Nvidia Backs $164.5 Billion Of AI Data Centers To Keep GPUs Selling

Nvidia has put as much as $164.5 billion of guarantees and commitments behind the data centers that run its chips, according to figures in its latest quarterly filing. The company has also invested or pledged more than $7 billion directly in AI cloud providers.

The timing of the disclosure is hard to ignore. In the same week, Australian GPU cloud Firmus shelved its IPO and turned to Nvidia for fresh capital. CoreWeave’s bonds traded at distressed levels, and Microsoft walked away from a planned Nscale site in England. Any CIO about to sign a multi-year GPU capacity contract now carries counterparty risk that runs through Nvidia’s balance sheet. That risk deserves the same diligence as price and performance.

How Nvidia’s Backstop Works

Let me break down the $164.5 billion, because the headline number hides a very uneven mix.

The largest piece is a guarantee on lease payments for the Ohio campus that SB Energy is building for OpenAI, first announced in August. If OpenAI fails to pay rent, Nvidia pays the landlord. The Information puts that commitment at up to $105 billion, or roughly 64% of the total.

The remainder is spread across three mechanisms. Nvidia has committed to buy cloud capacity that providers cannot sell to anyone else. It has signed data center leases directly and passes them through to customers. It also guarantees a smaller pool of leases signed by cloud providers themselves. The Information reports these at about $36 billion, $20 billion and $3.5 billion respectively.

CoreWeave is the clearest example of how this works in practice. Nvidia has agreed to buy any unsold CoreWeave capacity through April 2032. This puts a floor under the utilization of the largest independent GPU cloud. With this arrangement, lenders financing CoreWeave’s GPU purchases are partly underwriting Nvidia’s credit rather than CoreWeave’s.

Where The Strain Is Showing

The stress is surfacing among the neoclouds that depend on one or two large tenants.