Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Nvidia is the supremo in creating chips that power AI. Its $5.5tn market capitalisation reflects investors’ belief it will remain this way. The $13bn acquisition of Hugging Face on Thursday significantly increases its chances, at a cost that may be paid by some of its erstwhile comrades.
Hugging Face, while hardly a household name outside of tech, is the go-to repository for open-source and open-weight AI models, the kind users can tweak to suit their own ends as opposed to less customisable models offered by companies like OpenAI and Anthropic. Think of it as an app store for AI or a kind of giant online laboratory.
It’s easy to see why owning that, for what amounts to chump change, benefits Nvidia. First, it can promote usage of its own open-weight Nemotron models. It can also put a huge amount of cash behind Hugging Face to promote open models generally, which tend to run on Nvidia chips.
Closed and so-called frontier models do too, but Google, OpenAI, Anthropic and SpaceX are all making their own “custom silicon”. This isn’t enough to threaten Nvidia yet, but then, it’s CEO Jensen Huang’s job to worry about what happens a decade from now as much as what happens tomorrow.
That should worry the likes of OpenAI and Anthropic, though. Open models are roughly 20-30 per cent cheaper for users than the closed variety, Citigroup researchers reckon. The two leading model makers’ valuations have soared to nearly $1tn each because of their ability to charge users premium prices for “tokens”, the unit of AI workloads.
Nvidia’s backing of open models may thus mean more AI of all types gets used, but at a lower price per token. Companies are already fretting about the cost of running AI, even as they broadly still see the long-term benefits. Uber, which is laying off a tenth of its workforce, is one that has complained out loud about staff “tokenmaxxing”, running up huge bills for their AI experiments.
The oddity in all of this is that Nvidia is itself an investor, backer, supplier and guarantor for the very companies its new deal potentially disrupts. Huang has signed off on equity investments in both OpenAI and Anthropic; it is providing a financial backstop for OpenAI’s giant PORTS-Pike data centre. The model-making duo, plus Google, Amazon and all the rest, remain enthusiastic buyers of Nvidia’s chips.
But maybe that’s not so odd after all: Nvidia’s past investments, and other forms of supposedly circular financing, are a means to a commercial end. These interdependencies facilitate Nvidia’s own profit by securing a future market for its chips, but they don’t seem to hold Huang back from making investments that might put the goodwill of allies at risk. That is, for an Nvidia investor, quite reassuring.
True, if Nvidia’s foray into open models shaves billions off the value of its OpenAI and Anthropic, the chipmaking giant’s own balance sheet will take a hit. But that is presumably a small price to pay for making it more likely that Nvidia is still the world’s biggest company a decade from now. Huang’s main game is making chips, not friends.
