Analysis
Nvidia confirmed it has agreed to acquire Hugging Face for approximately $12.93 billion, formally closing a deal that had been reported as unsigned for more than a week, TechCrunch reported. Pulse first covered the pending deal on Aug. 27, when The Information reported Nvidia had agreed to buy Hugging Face but cautioned the agreement had "not been formally signed and could still change." That caveat is now resolved: Nvidia CEO Jensen Huang and Hugging Face co-founder and CEO Clement Delangue both issued on-the-record statements confirming the transaction, Nvidia's largest outright acquisition.
What Changed Since August 27
The deal size held steady at roughly $12.9 billion, but the confirmation adds two things the earlier reporting didn't have: an explicit neutrality commitment and named executive quotes. Huang said publicly that "Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. Nvidia compute will not be required to build on or deploy through Hugging Face." Delangue framed the deal as access to "more compute, more support, more collaboration, and more visibility" to keep scaling a platform that now hosts more than 3 million models, 500,000 datasets and 1 million applications for a developer base Hugging Face puts above 18 million.
“- Platform scale -- 3 million-plus hosted models, 500,000-plus datasets, 18 million-plus developers, 1 million-plus applications (Spaces).”
Company Background
Hugging Face was founded in 2016 by Delangue, Julien Chaumond and Thomas Wolf, originally as a consumer chatbot app, before pivoting into what became the default hosting and distribution layer for open-weight AI models -- effectively GitHub for machine learning. It last raised money in a $235 million Series D in August 2023 at a $4.5 billion valuation, from an investor list that reads like a who's-who of companies that didn't want a rival to own it outright: Google, Amazon, Nvidia, Intel, Salesforce, AMD, Qualcomm and IBM. Nvidia was already on the cap table before this acquisition -- a minority investor becoming an owner is a different kind of transaction than an outside buyer coming in cold, and probably explains why Nvidia moved to formal terms faster than a typical strategic acquisition process would allow.
The most striking data point in the deal's own history: Hugging Face reportedly turned down a $500 million acquisition offer from Nvidia in an earlier round of talks, according to the Financial Times. Whatever the exact timing of that rejected offer, the gap between it and the $12.9 billion final price -- roughly 25 times higher -- is a rough proxy for how much scarcity value the open-model distribution layer has accumulated as frontier labs, enterprises and individual developers converged on one place to publish and pull models.
Why Nvidia Wants the Distribution Layer, Not Just the Chips
Nvidia already owns the silicon layer through its GPUs and the software layer through CUDA. What it has never owned is the place developers actually go to find, fine-tune and deploy a model. Nearly every major open-weight release -- Meta's Llama family, Mistral, Alibaba's Qwen, Google's Gemma -- gets published to Hugging Face on day one regardless of which cloud or chip vendor a developer ultimately runs it on. That neutrality has been the platform's core value proposition, and it's exactly what Huang's public commitment is trying to preserve on paper, because the moment Hugging Face looks like an Nvidia sales funnel, competing labs and cloud providers have a reason to build or back an alternative registry.
That threat isn't hypothetical. AWS, Google Cloud and Microsoft Azure all run their own model registries and marketplaces, and Databricks -- itself freshly valued at $190 billion after a $5 billion raise this August, per Crunchbase's August funding tally -- has pushed its own open-source ecosystem (MLflow, Unity Catalog) as an alternative home for enterprise AI assets. If Hugging Face starts to look like Nvidia's front door rather than neutral ground, those competitors gain a straightforward pitch: use us instead, and don't hand your distribution layer to your GPU supplier.
Numbers in Context, and What the Confirmation Doesn't Resolve
- Deal value -- $12.9 billion, Nvidia's largest outright acquisition (its ~$20B Groq deal in December 2025 was a license-plus-talent agreement).
- Hugging Face's last priced valuation -- $4.5 billion in August 2023, meaning this transaction values the company at roughly 2.9x its last funding-round mark three years later.
- Platform scale -- 3 million-plus hosted models, 500,000-plus datasets, 18 million-plus developers, 1 million-plus applications (Spaces).
- Rejected earlier offer -- $500 million, per the Financial Times, a fraction of the eventual price.
What the "Nvidia locks up open-source AI" framing misses: Hugging Face's core repository infrastructure is largely open-source itself, and the models hosted on it belong to the labs that trained them, not to Hugging Face -- Nvidia can't unilaterally wall off Llama or Qwen weights that Meta and Alibaba choose to keep open regardless of who owns the hosting layer. The real leverage Nvidia gains is over discovery, developer experience and, longer-term, whichever premium services Hugging Face builds around inference and fine-tuning -- not over the open licenses themselves. Regulatory review is also an open question the confirmation doesn't resolve: neither company disclosed a closing timeline or addressed antitrust review, and a chipmaker acquiring the industry's default open-model distribution point is a more legible antitrust target than most AI infrastructure deals of the past year.
Worth watching from here: whether Hugging Face's Spaces and Inference Endpoints products get pushed toward Nvidia-only backends over time regardless of today's pledge, and whether AWS, Google or Databricks respond by materially investing in a competing open registry rather than just their existing marketplaces.
Update (September 4, 2026): Pulse has follow-up coverage — Thinking Machines Nears $1B Round at $40B.