Korva is a scroll of the internet’s best ideas.

The backstop deals

Nvidia provides a 25% backstop, takes equity in neoclouds, and guarantees it will buy their compute through 2030. The purpose is to give those companies a lower cost of capital so they can buy more chips. They get that lower cost of capital because Nvidia assumed the risk.

Risk on the balance sheet

In one world, AI takes off and never stops, and Nvidia captures the upside of its risk. In another, a neocloud backs up, a ton of compute comes to market, and the hyperscalers already have plenty. Nvidia is then paying for compute that no one wants and has lost a bunch of money. The expected value of that investment is somewhere between zero and 100%, and the part below 100% reduces Nvidia's profitability. Viewed holistically, it is a price cut.

The hyperscaler capital fight

Nvidia's biggest competitors are the hyperscalers. Google already made a deal to sell about 20% of its TPUs to Anthropic, and Amazon is expected to sell Trainium externally. They sell their chips as commodities rather than on differentiation, so they don't weaken the pull of their own clouds. Their real edge is scale and a lower cost of capital than the neoclouds, which makes it a capital fight.

How does a lower cost of capital sell more chips?

Buying chips is usually financed with debt, and lenders set the interest rate by how risky they think the borrower is. A backstop from a large supplier shifts some of that risk, so lenders ask for less. Cheaper financing makes more purchases pay for themselves, and the customer buys more.

Why can hyperscalers borrow more cheaply than neoclouds?

Credit rating agencies score borrowers on cash flow, existing debt, and balance sheet strength. The largest tech companies generate huge cash flows from established businesses, which puts them at the favorable end of those scales. A younger company that only rents out chips has a thinner record, so lenders charge it more.

How has vendor financing gone wrong before?

In the late 1990s, telecom equipment makers such as Lucent lent money to customers so they could buy network gear. When the telecom bubble burst in 2000 and 2001, many of those customers failed. Their suppliers were left holding loans that were never repaid.

More from Invest Like the Best with Patrick O'Shaughnessy

Business & StartupsApple ships deterministic productsAI is probabilistic, and Apple is the king of deterministic products: ship an iPhone once and it has to be good.
Enjoyed it?