Nvidia pays 1.84 times its rejected offer to own the hub 13 million developers pull from

The Information reported Wednesday night, citing a source familiar with the matter, that Nvidia has agreed to buy Hugging Face for $12.9 billion.

Neither Nvidia nor Hugging Face responded to requests for comment, and Nvidia has moved quickly in the past to address reports it considers inaccurate.

Nvidia invested in Hugging Face's 2023 Series D, which raised $235 million from investors including Google, Amazon and Salesforce, valuing the company at $4.5 billion.

OpenAI, Google, Amazon and Anthropic are all in the process of building their own AI chips to lessen their reliance on Nvidia.

Nvidia has already poured tens of billions of dollars into building its own open source AI models.

Chinese labs including Z.ai and Alibaba's Qwen have shipped open models that are approaching and even outscoring Anthropic's Claude and OpenAI's GPT on open benchmarks.

Chinese labs such as Moonshot AI released systems, including its Kimi K3 model, that matched leading US models on benchmarks while costing a lot less to run.

Washington officials reportedly weighed restrictions on open-weight models.

Nvidia joined Meta and Microsoft in lobbying Washington against rules that would restrict open-weight releases, arguing the government should not throttle open-source AI.

Hugging Face CEO Clem Delangue pointed to a letter signed by Nvidia CEO Jensen Huang and 24 other companies, including Hugging Face, urging the US government to support open models rather than restrict them.

Hugging Face operates the infrastructure of the open ecosystem: a model hub where teams publish weights, a datasets library, the Transformers library that became the default way to load those models, and Spaces, where demos run.

Nvidia's GPUs and CUDA software are the default substrate for AI compute, and Hugging Face is where most models are stored, versioned and downloaded.

In a CNBC interview in late July, Delangue warned that China is "clearly dominating" open source AI.

Hugging Face cofounder Thomas Wolf declined to comment on the rumoured sale to Nvidia but said the company has "always had a lot of interest".

Some Hugging Face users said the deal "screams monopoly" and expressed concern that Nvidia could censor the repository or use it to exert control over developers.

The reported $12.9 billion purchase price is 1.84 times the $7 billion valuation implied by the investment offer Hugging Face rejected late last year.

At $12.9 billion the price equals 13.4 percent of Nvidia's $96.2 billion record quarterly revenue, or roughly twelve days of revenue at that quarterly rate.

The $12.9 billion price is 3.5 percent of the $366 billion in future commitments Nvidia disclosed.

The reported price is 2.87 times the $4.5 billion valuation Hugging Face carried at its 2023 Series D.

Across Hugging Face's stated 13 million users, $12.9 billion is about $992 per user.

Across the more than 3 million models Hugging Face hosts, $12.9 billion is about $4,300 per model.

Business Insider, which first reported over the weekend that Hugging Face was fielding takeover interest, reported Wednesday night that the talks, which would value the company at more than $13 billion, had not yet produced a signed agreement and could still fall apart.

A neutral repository lets any developer pull any model and run it anywhere, while a vendor-owned repository can steer that flow toward the owner's cloud, tooling and accelerators, not by blocking rivals but by making the path of least resistance the one that runs on Nvidia.

Open-weight licences do not change with ownership and a permissively licensed model stays usable off-platform, but the default doorway changes: a startup that today pulls a community checkpoint may afterwards do so inside an Nvidia account, with Nvidia-hosted inference one click away.

Distribution on the hub is effectively free today but could be priced or prioritised later, and a lab publishing there competes with others that Nvidia also supplies and would now partly own.