Nvidia agrees to buy Hugging Face for $12.9bn

By Takeshi Mori · 2026-08-28

What is known and what is not

The deal is unconfirmed. A person with knowledge of the matter told The Information on 26 August 2026 that Nvidia has agreed to acquire Hugging Face for $12.9 billion.^[1] As of 28 August 2026, neither Nvidia nor Hugging Face had issued a press release or filed an SEC report confirming the acquisition.^[2] Until there is a filing or a public statement, every downstream detail, including price, structure and timeline, is single-source.

Treat the deal as probable enough to plan for, but do not restructure your model stack around it yet. The practical question is what changes if it closes.

The scale of the deal

Nvidia's previous record acquisition was Mellanox Technologies, completed in April 2020 for approximately $7.0 billion.^[3] The reported Hugging Face price is nearly double that. Hugging Face raised $235 million in Series D funding in August 2023 at a $4.5 billion post-money valuation led by Salesforce Ventures,^[4] making the reported $12.9 billion price a near-threefold markup in roughly three years.

Nvidia reported a record $96.2 billion in revenue for the quarter ended 26 July 2026. At that scale, $12.9 billion is a meaningful but manageable bet. Nvidia already owns the hardware layer; Hugging Face gives it the distribution layer where developers actually find and pull models.

The neutrality tension

Hugging Face hosts over one million open-source AI models on its platform, making it the closest thing the AI research community has to a neutral commons.^[5] That neutrality is the asset Nvidia would be buying, and the asset most at risk if the acquisition closes.

Hugging Face co-founder and CEO Clément Delangue has framed the company's mission explicitly in terms of openness. "We conduct our mission primarily through openness, which takes many forms, including public communication about the technology, open science, and open source," Delangue said.^[7] Delangue has also said that, as the fastest-growing open-source library of pre-trained models in the world, openness and access have been key to making systems work better for more people.^[7] Those commitments sit uneasily with ownership by the company that sells the chips every model ultimately runs on.

For Australian teams building on open models, the practical risk is platform dependency. A Hugging Face under Nvidia ownership could favour Nvidia hardware in compute recommendations, pricing tiers or model optimisation tooling. No such policy has been announced, and the concern rests on the structural incentive that comes with concentrated ownership, and builders should be mapping their alternatives now.

Regulatory posture

On 23 July 2024, the US Department of Justice, the Federal Trade Commission, the European Commission and the UK Competition and Markets Authority issued a joint statement pledging to promote effective competition in the AI ecosystem.^[6] That statement was written with exactly this kind of vertical integration in mind. FTC Chair Lina M. Khan said at the time that "fair, open, and competitive markets will help unlock the opportunity, growth and innovation that these technologies could provide".^[8]

A chipmaker acquiring the dominant open-model repository gives each of those four bodies a clear theory of harm: the same firm controls both the hardware a model runs on and the platform where developers discover and deploy it. Whether that theory survives merger review is a question for regulators. Nvidia's Mellanox deal closed in April 2020 with US regulatory approval after roughly a year of review.^[3]

Takeshi Mori

Takeshi Mori writes about technology and start-ups. He is curious about how products get built and who they are really for, and he would rather see a thing working than hear it described.