Nvidia backs up to $105bn in AI data-centre financing as custom chips threaten demand
- Nvidia agreed in mid-August to backstop up to $105bn for an Ohio data centre that will use its chips, and separately outlined a plan with six Wall Street firms to mobilise more than $500bn for AI infrastructure by guaranteeing equipment values.
- Over the past three years, Nvidia has pledged more than $70bn for startup investments and provided $300bn in financial support to customers, including guarantees and agreements to top up customer income when targets are missed.
- Amazon, Google, Meta and Microsoft generate roughly half of Nvidia's revenue and are expected to spend about $800bn on AI infrastructure this year, while also designing chips that cost one-fifth to one-third as much as Nvidia's.
- Bloomberg Intelligence expects custom AI chips to reach about 50% of the AI-processor market by the end of the decade, up from roughly 40% this year, as Google and Amazon begin selling their chips outside their own operations.
- Nvidia made about 90 startup investments last year, nearly twice its count two years earlier, and had already agreed more than 60 further investments this year to fund customers and expand demand for its chips.
Hacker News opinions
Calling Nvidia a central bank is generous. It cannot expand chip supply at will like a normal central bank, and it does not really control interest rates.
I think TSMC is the currency press operator in this analogy. It is the manufacturer that makes the chips Nvidia's whole financing machine depends on.
I'm worried Nvidia will eventually treat gaming as an afterthought. It already removed standalone gaming revenue from its reports, and I do not see AMD or Intel easily filling a sudden gap.
I disagree that AMD cannot step in. AMD already makes console SoCs, and much of the Steam user base runs older hardware anyway.
Nvidia will keep shipping gaming products. Blackwell GPUs, RTX Spark laptop APUs, and the Nvidia-powered Switch 2 show that, and consumer hardware remains part of CUDA's value.
The financing is temporary. Private credit needs returns eventually, and AI overinvestment is likely to produce a capital-cycle downturn.
If Nvidia is a bank, it looks more like a profit-and-risk-sharing lender than one earning conventional interest. A broad AI failure could leave it holding compute and assets from bankrupt neoclouds.
I am seeing too many signals like this to ignore. The obvious question is how many warnings investors need before they reduce exposure.