AI · AI models

By Conduit Report Editorial

Nvidia is buying Hugging Face, and it changes who controls open-source AI

A ~$12.9B definitive agreement puts the biggest open-model hub under the biggest AI chipmaker. What's confirmed, and why it's a bigger deal than the price.

Stated by the maker, publisher, or platform holder.

Nvidia has agreed to buy Hugging Face — the platform where much of the open-source AI world hosts its models, datasets and tools — in a deal reported at around $12.9 billion. It's a definitive agreement, not a rumour, and if it closes it puts the industry's central open-model hub under the company that already dominates the hardware AI runs on.

The price is the least interesting part. What matters is what it means for who controls the plumbing of open-source AI.

What's confirmed

The number is described consistently across reporting, and "definitive agreement" is a specific legal stage — both sides have signed, and it now needs regulators to clear it. That's further along than a rumour or a reported interest, which is why this carries a Confirmed stamp on the fact of the deal, even though the close is still pending.

What Hugging Face actually is

If you're not in AI, Hugging Face is easy to underrate. It's often called "the GitHub of AI" — the default place where developers publish and download open models, share datasets, and grab the tools to run them. When a new open model launches, it lands on Hugging Face. When a developer wants to build on someone else's model, that's usually where they get it.

That makes it a control point. Not the models themselves — a neutral hub through which an enormous share of open-source AI development flows. Whoever owns it owns a chokepoint in the ecosystem.

Why it's a bigger deal than the price

Nvidia already sells the chips almost all AI trains and runs on. Owning Hugging Face on top of that means it would control both the hardware layer and the main distribution layer for open models. That's vertical integration of a kind the industry hasn't seen — the equivalent of the company that makes the engines also owning the biggest marketplace where cars are bought and sold.

The optimistic read: tighter integration between models and Nvidia hardware could make open models faster and easier to deploy. The pessimistic read: a neutral, community-oriented hub coming under the control of the industry's most powerful company shifts who sets the terms — on licensing, on hosting, on which hardware things are optimised for. Both readings are reasonable, and which one plays out depends on how Nvidia runs it.

The pattern

This fits everything Nvidia has done in 2026: it isn't just selling chips anymore. Between its data-centre guarantees, its push into PC silicon, and now the biggest open-model hub, Nvidia is building an ecosystem that spans hardware, infrastructure, and distribution. The company spent years being treated as the only serious option for AI compute. It's now moving to be the default option for several more layers of the stack.

Whether regulators let a deal this consolidating go through unmodified is the open question. "Definitive agreement, subject to approval" means the signatures are done and the scrutiny is just beginning.

Published 05 September 2026. We add a dated update line to this page when something it covers materially changes.