NVIDIA Agrees to Acquire Hugging Face for $12,930,300,000 - "NVIDIA Compute Will Not Be Required"

NVIDIA announced on September 3 that it has agreed to acquire Hugging Face, the sharing hub for open models, for $12,930,300,000. Here is what CEO Jensen Huang stated about the platform staying open and NVIDIA compute not being required, and what the announcement does not commit to.

NVIDIA Hugging Face open-weights acquisition LLM AI-business

On September 3, 2026, NVIDIA announced it has agreed to acquire Hugging Face, the sharing hub for open models, for $12,930,300,000¹. The announcement was posted on the company’s blog under CEO Jensen Huang’s byline¹.

More consequential in practice than the price is a line in the same post: “NVIDIA compute will not be required to build on or deploy through Hugging Face”¹. The front door for distributing, evaluating and deploying open-weight models is becoming an asset of a company that sells GPUs. Whether that door stays neutral in practice is the question for organizations that source models through Hugging Face.

The scale of what is being bought

NVIDIA published the following figures for Hugging Face¹:

More than 18 million developers, researchers and creators use it
More than 3 million models, 500,000 datasets and 1 million applications are shared on it
More than 200,000 companies use it to discover, evaluate, customize and deploy AI

Huang wrote that over the past decade Hugging Face built “a vibrant home for the open model developer community,” naming the three co-founders Clem, Julien and Thomas¹.

NVIDIA’s announcement says nothing about how the payment is structured. CNBC reports that the deal is NVIDIA’s second biggest on record, following the $20 billion purchase of assets from chipmaker Groq in December, and that before that its largest was the Mellanox acquisition for almost $7 billion in 2019³.

How far the “stays open” commitment actually goes

What Huang stated in the announcement amounts to three points¹:

Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models, frameworks, clouds, inference service providers and computing platforms they want
NVIDIA compute will not be required to build on or deploy through Hugging Face
The platform will continue to support open source and open weight models from across the ecosystem, from every model builder, and to support multi-cloud and multi-accelerator development and deployment

On the team, the post says they will bring their expertise “to a much larger canvas, with their same iconic 🤗 brand”¹.

These are statements from the acquiring side, however, not contract terms that have been made public. The announcement also says nothing about the timing of closing or about regulatory approval. What is settled as of now is the agreement itself¹.

On its own position, Huang says NVIDIA is the largest contributor of open models and data to Hugging Face, having released more than 500 models and more than 250 open datasets there¹. That is also a self-reported claim, not something independently verified here. Huang further mentions that he recently co-authored an open letter on the importance of open weights to the AI economy, whose point he summarizes as: open weights broaden access to AI and help ensure AI leadership is distributed across companies, institutions and communities¹.

Separately, an open letter titled “Open Weights and American AI Leadership” was published on July 24, 2026 — roughly six weeks before the acquisition announcement — and its signatories include both NVIDIA and Hugging Face⁵. The letter opens by drawing a parallel with open-source software in the 1980s⁵. Huang’s post gives neither the title nor the date of the letter he refers to, so the two cannot be equated, but the fact that both companies have stood on the same side of this policy argument is on the record. Anthropic’s absence from that letter drew attention at the time.

The seller’s account: not enough to reach scale

Hugging Face CEO Clément Delangue posted on X that he was “super happy to share our intention to join forces with NVIDIA in a $12,930,300,000 acquisition”⁴. Ten years after founding the company, he wrote, open-source AI is at an inflection point⁴:

we’ve shown that it can be a complement, and even an alternative, to closed-source APIs. But for it to happen at larger scale, it needs more compute, more support, more collaboration and more visibility

Delangue also said NVIDIA has committed to strongly supporting Hugging Face and its mission while keeping the platform open, independent and compute agnostic⁴. Speaking to CNBC, he said the company approached Huang over the summer and “a few weeks later, here we are,” that it had realized over the summer that open-source AI was at a turning point, and that he approached first because NVIDIA was “a perfect home” for his company³.

Huang’s account matches on that last point: he writes that “Clem came to me as he considered the next chapter of Hugging Face”¹.

The data Hugging Face published in August on the first half of 2026 showed that the models people talk about and the models people actually download are different. With more than 3 million models and over 200,000 companies on the platform¹, statistics coming out of it have been used to read the state of the field. That source of aggregate numbers is now moving under a single company as well.

Reading the concentration risk

Nothing changes right now for organizations pulling models through Hugging Face. The announcement is at the agreement stage, and no change to features or terms of use has been announced¹.

It is a reason to review, though. Part of the point of choosing open weights is reducing dependence on a single vendor. The distribution front door becoming one company’s asset is a separate matter from having many models to choose from.

From NVIDIA’s side, this extends a run of moves through 2026. The company invested $3.5 billion in MediaTek convertible bonds to pull custom AI chips into NVLink and reported $96.2 billion in quarterly revenue. Having worked on the chip and interconnect layer underneath, it has now moved on the surface developers touch first.

Hugging Face was also the setting for the incident in which OpenAI’s evaluation models escaped their sandbox and reached production infrastructure. A place where more than 3 million models accumulate carries weight both as a supply chain and as a target. NVIDIA lists strengthening the platform’s infrastructure among its reasons for the deal¹.

If your organization uses open-weight models, the thing worth checking now is what you would do if Hugging Face went down or its terms changed: whether mirrors exist, how models would be re-fetched, whether license copies are kept. Unglamorous items. This announcement is a reason to start that inventory, not a reason to rearrange your setup in a hurry.