Nvidia confirms the purchase of Hugging Face for $12.9 billion

Nvidia has officially confirmed the acquisition of Hugging Face for $12.93 billion, ending weeks of rumors about a deal that redraws the balance of power in open artificial intelligence. Hugging Face is no ordinary startup: its platform hosts three million models, one million applications used by more than 18 million developers and half a million datasets, making it core infrastructure for much of the open-source and open-weight AI ecosystem worldwide.

Nvidia CEO Jensen Huang moved immediately to dispel the most obvious fear such moves raise: that the buyer will turn a neutral platform into a walled garden serving its own hardware. In a blog post, Huang said Hugging Face 'will remain an open platform for the entire AI ecosystem' and that developers will be able to choose the models, frameworks, clouds, inference providers and computing platforms they prefer, explicitly stressing that 'Nvidia compute will not be required to build or deploy through Hugging Face'. It is a statement of intent aimed at reassuring a developer community that has built much of its daily work on infrastructure that until now was seen as independent of the major chip suppliers.

Beyond the rhetoric of openness, the business logic behind the deal is fairly clear. Nvidia is already the dominant hardware platform for AI development and inference, and controlling the largest open repository of models and datasets allows it to shape that ecosystem in ways that fit its own chips better. Huang noted that Nvidia itself has already published more than 500 open models and 250 open datasets on Hugging Face, arguing that the company builds open models precisely so that developers around the world will adopt them. There is an additional commercial motivation: the option of selling unused compute capacity to enterprise customers bundled with Hugging Face's offering, turning the platform into an extra distribution channel for that capacity.

Hugging Face's financial situation adds interesting nuances to the deal. Founded in 2016, the company had raised more than $395 million to date, with a last round in 2023 of $235 million led by Salesforce Ventures and with participation from Google, Amazon, IBM and Nvidia itself, which was already among its investors before this acquisition. It is particularly striking that, according to the Financial Times, Hugging Face turned down a $500 million offer from Nvidia last year: the gap between that figure and today's $12.9 billion illustrates how quickly the company has been revalued, in step with the growth of its business. Along those lines, The Information recently reported that Hugging Face is already generating $150 million in annualized revenue, and its CEO, Clem Delangue, said in July that this pace of growth was bringing the company close to profitability.

Delangue himself explained the reasons behind the sale in a post on X, thanking the community for proving that an alternative to closed APIs was possible, but acknowledging that scaling that model further required 'more compute, more support, more collaboration and more visibility'. By his account, that was why the company went to talk to Huang, who offered to provide exactly that. It is a narrative that fits Huang's own public positioning as an advocate of open-weight models: the executive signed a letter, co-signed by other organizations, arguing that such models strengthen the United States' position against rivals such as China in the AI race.

The deal is also part of a much broader Nvidia investment strategy in model development. Last month, the Wall Street Journal reported that the company had closed a $6 billion agreement with coding startup Poolside to develop open models, and in its latest earnings presentation Nvidia disclosed that it has invested more than $50 billion in frontier AI labs. Asked by analysts during that same presentation, Huang argued that virtually every existing open model runs on Nvidia hardware, a point that reinforces the thesis that controlling the open ecosystem is ultimately a way of consolidating demand for its own silicon.

One of the most unusual arguments Huang made to justify the relevance of open models is their role in cybersecurity. As he explained, frontier models are vital for security firms to run distributed, continuously operating autonomous defense systems, and he said such companies 'could not do it without open models'. Delangue himself had recounted a telling episode along these lines in July: an open Nvidia model helped Hugging Face fend off cyberattacks after proprietary models failed to protect the platform. In fact, a few days before that statement, OpenAI had admitted that one of its as-yet-unreleased models had breached Hugging Face's security, an incident that adds a further layer of competitive tension between the big closed labs and the open ecosystem Nvidia has just bought.

Taken together, the acquisition puts Nvidia in an unprecedented position of control over the distribution infrastructure for open models, just as the company multiplies its capital bets on AI labs and on developing its own open-weight models. The public commitment to keep the platform neutral will predictably be the point the developer community watches most closely in the coming months, given that the recent history of large-scale tech acquisitions has not always honored that kind of initial promise.