Korva is a scroll of the internet’s best ideas.
Nvidia lines up $500 billion for compute
Jensen Huang laid out the plan in an article on X titled AI Factory Compute Is Becoming an Investable Asset Class. The firms will set up independent financing platforms that raise third-party capital.
A company that wants compute borrows money, buys Nvidia-based systems, and rents out the compute to generate revenue. It repays the loan from those cash flows. Nvidia's role is to connect customers who need compute with lenders.
Nvidia has also offered the lenders a residual value guarantee on deals brought to it. After three or four years, it guarantees that the GPUs can be rented at a certain rate.
Why would anyone guarantee what chips are worth later?
A residual value guarantee is a promise about an asset's worth at the end of a lease or loan term. Lenders worry about the price of used equipment, so a guarantor who absorbs that risk makes the loan safer. The guarantor typically charges for taking on the risk, often through fees or a share of upside. Residual values are central to how auto leases are priced.
How do financing platforms keep assets separate?
Platforms usually house the equipment in a special purpose vehicle, a legal entity that owns only the assets and the debt tied to them. If the borrower's parent company runs into trouble, lenders still have a claim on the equipment and its rental income. Investors such as pension funds can then buy into the vehicle without taking on the parent's wider business risks. Special purpose vehicles have been used in project finance for decades.
How does GPU rental produce loan repayments?
Providers bill customers by the GPU-hour, so each chip's revenue equals its hourly rate times the hours it is rented. Longer reserved contracts give lenders predictable income, while on-demand rentals can bring in more per hour with less certainty. Utilization, the share of hours a chip is rented, is the key figure in these models.
More from All-In with Chamath, Jason, Sacks & Friedberg
