Nvidia reportedly agrees to buy Hugging Face for $12.9B: the open-model commons gets a chip-vendor owner

Nvidia reportedly agrees to buy Hugging Face for $12.9B: the open-model commons gets a chip-vendor owner

Nvidia has reportedly agreed to buy Hugging Face for $12.9 billion, and the entire "agreed" rests on one anonymous source at The Information, relayed by Ars Technica and The Verge 1 2. Neither company has confirmed. The number is still stark against the last valuation on the public record: $4.5 billion, set in August 2023, which makes the reported price about 2.9 times that figure, a 187% premium arriving three years to the week after the round 3.

Hugging Face is the AI startup behind a GitHub-like hub where developers find, download, fine-tune, and re-upload open models, and Ars Technica calls it "the de facto place to store, download, or work on open-weight models" 3 1. The hub already hosted over 1 million repositories as of August 2023. On the last headcount and funding figures the company disclosed publicly, also from August 2023, $12.9 billion works out to roughly $76 million per employee, about 33 times the $395.2 million Hugging Face had raised to that point 3. The company was founded in 2016 and is reportedly not yet profitable 1.

What is actually agreed, and what is only talks

The deal's certainty depends entirely on which outlet you read. The register, in one place:

• The Information: Nvidia has agreed to acquire Hugging Face for $12.9 billion, citing one person with knowledge of the matter. ProvenBrief has not opened the original; we read it through Ars Technica's and The Verge's attribution lines 1 2.
• Business Insider, first: the two companies were in talks but had not yet reached a deal, also citing a single anonymous source, published almost an hour before The Information's story 2.
• CNBC: a separate source confirmed talks are underway 1.
• Both companies: no confirmation. Ars Technica notes the deal is not finalized and could still run into trouble ahead 1.

One thread from the 2023 files ties the register together: Salesforce, which Ars Technica reports was also trying to buy Hugging Face, participated in that August 2023 round, the same $235 million raise whose $4.5 billion valuation Salesforce CEO Marc Benioff seemingly verified on X at the time 1 3. The reported rival bidder was inside the last valuation round.

Why a GPU vendor wants the model commons

Ars Technica states the strategic frame in one sentence: OpenAI, Anthropic, and other frontier labs have begun investing in their own specialized hardware, "seeking vertical integration and freedom from Nvidia's leverage" 1. The escape is real at the silicon layer: custom chips can route around the GPU tax.

Our read is that the reported purchase aims one layer up. Labs that build their own silicon still pull models, datasets, and fine-tuning tooling through one commons, and if the chip vendor owns the commons, the leverage everyone is escaping at the hardware layer reappears at the library layer. Nvidia already ships its own open models, the Nemotron family, and its earlier plan for its own cloud AI business stalled 1. Owning the hub that open-model workflows route through would be a way to re-enter the services business the cloud providers won, without winning a cloud war. The Verge's item on the deal raises the wider optics, noting Nvidia recently debuted "compute as an asset class" amid circular-financing criticism 2. That is context, not a confirmed motive: one anonymous source is all the motive anyone has on record.

Robotics is the second layer

The robotics stack got carved into layers in the same week this story broke. Hugging Face's own LeRobot project hosts "state-of-the-art machine learning for real-world robotics," with models, datasets, and tools in PyTorch on a Hugging Face org page; that is the vendor describing its own product 4. Ars Technica notes Hugging Face has increasingly invested in exactly the model types used in robotics and other physical AI applications, a space where Nvidia is already one of the biggest players 1. And on August 27, Anthropic opened a research preview of what it calls the Model Hardware Standard, "a shared specification for AI agents to safely operate physical devices" 5.

A chip vendor that owns the robotics model commons, while rivals define the interface between AI agents and physical machines, would sit across the robotics stack the way it sits across GPUs today. That connection is ours; the sources state the pieces separately.

The asymmetry is the story. Labs can dual-source silicon, custom or merchant. They cannot dual-source the commons: there is one Hugging Face, and per one anonymous source, the GPU vendor just agreed to buy it. What $12.9 billion would buy, if this closes, is the neutral ground every escape route runs through.