Why Nvidia paid $12.93 billion for Hugging Face, the home of open AI models

The Verge’s Jess Weatherbed reports that Nvidia has agreed to acquire Hugging Face for $12.93 billion. Founded in 2016, Hugging Face is the library where much of the AI world browses and downloads open-source models, datasets and tools, which is why it is often called the GitHub of AI. Its last official valuation was $4.5 billion, set in a 2023 round in which Nvidia itself invested.

Two other figures frame the price. The Financial Times reported that Hugging Face rejected a $500 million investment from Nvidia last year that would have valued it at $7 billion, citing worries about one investor becoming dominant. The Information estimates recent annualized revenue at about $150 million. Ars Technica, working from the FT, describes it as Nvidia’s largest acquisition, bigger than Mellanox, with a close targeted by 2027 (our write-up of that version).

Nearly tripling the 2023 valuation for a business with that revenue only makes sense if Nvidia is buying position rather than profit. Weatherbed’s framing is the useful one: closed AI labs are working on their own chips, and Nvidia’s answer is to own the place where open-source developers already spend their time. Whoever runs the default hub for open weights influences what gets built, how it is packaged and what hardware it is tuned for first.

Seen that way, the $150 million revenue number is a distraction. The asset is the index of open models and the community habits around it. If the next wave of models, agents and evaluations keeps defaulting to Hugging Face, and Hugging Face keeps defaulting to tooling that works best on Nvidia, Nvidia’s software moat extends well beyond its own products.

The rejected investment is the telling detail. A year ago Hugging Face did not want Nvidia as a dominant shareholder at $7 billion. Now it accepts Nvidia as its owner at $12.93 billion. That is less a change of heart about independence than a recognition of what scale costs. CEO Clément Delangue’s public message emphasized more compute, more support and more visibility, which are real benefits that come at the price of no longer being neutral ground.

Huang’s pledge that the platform will stay open, and that Nvidia compute will not be required, is therefore the core of the deal rather than a footnote. The first test will be product surfaces like Spaces, Inference Endpoints and search ranking. If they start to look like an Nvidia storefront, the developers who made the hub valuable can move their models and their attention elsewhere far more easily than they can leave Nvidia GPUs. WIRED’s angle, covered in Nvidia’s $12.9 billion bet on open-source AI, focuses on how this fits Nvidia’s own open-model work.

Talk of the deal leaked in late August, with Business Insider describing a roughly $13 billion process and The Information reporting an agreed $12.9 billion figure before Nvidia confirmed it. Nvidia’s own announcement, which we analyzed in NVIDIA to Acquire Hugging Face, put the price at exactly $12,930,300,000 and repeated the neutrality promise. See all of our Nvidia coverage for the wider pattern of the company buying and funding its way up the AI stack.