Huang rejects circular financing charge, says NVIDIA's customer stakes are small relative to demand

The materiality claim can be checked in one direction. Add the two equity cheques NVIDIA has disclosed into its own customer base and you get $32bn ^[14], and set that against S&P Global's estimate that the five largest hyperscalers will spend an additional $5.3 trillion of capex through 2030 ^[12]: NVIDIA's own money is about six-tenths of one per cent of the spending it is accused of manufacturing ^[15]. On those two numbers Huang is right.

The revenue side does not check. NVIDIA's quarterly filings put three direct customers at 16, 15 and 13 per cent of first-half fiscal 2027 revenue ^[9], so 44 per cent of the top line sits with three counterparties the company does not name ^[13]. The disclosure does not say whether any of the three has taken NVIDIA equity, and it does not separate an order placed with NVIDIA's money from one placed with a customer's own. Huang's test, small stakes against large demand, is one his filings do not let an outsider run.

The $500bn does less work than its size suggests. The August platform, built with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, is a vehicle for third-party capital, and the figure is a target for raising financing, not a sum NVIDIA has committed ^[6]. What NVIDIA itself has at risk is the $32bn, and $30bn of that is a 27 per cent slice of a round in which Amazon's $50bn was the bigger cheque ^[16].

The IMF's April report said companies across the AI value chain are increasingly using circular financing and that AI investments could struggle in a downturn, while describing the financial stability impact as insignificant at the current stage ^[10]. The BIS went at the funding side on September 10, pointing to heavier use of debt and private credit to pay for AI capital outlays and warning that misaligned returns would make the boom end in a much bigger financial crisis ^[11].

"In AI, compute is revenue," Huang said ^[7]. That moves the question from whether a chip was sold to whether the chip earns, since he treats GPUs as productive assets throwing off regular compute revenues instead of one-off hardware purchases ^[8]. It is a question about utilisation and price, and the material here carries neither a utilisation figure nor a rental rate.

The reading I would defend is narrow: NVIDIA has won the balance-sheet argument and has not engaged the disclosure one, and the composition of that 44 per cent is the only number that would settle either. The counter-thesis is decent. If the three unnamed customers are hyperscalers reselling compute to enterprises paying cash, then the stakes are venture spend and the concentration is ordinary customer risk. What would break my reading is a filing that names the three and shows none of them has taken NVIDIA money, or OpenAI buying capacity out of its own revenue instead of out of the $110bn it has just raised ^[5].