NVIDIA to acquire Hugging Face for $12.93 billion

On September 3, 2026, NVIDIA CEO Jensen Huang wrote that NVIDIA had agreed to acquire Hugging Face for $12,930,300,000. The NVIDIA blog post is the primary document: more than 18 million developers, researchers, and creators use the hub to share more than 3 million models, 500,000 datasets, and 1 million applications, and more than 200,000 companies use it to discover, evaluate, customize, and deploy AI. Huang named Hugging Face co-founders Clément Delangue, Julien Chaumond, and Thomas Wolf. The New York Times reported that Delangue said the companies signed on Wednesday, September 2, and that Huang told CNBC other bidders had also tried to buy the company.

This briefing is for model builders who ship on the Hub, platform teams that pull weights and datasets from it, and anyone who has to explain a hardware-neutral promise next to a chip vendor's check. It covers the price and structure, what NVIDIA says will not change, why the deal exists now, who already sits on the cap table, and what is still unsigned — close timing and regulators.

The deal

Reuters broke the cash split the same day: about $11.9 billion to Hugging Face investors and an equity-based retention program of up to $1 billion for employees joining NVIDIA. The Associated Press attributed a $1 billion retention plan to Huang on CNBC. Forbes, citing an SEC filing, said NVIDIA expects the deal to close in the first half of 2027. Those three numbers — $12.93 billion headline, ~$11.9 billion to investors, up to $1 billion retention, H1 2027 close — are the ones that appear in the announcement-day record. Do not treat a rounded $13 billion headline as a different price.

Huang's post is also the source for the independence language that is actually in writing. Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models, frameworks, clouds, inference providers, and computing platforms they want. NVIDIA compute will not be required to build on or deploy through Hugging Face. The hub will keep hosting open-source and open-weight models from every model builder and will keep supporting multi-cloud and multi-accelerator work. That is hardware-neutral as a policy statement, not a completed merger filing. Delangue told CNBC, as GamesIndustry.biz and other outlets quoted, that Hugging Face approached NVIDIA because it needed more resources, more scale, and more visibility, and that NVIDIA was a perfect home.

Why this round now

NVIDIA is already the largest contributor of open models and data on the Hub, Huang wrote: more than 500 models and more than 250 open datasets. The acquisition puts that distribution channel inside the company that already fills it. TechCrunch's same-day piece is blunt about the industrial logic: an open ecosystem NVIDIA controls is useful if you sell the dominant training and inference GPUs, and unused capacity can be packaged with Hugging Face's enterprise surface. The Financial Times, via Ars Technica's write-up, called it NVIDIA's largest outright company purchase, ahead of the $6.9 billion Mellanox deal in 2020, and said Hugging Face had last year turned down a large NVIDIA investment at a $7 billion valuation to stay independent.

Open models are the product NVIDIA is betting customers will keep choosing against closed APIs. Huang's quoted line in the Wired and NVIDIA materials is that open models let startups, businesses, universities, and public institutions build without training every model from scratch, and that AI advances faster when people can build together. That is the strategic sentence. The timing is the rest of 2026's open-weight race — not a new Hub feature launch.

What the money is for

Huang said NVIDIA will scale Hugging Face's platform, strengthen its infrastructure, and expand access for developers and institutions. NVIDIA's infrastructure, engineering, and global reach are supposed to improve reliability, safety, model evaluation, inference, and deployment while preserving the open ecosystem. Those are the buckets in the blog post. There is no public split of the $12.93 billion across those buckets, and no published headcount or office-closure number. The retention pool is the only earmark that reporters attached a dollar figure to.

For a Hub user the practical spend is on uptime, evals, and deploy paths — not on a mandate to buy NVIDIA GPUs. If that promise holds, a team on AMD, Intel, a hyperscaler TPU, or a third-party inference API should keep the same upload and download loop. If it does not, the first place it will show is default hardware in Hub deploy UIs, not in the blog post.

Competitive context

Hugging Face had raised about $395 million to $400 million by late 2025 (TechCrunch/Crunchbase and PitchBook via Wired). A 2023 round valued it at $4.5 billion, Forbes wrote, with backers that included NVIDIA, Salesforce, Google, and Amazon. SiliconANGLE noted that AMD, Intel, and Qualcomm were in the latest round, and that Salesforce had reportedly looked at buying the company before NVIDIA moved. Some of the $12.93 billion therefore lands with NVIDIA's chip competitors. That is a feature of the cap table, not a rumor.

Closed-model vendors — OpenAI, Anthropic, Google's Gemini stack — are the other side of Huang's bet. NVIDIA is not buying a rival foundation-model lab. It is buying the GitHub-like index those labs and every open-weight lab already publish into. Control of that index is leverage over which models get tried first. Hugging Face also still runs Transformers, Diffusers, Datasets, and enterprise Hub features; none of those product lines got a separate price in the announcement.

Open questions

Close is targeted for the first half of 2027 and will draw competition review. The Financial Times said as much on announcement day. Until then Hugging Face is a signed deal, not a subsidiary. Watch whether Hub deploy defaults, model cards, or inference widgets start ranking NVIDIA runtimes first, and whether multi-accelerator CI stays first-class. Those are verifiable product changes. A blog sentence that NVIDIA compute is not required is not a regulator's remedy.

Builders should keep cloning and pinning model revisions they actually run, and should keep a non-Hub mirror of anything that is load-bearing, the same way they would for a GitHub org they do not control. Confirm the $12,930,300,000 figure and the hardware-neutral paragraph on blogs.nvidia.com, not on a recap. If you are an employee in the retention pool, the $1 billion program is the document to ask for — it is not in the public blog. If you are an AMD, Intel, or cloud customer of the Hub, the test is whether your current workflows still work on the day the deal closes, not whether the press release was polite.