NVIDIA to Acquire Hugging Face for $12.93 Billion

NVIDIA is acquiring open-source AI hub Hugging Face. CEO Jensen Huang announced the deal himself on September 3, putting the price at $12.9303 billion. The deal closes a loop that began just ten days earlier, when Hugging Face was reported to be exploring a sale at a valuation above $13 billion.

Hugging Face runs the Hub, a platform where developers upload and download AI models, datasets and demo apps — essentially an AI-flavored GitHub. Model weights, datasets and no-code “Spaces” for trying a model in the browser mostly flow for free; the company monetizes through enterprise subscriptions, paid GPU compute and hosted inference endpoints. Its revenue has never been disclosed, but annualized recurring revenue is estimated at $100-150 million, and Delangue recently said on a podcast the company is “close to profitability.” At $12.9 billion, the deal values Hugging Face at 86 to 129 times that estimated revenue.

Hugging Face was founded in 2016 in New York by three French entrepreneurs — Clément Delangue, Julien Chaumond and Thomas Wolf — originally as a chatbot app for teenagers. When Google open-sourced its BERT language model in 2018, Wolf’s team rebuilt it in PyTorch and posted the code publicly; the reaction dwarfed anything the chatbot had generated. That code grew into the Transformers library, and by 2019 the company had dropped its consumer app to build infrastructure instead.

According to Huang, more than 18 million developers, researchers and creators now use Hugging Face to share more than 3 million models, 500,000 datasets and 1 million applications, with more than 200,000 companies using the platform to discover, evaluate and deploy AI.

Huang stressed that Hugging Face will remain an open platform after the acquisition. Developers will keep choosing whatever models, frameworks, clouds, inference providers and compute platforms they want, and NVIDIA compute won’t be required to build on or deploy through the platform. He added that support for open source and open-weight models, and for multi-cloud, multi-accelerator development, will continue unchanged.

This isn’t NVIDIA’s first tie to Hugging Face. It took a small stake in Hugging Face’s 2023 Series D alongside Google, Amazon, IBM, Intel, AMD and Qualcomm, announced a DGX Cloud partnership at SIGGRAPH that same year, and expanded the relationship in 2025 with Training Cluster as a Service. NVIDIA’s Nemotron open-model family is published directly on the Hub, and Hugging Face’s own analysis named Nemotron 3 Ultra the largest American open model released in 2026.

Yet late last year, when NVIDIA offered to invest $500 million at a $7 billion valuation, Hugging Face turned it down — reportedly to avoid letting any single investor sway its decisions. Delangue has long called Hugging Face the “Switzerland of AI” for its neutrality across clouds and chips. As we reported last month when sale talk first surfaced, that neutrality was always going to be the central tension in any deal: the strategic investors in Hugging Face’s 2023 round — Google, Amazon, NVIDIA, AMD, Intel, IBM and Qualcomm — already represent every layer of AI infrastructure, so any one of them taking full ownership undermines the premise itself. Less than a year later, Hugging Face sold the whole company to one of those very shareholders.

What makes this deal notable is that NVIDIA has spent the past eight months doing the opposite three separate times. It paid $900 million for chip designer Enfabrica last September, then in December signed a non-exclusive $20 billion technology license with AI inference chip maker Groq, absorbing key talent including founder Jonathan Ross. “We’re bringing on talented people and licensing Groq’s IP — we are not acquiring the company,” Huang said at the time. That acqui-hire structure — capturing technology, talent and equity upside without a formal acquisition — drew scrutiny for sidestepping merger review, prompting an inquiry from Senators Elizabeth Warren and Richard Blumenthal. Last month, reports surfaced that NVIDIA was pursuing a similar $7 billion arrangement with AI coding startup Poolside, and around the same time it was confirmed to be discussing partnership, investment or an outright buyout with Korean chipmaker Rebellions. The Hugging Face deal breaks from that pattern — it’s a full, 100% equity acquisition, which means it will likely have to clear antitrust review head-on.

The acquisition extends NVIDIA’s push to plant a flag at every layer of the AI stack, from energy and chips to infrastructure, foundation models and applications. With OpenAI unveiling its own inference chip, Jalapeño, built with Broadcom, and Google and Meta pushing their own TPU and MTIA silicon, NVIDIA has earmarked $18 billion for equity investments for the remainder of this fiscal year — part of what’s been described as a shift in NVIDIA’s moat from chips to capital. NVIDIA, which posted a record $81.6 billion in quarterly revenue back in May, just topped that again with $96.2 billion (up 106% year-over-year) and is guiding to $108 billion next quarter. Its market cap stands at $5 trillion, the largest of any company in the world.

Hugging Face’s rivals in model-sharing include Google’s Kaggle, Replicate, Alibaba’s ModelScope and GitHub Models, but none has matched its scale in registered models, datasets or developer community size.

Huang said Delangue came to him while weighing Hugging Face’s next chapter and believed NVIDIA would be a great home for the company, its community and the future of open models. The Hugging Face team will keep its brand as it moves to a larger stage. The deal is expected to go through customary closing procedures, though no specific timeline or regulatory approval schedule has been disclosed.