Nvidia cosigns AI lab leases as credit backstop

Nvidia is providing a $250 billion（約40兆円） backstop to enable OpenAI to lease space at what the article describes as the largest data center ever built, which SoftBank is constructing. The core problem: OpenAI has no investment-grade credit rating and therefore cannot secure a commercial lease on its own. SoftBank, the landlord, views OpenAI as a tenant risk. Nvidia is cosigning the lease — essentially pledging its own credit to guarantee OpenAI's obligation — much as a parent might cosign a first apartment lease for an adult child with no credit history.

This is not an isolated incident. Broadcom executed an identical arrangement for Anthropic a few weeks prior. Nvidia, Broadcom, and Google are now collectively absorbing the credit risk of AI labs that cannot borrow independently. The mechanism is credit enhancement: a risky obligation (OpenAI's lease) is wrapped in the balance sheet of a safer company (Nvidia), so the risk is repriced downward and the loan becomes feasible.

The article raises a critical analogy to the 2008 financial crisis. AIG, an insurance company, backstopped risky mortgages and stamped them AAA, distributing the underlying risk throughout the financial system. Here, risky AI financing is flowing into Nvidia, Broadcom, and Google and emerging "similarly shined up." The difference, however, is that Nvidia has genuine operational reasons to be part of the AI buildout — OpenAI will fill the Ohio data center with Nvidia chips. But this entanglement is precarious. The article notes that General Electric and General Motors were nearly toppled in 2008 when their finance arms became too exposed; Nvidia is now testing its balance sheet to finance its customers' growth. The implicit risk: if AI spending does not sustain, or if OpenAI and Anthropic fail to reach profitability, Nvidia's credit enhancement could collapse.

OpenAI and Anthropic are burning through cash at a pace their revenues cannot support, and they have no collateral or track record to convince traditional lenders to finance them. The solution emerging from Nvidia, Broadcom, and Google is to lend them the creditors' own credit ratings — a form of credit enhancement that, in principle, can work. Spreading risk is how insurance functions. But the article draws a critical distinction: these chip makers are not passive risk-takers. Nvidia does need to be part of the AI infrastructure buildout; OpenAI will fill the Ohio data center with Nvidia chips. This entangles the chip maker's own balance sheet with its customers' survival, mirroring the structure that nearly broke General Electric and General Motors when their finance arms became too exposed to 2008 downturn.

The historical echo is vivid. In 2003, Fed Chair Alan Greenspan praised credit derivatives for distributing risk to "those who are willing and are capable of doing so." Everyone is willing when growth looks infinite. Nvidia's willingness to backstop OpenAI's lease assumes sustained AI spending and the labs' path to profitability — assumptions that look safe only while the numbers keep rising.

No comments yet. Be the first to share your thoughts!