NVIDIA Buys Hugging Face for $12.9 Billion

NVIDIA's official announcement · NVIDIA's official website

NVIDIA had long been one of the library's biggest donors. Huang wrote that NVIDIA is the platform's single largest contributor, having uploaded more than 500 models and more than 250 open datasets. This time, that donor bought the entire building.

A year ago, NVIDIA's $500 million offer was turned down

Hugging Face rejected a $500 million investment offer from NVIDIA last year. According to the Financial Times, that offer valued the company at $7 billion, and Hugging Face turned it down because it didn't want a dominant investor who could sway its decisions.

CEO Clem Delangue himself explained why the answer changed within a year. Appearing on CNBC's Squawk Box on the 3rd, he said he approached Huang first over the summer, and the deal came together within weeks. "Over the summer, we realized that Hugging Face and open-source AI as a whole were at an inflection point, and that we needed more resources, more scale, and a bigger presence," he said. He added that he saw NVIDIA as the perfect home for the company, which is why he reached out first.

The numbers also made the decision easier. The Information reported that Hugging Face's annualized revenue is around $150 million, up from roughly $100 million two months earlier. Its last funding round in 2023 valued the company at $4.5 billion, meaning this deal prices it at nearly three times that.

A chipmaker just moved up a layer

This is NVIDIA's second-largest acquisition on record. The largest was its $20 billion purchase of Groq's assets last December, and before that, the biggest deal had been its roughly $7 billion acquisition of Israel's Mellanox in 2019.

For a chip company, buying a developer platform means moving up a layer above its own product. Hugging Face already offers a feature that runs models on borrowed compute, and TechCrunch noted this could give NVIDIA a way back into a cloud business it scaled down about a year ago — without having to rebuild it from scratch.

Hugging Face was founded in 2016. According to Crunchbase, it has raised more than $395 million to date. Its $235 million round in 2023 was led by Salesforce Ventures, with Google, Amazon, IBM, and NVIDIA all participating. This deal marks a case where a former investor has become the owner.

The openness pledge was written into the deal

Huang wrote that Hugging Face will remain an open platform for the entire AI ecosystem. He included a specific commitment: developers will keep choosing whichever models, frameworks, cloud providers, inference services, and compute platforms they want, and NVIDIA will not require its own compute to be used for building or deploying on Hugging Face.

NVIDIA said it will continue supporting multiple clouds and multiple accelerators together, rather than favoring one. The platform will keep accepting open-source and open-weight models from all creators, not just a single company's models, and it will keep its familiar logo — the two hands framing a face.

That pledge doesn't sound like empty talk, and a recent example explains why. Huang noted that far more people defend systems than attack them, so when a community collaborates transparently around open models, the defenders end up with an asymmetric advantage.

Editor's take

This deal leaked before it was confirmed. After The Information first reported the agreement, rumors circulated that the founders were hesitating and that the revenue multiple looked too steep — but the official announcement came about a week later, at the same price the rumors had already named. So the question worth revisiting isn't the price. It's why this deal came together now.

For NVIDIA, the open-model camp is effectively its own team. Closed frontier labs like OpenAI, Google, Amazon, and Anthropic are all building their own chips. If those labs end up dominating the market, there's less reason to buy NVIDIA hardware. But if more companies adopt open models instead, NVIDIA's customer base for its own chips grows thicker. NVIDIA recently disclosed in an earnings call that it has put more than $50 billion into frontier AI labs — and now it directly owns the channel through which those models circulate.

That's what separates this openness pledge from a goodwill gesture. NVIDIA isn't keeping Hugging Face open out of altruism — it's keeping it open because openness is the structure in which NVIDIA wins. That actually makes the pledge more credible. Promises made out of goodwill tend to bend when circumstances change; promises aligned with self-interest tend to last.

For teams working with this in practice, there's one thing to check. If your pipeline pulls models and datasets from the Hugging Face Hub on the fly, that means your whole workflow rests on a single point. A change of ownership is a good moment to review mirroring and version pinning — not because you doubt the openness pledge, but because that's basic hygiene for any hub. Teams that already keep a record of the licenses and provenance of the models they use for training or production shouldn't see any disruption.

The landlord of the open-model library has changed. What tenants should check isn't the rent — it's the wording of the lease, and that wording is already public. The line promising not to require NVIDIA compute is the standard by which this acquisition will be judged going forward.