Analysis
Nvidia's $12.9 billion acquisition of Hugging Face was confirmed Sept. 3, and Pulse covered the deal terms then. What has changed since is the shape of the objection.
The initial reaction was about the number. The reaction now, laid out in The Register's Monday discussion of the deal, is about governance. Hugging Face is where roughly 80 million users publish and pull open models -- the closest thing the field has to a neutral registry, described in that piece as essentially the GitHub of AI. Nvidia sells the hardware those models run on. Owning the distribution layer for open weights while selling the accelerators creates a conflict that no amount of stated independence resolves on paper.
“The deal is not expected to close until 2027, pending regulatory approval.”
Two details from the transaction are worth restating. About $1 billion of the consideration is earmarked for Hugging Face employees joining Nvidia, which is a retention structure, not a purchase price -- it tells you Nvidia's model of the asset is the team and the community, both of which can leave. And Hugging Face turned down a $500 million Nvidia investment in 2025 before accepting acquisition this year, which is a large swing in eighteen months for a company whose CEO says it is targeting 100 million users.
The deal is not expected to close until 2027, pending regulatory approval. That is a long window in which a credible alternative registry could get funded -- and a long window in which model publishers who care about neutrality can migrate. The open question nobody has answered is what happens to a maintainer whose model runs best on a competitor's silicon.
