Nvidia wants Wall Street to mortgage AI chips like houses

Nvidia chief Jensen Huang wants Wall Street to treat a rack of AI chips like a house. In a post titled "AI Factory Compute is Becoming an Investable Asset Class," Jensen Huang laid out a plan for Goldman Sachs, BlackRock, and others to create independent financing platforms that would raise $500 billion from outside investors for AI data centers.

The structure works like a mortgage. A buyer borrows money to buy computing systems built around Nvidia chips, rents out the computing power to customers, and repays the loan from that rental income. Nvidia would act as matchmaker, introducing buyers that need computing to banks and investment firms that can fund it.

The feature that makes the loans cheaper is a floor price from Nvidia, called a residual value guarantee. A lender can bring a deal to Nvidia, and Nvidia promises a minimum rental rate for those chips in three or four years, absorbing about 25 percent of the difference if the market falls below that level. Nvidia can price that promise because it sees detailed data on computing supply and demand.

In return, Nvidia would keep a share of revenue above that floor. In the view of David Sacks, a former White House AI adviser and investor, those payments would work like royalties and could grow into a large cloud business that requires little of Nvidia's own capital.

Gavin Baker, an investor, argues the bet can last because Nvidia's flexible chips stay useful as AI designs change, giving them a long, financeable life. The analogy from private equity executives is to airplane financing, where a loan rests on the lasting value of the plane itself, not only on the credit of the airline. For chips, that means Nvidia must standardize designs so Wall Street can bundle many similar loans and sell pieces of them to investors, as it does with home mortgages.

In Sacks's description, Nvidia itself would act as the central bank of AI, working through private banks much as the Federal Reserve works through the banking system. He argues that financing, not customer demand, is now the limit on growth, with Nvidia effectively lending buyers the money for its own products, which is what seller financing means, and extending it as a line of credit. To illustrate the scale, he estimated that entrepreneur Elon Musk's xAI wants to add six to eight gigawatts of power next year, a gigawatt being a unit of electrical power used to measure data center size, at a cost of $300 billion to $400 billion, compared with what he described as roughly $100 billion recently raised.