Nvidia’s $500 Billion AI Financing Plan Runs on Pension Money, and Nvidia May Back Up to a Quarter of a Deal
Nvidia signed memorandums of understanding on Aug. 10 with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to set up financing platforms meant to mobilize more than $500 billion of third-party capital for AI infrastructure “over time,” the chipmaker said. On Sunday, a Fortune commentary by Mehran Gul argued that the deal shows the AI race “shifting from frontier labs to credit desks.” The partnerships remain “subject to execution of the final agreements,” Nvidia’s release says.
Gul writes that Nvidia “will not, for the most part, be lending its own money,” and that the capital comes from pension funds, insurers, sovereign wealth funds and household savings. In a blog post the next day, chief executive Jensen Huang wrote that the $500 billion “is not NVIDIA revenue, a single fund or a commitment to a single customer,” and that the lenders would judge each project’s customer, demand, utilization, cash flow and residual value on their own.
The five largest U.S. cloud companies are on track to spend $730 billion on capital expenditure in 2026 and are expected to cover up to a third of it with debt, Fortune reports. Goldman Sachs chief executive David Solomon said in Nvidia’s release that the bank sees “the new opportunity to create a market for credit backed by NVIDIA compute.”
The part Nvidia keeps
Huang wrote that Nvidia “may provide a residual-value support mechanism for up to 25% of an opportunity,” judged project by project, where residual value means what the equipment is still worth when the financing runs out. Huang said on X that the company has the option to backstop up to $125 billion, or 25% of the potential deals, Reuters reported.
That support sits on the asset Gul calls the weak point. Power plants are built to last decades and earn predictable returns, he writes, while “graphics processing units (GPUs) are short-lived assets; AI demand is unpredictable; and it is too early to project returns.” Huang’s blog argues the other side, saying the A100 chip Nvidia introduced in 2020 is still in commercial use six years later and that one-year rental prices for its newer H100 chip rose from about $1.70 per GPU-hour in October 2025 to about $2.35 in March 2026.
Who the platforms are for
The platforms are aimed at “qualified AI labs, enterprises and AI clouds,” Huang wrote, many of which “do not yet have access to financing at the scale or cost required to build quickly.” Gul writes that the structure widens the field to “a wider bench of competitors beyond the hyperscalers,” the biggest cloud operators, who can borrow on their own names.
That makes Nvidia a seller that, in some cases, also backs what its used chips are worth for buyers who could not raise the money at that scale alone. If a GPU from this year’s batch is worth far less once newer generations ship, that backing is what the pension funds and insurers lean on.
Sources
- NVIDIA press release, Aug. 10, 2026
- Jensen Huang, NVIDIA Blog: NVIDIA AI Factory Compute Is Becoming an Investable Asset Class (Aug. 11, 2026)
- Fortune: The next AI race will be won at the credit desk (Mehran Gul, commentary, Oct. 18, 2026)
- Reuters: Nvidia partners with Wall Street giants to raise $500 billion for AI buildout (Aug. 10, 2026)
Discover more from Bitter Fool
Subscribe to get the latest posts sent to your email.