Artificial intelligence has a funding problem that dwarfs most corporate budgets. Bank of America forecasts that AI capital spending could top $5 trillion between 2026 and 2030, with roughly $1.2 trillion of it needing outside financing.
Most of the gap sits with Oracle (ORCL), SpaceX (SPCX), CoreWeave (CRWV) and Nebius (NBIS), where spending runs about $700 billion more in projected capital spending than operating cash flow.
Nvidia’s answer arrived on August 10, when it signed deals with six major asset managers to mobilize more than $500 billion in third-party capital for AI infrastructure, according to a Nvidia statement. The pitch is bold. Nvidia shares closed down 2.86% that day, while Apollo and KKR both rose more than 6%.
Lenders want more protection than Nvidia offered
Some lenders want higher guarantees than Nvidia first outlined, even for its industry-leading AI chips. Banking sources and credit managers said the August plan, announced with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, had envisioned chips as collateral with only a limited guarantee.
The structure works like a lease. A financing platform buys the chips, leases them to the customer and collects payments tied to the compute revenue. Nvidia backstops part of the residual value in some arrangements, which also lets the lender charge a lower rate.
The customer gets the hardware, the lender gets a long-dated return, and Nvidia books a sale, as long as the revenue holds.
Chief Executive Jensen Huang has said Nvidia may provide support of up to 25% of an opportunity. He also said he approached only six firms and none turned him down. The money will all be third-party capital, Huang said.
Huang’s case is simple. Chips have become an investable asset class for the first time, he said. Blackstone’s Jon Gray compared the idea to mortgage lenders evaluating homes. Gray said usage at Blackstone portfolio companies surged sevenfold this year, evidence that demand is outrunning supply, CNBC reported.
The fight over how long a chip keeps earning
Huang has said graphics processing units can last up to a decade. Some credit investors see a shorter life. Andrew Chang of S&P Global Ratings said service beyond five years has held true so far, yet S&P takes a conservative view of the chips’ value.
Big cloud companies already disagree with the skeptics. Google, Microsoft and Oracle estimate their AI chips last about six years. Investor Michael Burry says two to three years is closer to the truth.
