Hugging Face acquisition explained “A man making $120BN a year selling compute decides to buy a company that helps make compute more cost-effective so he can sell more compute. If end users have $1TN to spend on tokens, Nvidia would prefer that money flow through open-source people at 30%... gross margins, rather than 70% gross margins at OpenAI or Anthropic. Open source is good for compute salespeople. If you are selling GPUs, you want everyone else’s margin to be lower so yours can be higher.” Rory O'Driscoll Love to hear your thoughts Vipul Ved Prakash Naveen Rao Bindu Reddy George Hotz 🌑show more

Started casually listening to series a year ago and now its a weekly must listen. Love the analysis and discussion Rory and Jason bring every episode. Thank you!

@rodriscoll @jasonlk are incredible.

That and I'm sure nVidia is reacting to the major models developing their own silicon and using alternative AI chip platforms as well.

commoditize your complement, infra edition. every point of margin the model layer loses to open source becomes gpu demand. nvidia doesn't need to own the models, it needs them cheap and everywhere. hf is the cheapest way to keep it that way

Thanks for sharing this

this is also the price justification. ~13B on ~$150M ARR (~86X) only makes sense if youre not valuing it as a pure software business. Its a distribution channel for margin compression on your competitors

The flywheel is pretty clear: make open models easier and cheaper to run, expand usage, and ultimately create more demand for compute.

yeah that tracks

And that’s exactly why this strategy is so powerful. 😂 Nvidia doesn’t need to own the biggest model.,.it can make the entire AI ecosystem hungrier for compute. Lower inference costs… more adoption…more workloads… more GPUs. That’s the flywheel.

It seems selling the shovels in the AI gold rush was not enough. 🪏 Now NVIDIA has bought influence over which models become tomorrow’s enterprise standard!

Is the real play here making compute cheaper enough that overall demand for it grows even faster?

I like this acquisition, adds more horizontal capture for nvidia #dyor

The mechanic worth watching is who owns the discount. If Nvidia routes token spend through its own cheaper-compute subsidiary, every other AI infra buyer is underwriting a competitor's cost curve, not the market's. Different question than the one about Hugging Face alone.

this also makes nvidia less dependent on any one closed model company. open-source demand still becomes compute demand, and now they own part of the distribution layer too

The bet only works if open weights actually increase total inference volume instead of just redistributing it.

NVIDIA likes open models because they commoditize the layer above compute. But $12.93B buys more than cheaper tokens: it buys the place 18M developers decide what to run. The moat isn’t margin compression, Harry. It’s owning demand formation before the GPU gets chosen.

Free market for the win 🇺🇸

The strategic beauty is that Nvidia doesn’t necessarily need to own the highest-margin application layer. It benefits when thousands of companies can afford to build, train and run more models. Sometimes the best position in a gold rush is making sure everyone can afford to keep digging.

Nvidia doesn’t need inference to stay expensive. It needs inference to get cheap enough that almost everything becomes worth running AI on.

so nvidia buys the shovel factory to sell more shovels during the gold rush. classic move.

Nvidia may have bought itself a very expensive headache. HF doesn't just bring distribution. It brings copyright exposure, malicious-model risk and an enormous software supply-chain attack surface. And the irony? Nvidia probably would've sold the compute either way.

thats a pretty funny way to put it and also probably exactly whats happening

As Bezos said - someone’s margin is someone else’s opportunity

That’s business 101. Super obvious why Jensen did it. Let’s he makes open source #1 🙌🏽