NVIDIA to Acquire Open Source Platform Hugging Face for $12.9 Billion, Founded in 2016 by Three French Founders in New York
NVIDIA has signed an agreement to acquire the open-source AI platform Hugging Face for approximately $12.93 billion, with about $11.9 billion paid to shareholders and up to $1 billion in retention equity for employees joining NVIDIA, targeting completion in the first half of 2027, subject to regulatory approval.
This marks NVIDIA's second-largest recorded acquisition, below the previous $20 billion purchase of Groq-related assets, and above the $6.9 billion acquisition of Mellanox in 2020. Jensen Huang stated that the platform has over 18 million developers, around 2 to 3 million models, 500,000 datasets, and 1 million applications, with over 200,000 companies using it to discover, evaluate, customize, and deploy models; NVIDIA itself has uploaded over 500 models and 250 open datasets, making it one of the largest contributors.
The company was founded in 2016 by three French founders, Clément Delangue, Julien Chaumond, and Thomas Wolf, in New York, initially starting with a chatbot aimed at teenagers, later evolving into a community for hosting models and datasets. In 2023, it raised $235 million in funding at a valuation of $4.5 billion, with investors including Google, Amazon, NVIDIA, AMD, Intel, Qualcomm, IBM, and Salesforce. Last year, NVIDIA proposed a $500 million investment at a valuation of about $7 billion, which was rejected to maintain independence; Salesforce had briefly considered an acquisition, but NVIDIA secured a deal for full ownership.
Delangue stated that in the summer, they judged that open-source AI had reached a turning point requiring more resources and visibility, proactively contacting Jensen Huang, and the deal was finalized within weeks, with the three founders joining NVIDIA. The platform's annual revenue is reported to be about $150 million, with a valuation multiple of approximately 86 times. NVIDIA has committed to maintaining openness: users can continue to upload and download selected models and datasets, and support other chip and cloud vendors. The deal may face antitrust scrutiny.
Mechanically, the acquisition buys the distribution entry for open-source models, selling a premium for a "neutral Hub independent of chip manufacturers." The funds will come from NVIDIA's ecological investment budget into the hosting platform, rather than purchasing another wafer fab. Beneficiaries include chip manufacturers needing to distribute open weights to developers, as well as cashing out existing shareholders (including competitors like AMD, Intel, Qualcomm); those under pressure include closed-source labs wanting to build their own distribution channels and regulators concerned about a single accelerator manufacturer controlling the entry point.
Supplement: The three founders and their families were estimated by French media to be worth about €650 million before the transaction; the deal price is nearly three times the 2023 valuation, significantly increasing the founders' worth with cash and retention equity. Recently, the platform also experienced a security incident, which Delangue attributed to engineering errors, and was repaired using NVIDIA's version of the Chinese open-source model.
Source: Public Information
ABAB AI Insight
The transition from rejecting NVIDIA's investment to selling Hugging Face to NVIDIA took less than a year, indicating that "independent open-source neutrality" is a priceable asset on the capital ledger, not just a doctrine. The three French founders initially created an emoji chatbot in 2016, then turned the model repository into the default address for developers, with a cost structure unlike OpenAI's pre-training burn, resembling GitHub's distribution and tools. By inviting chip and cloud companies into the shareholder roster at a $4.5 billion valuation, they effectively locked potential buyers at the table.
The capital path is Jensen Huang using acquisitions to supplement "default entry points beyond chips." Mellanox acquired data center networking, Groq acquired inference silicon, and Hugging Face acquired model listing rights. Closed-source labs place chip orders with NVIDIA while developing their own chips to escape Huang; bringing the open-source Hub into the group means capturing the door developers pass through to download models before competitors' self-developed silicon lands. The commitment to continue supporting competing chips is to pass regulatory scrutiny and avoid turning the Hub into a counter that only sells their own cards.
Comparable cases include Microsoft's acquisition of GitHub, Google's acquisition of Kaggle-level communities, and Meta's use of Llama to exchange ecosystems. The difference is that the buyer is the world's most expensive accelerator company, and the target is a shelf of open weights. The industry phase has shifted from "training compute is king" to "who controls the discovery and download of models." Salesforce's failed acquisition attempt indicates that software giants also see the same door, but lost out to chip cash in terms of price and speed of closing.
The structural change is a reconstruction of the industry chain. The mechanism is that open-source does not eliminate entry rent; it merely shifts the rent from model weights to hosting, evaluation, and deployment tools; whoever owns the entry can direct traffic to their inference stack while verbally maintaining neutrality to lower antitrust temperatures. The three founders joining the buyer temporarily mortgages the community's credibility to the expansion calendar of the chip empire.
ABAB News · Cognitive Laws
Once a neutral entry can be priced, independence becomes a transaction term.
Chips sell compute power, while repositories sell download rights.
Rejecting one round of financing often raises the price for the next complete sale.