with an option on another 3.75 gigawatts. Capacity comes online in phases starting 2028. OpenAI will run Nvidia's full-stack DSX platform. Nvidia becomes the exclusive compute provider. SB Energy and SoftBank commit to building at least 10 gigawatts of new generation and investing $4.2 billion in regional grid infrastructure with AEP Ohio.
Jensen Huang called AI infrastructure "land, power, and shell capacity." He is right about the inputs. He is silent about the counterparty risk. The guarantee is a residual value guarantee — a common structure in project finance for physical assets like ships or aircraft. But here, the asset is a data center designed for AI compute. The residual value of a specialized AI data center in a post-hype cycle is unknown. The guarantee assumes Nvidia's chips will remain valuable enough to attract a new tenant if OpenAI defaults. That is an assumption based on perpetual demand growth.
Core: The Incentive Structure — A Moral Hazard Machine
From my perspective as a cross-border payment researcher, I see this as a liquidity engineering problem. The guarantee creates a three-party risk triangle: Nvidia, OpenAI, and SB Energy. SB Energy gets a guaranteed lease from a $3 trillion company. OpenAI gets a data center it cannot afford on its own balance sheet. Nvidia gets a captive customer and a slice of the energy infrastructure.
But the guarantee also introduces a classic moral hazard. OpenAI has less incentive to maintain its creditworthiness because Nvidia is backstopping the lease. The termination clause — "once OpenAI achieves a satisfactory credit rating" — is a carrot that may never be eaten. Credit ratings are lagging indicators. By the time a rating agency downgrades OpenAI, the guarantee is already being drawn.
My 2022 Terra-Luna post-mortem taught me that algorithmic guarantees fail when the underlying asset loses value. Here, the underlying asset is Nvidia's own compute hardware. If the AI market cools, Nvidia's chips lose resale value. The residual value guarantee becomes a poison pill. The same logic applies to the $1.5 billion equity investment in SB Energy. Nvidia is doubling down on the same thesis: AI compute demand is insatiable and permanent.
Data Deep Dive: The 4.25 GW Commitment
Four point two five gigawatts of information technology load is enormous. For context, that is roughly the power consumption of three million US homes. The total campus with the option could reach 8 GW — equivalent to the output of eight large nuclear reactors. The capacity is expected to come online beginning 2028. That is a four-year construction timeline. The commitment extends 20 years.
Liquidity evaporates faster than hype. If OpenAI's trajectory changes before 2028, the guarantee is worthless. The lease payments are conditional on OpenAI's survival. The guarantee is conditional on Nvidia's willingness to pay. The structure assumes both companies remain solvent and aligned for two decades. That is a long time in a industry where product cycles are measured in months.
Contrarian: The Guarantee Is a Signal of Weakness, Not Strength
Most coverage will frame this as a bold bet on AI. I see it as a sign that OpenAI's standalone credit is insufficient. The guarantee exists because the market would not lend to OpenAI on its own terms. Nvidia is effectively providing a credit enhancement that the capital markets refused to offer. The termination clause — "once OpenAI achieves a satisfactory credit rating" — is the market's judgment: OpenAI is not yet investment grade.
Consider the alternative. If OpenAI were creditworthy, it could issue corporate bonds or secure project finance directly. Instead, it relies on a chipmaker to guarantee its rent. This is analogous to a startup founder using their personal credit card to cover office lease. It works until the card is maxed out.
Nvidia's previous announcement on August 10 — financing platforms with six asset managers to mobilize over $500 billion for AI compute — was a different structure. That was a fund, not a guarantee. This is a direct balance sheet commitment. The risk is concentrated, not diversified.
Why would Nvidia do this? Because it is the only way to lock in demand for its next-generation hardware. The guarantee ensures that OpenAI will buy Nvidia's DSX platform exclusively for the campus. The $1.5 billion equity investment in SB Energy gives Nvidia influence over the energy supply chain. The guarantee is a vertical integration strategy disguised as a lease.
But vertical integration carries its own risks. Nvidia is now exposed to OpenAI's operational risk, SB Energy's construction risk, and the Ohio grid's reliability risk. The $4.2 billion grid infrastructure investment by AEP Ohio is a separate commitment, but if the grid is delayed, the data center comes online late. The guarantee clock starts ticking regardless.
My Experience: Why This Feels Familiar
In 2017, I audited three ICOs that promised guaranteed returns. Their tokenomics models assumed infinite liquidity. When the market turned, the guarantees evaporated. I learned that structural guarantees are only as strong as the guarantor's willingness to pay. The 2020 DeFi yield farming experiment taught me that high-yield pools are often artificially inflated by emission tokens. The 2022 Terra-Luna collapse showed that algorithmic guarantees can fail in hours. The 2024 ETF mapping work revealed that institutional bridges often mask counterparty risk.
This deal is no different. The guarantee is a financial instrument that transfers risk from OpenAI to Nvidia. The risk is real, but the accounting is opaque. The guarantee is a contingent liability — it does not appear on Nvidia's balance sheet until it is triggered. Investors should ask: what is the probability of default? The guarantee is not priced. The market is not pricing it because it is not traded. The only signal is the termination clause: when OpenAI gets a satisfactory credit rating. Until then, the risk is borne by Nvidia's shareholders.
Takeaway: The Cycle Will Test This Structure
Regulation lags, but penalties lead. If the AI market enters a downturn, this guarantee will be tested. The residual value of a specialized AI data center in a post-hype cycle is unknown. The guarantee assumes that Nvidia's chips will remain valuable enough to attract a new tenant. That is a bet on perpetual demand growth. History suggests that demand cycles always revert.
The question is not whether this deal will succeed. The question is who bears the loss if it fails. The answer is Nvidia's shareholders. The guarantee is a call option on OpenAI's survival. The premium is the $1.5 billion equity investment. The strike price is the lease payments. The expiration is 20 years.
Volatility is the fee for entry. The fee here is $105 billion of contingent liability. I will be watching the credit rating agencies. If OpenAI is downgraded, the guarantee becomes a liability. If the guarantee is triggered, the loss cascades to Nvidia's balance sheet. The structure is elegant. The risk is real.
Code is law until the wallet is empty. This guarantee is not coded. It is written in legal prose. The law will enforce it if Nvidia refuses to pay. But the law is slow. The market is fast. By the time the court rules, the liquidity will have evaporated.
