NVIDIA's Hugging Face deal alarms

Nvidia’s reported agreement to buy Hugging Face has pushed antitrust questions from AI chips into the software platforms developers use to find and deploy models. CNBC reported on August 27 that Nvidia had agreed to buy the open-source AI platform for $12.9 billion, citing The Information and a source familiar with the matter. Neither Nvidia nor Hugging Face immediately commented to CNBC on the report. If completed, the deal would place one of the best-known repositories for open-source AI models under the control of the company that dominates AI accelerators. ### Why are regulators and analysts focused on Hugging Face rather than just Nvidia’s chips? Hugging Face is used by developers to collaborate, test and share AI tools, according to CNBC’s account of the reported deal. Pomegra described the company as a Paris-founded, New York-based platform used by more than 13 million developers and hosting more than 500,000 AI models. That makes it a distribution point as well as a software platform. (cnbc.com) Nvidia already holds a dominant position in AI accelerators, and Pomegra said ownership of Hugging Face would give it a direct link from chips to the software layer above them. Pomegra said that combination could let Nvidia influence what gets promoted or optimized on a platform used by rivals including AMD, Intel, Google and Amazon. The publication said those concerns are likely to shape competitor lobbying in any regulatory review. (cnbc.com) ### Why is the reported price drawing so much attention? The $12.9 billion figure would make Hugging Face Nvidia’s largest acquisition to date, according to CNBC and Pomegra. Pomegra said the price would exceed Nvidia’s $6.9 billion Mellanox purchase and stand at nearly three times Hugging Face’s $4.5 billion Series D valuation from August 2023. A separate set of reports has cited a much lower figure, around $2.9 billion, which has added to uncertainty around the transaction’s terms. (pomegra.io) The consistent point across the published reports is not the exact valuation but the structure: Nvidia would move beyond hardware into ownership of a major model-distribution venue. That is the feature drawing antitrust attention in the coverage reviewed for this article. (cnbc.com) ### What does “moving up the stack” mean in this case? Tearsheet wrote on September 2 that “the antitrust question is moving up the stack,” describing a broader shift from scrutiny of semiconductors alone to scrutiny of software, financing and distribution layers around AI. The article pointed to Nvidia’s paused AI Compute Partnership Program as another example of regulators and lawmakers asking how much of an ecosystem one company should control. (pomegra.io) California lawmakers are considering AB 1776, known as the COMPETE Act, which Tearsheet said would broaden the state’s antitrust framework to reach more single-firm conduct. Tearsheet said the bill reflects a wider policy question: when support for an ecosystem becomes the use of market power to shape that ecosystem. (tearsheet.co) ### What would competitors worry about if the deal closes? AMD, Intel, Google and Amazon all distribute models through Hugging Face that are tuned for their own chips or cloud environments, Pomegra said. A Nvidia-owned platform could create incentives to favor CUDA-compatible workflows or Nvidia-linked tooling, even without formal exclusivity, according to that analysis. Siddy Jobe, a fund manager at Eonopolis Exponential Technologies funds, told CNBC that Hugging Face fit Nvidia’s broader effort to be “integrated in the entire stack vertically, going from energy to foundational models and also to applications.” His comment was supportive of the industrial logic of the deal, but it also underscored why rivals and regulators are examining the breadth of Nvidia’s reach. (tearsheet.co) (pomegra.io) ### What happens next? Any next step depends on a formal agreement and regulatory filing. Pomegra said no signed agreement had been reached as of its publication, while CNBC said talks were ongoing and recent. Both companies had not publicly confirmed a completed transaction in the reports reviewed here. The next concrete milestone would be a signed merger agreement or a filing that triggers review by U.S. and European authorities. (cnbc.com) California’s AB 1776 debate is proceeding separately, but Tearsheet said it is part of the same policy push over how antitrust law should treat powerful companies operating across multiple layers of AI infrastructure. (tearsheet.co)