The Information reports Nvidia agreed to buy Hugging Face for $12.9 billion

The Information reported on Wednesday that Nvidia has agreed to buy Hugging Face for $12.9 billion, citing a person familiar with the agreement. Amanda Caswell covered the implications for The New Stack on 27 August. Neither company had publicly confirmed the deal at the time of publication.

The conflict

Hugging Face does not push developers towards one chipmaker. Its Optimum libraries work with Nvidia's TensorRT-LLM and also support AMD, Intel and AWS hardware, with Optimum AMD and Optimum Intel letting Transformers and Diffusers models run on non-Nvidia chips.

That neutrality is a large part of what makes the platform valuable, and it is what a chipmaker owning it puts in question.

There is already a gap between the two parts. Through the open-source libraries, developers get a wide hardware choice. Through Inference Endpoints, the hosted product, most listed GPU options are Nvidia: T4, L4, A100. Nvidia's NIM already pulls models straight from the Hub through an hf:// path.

Nvidia has announced no plan to make NIM the default, and AMD and Intel support could stay as it is. The article's point is subtler and more likely: a competing chip can lose even while it stays on the platform. It only needs the Nvidia path to take fewer steps.

The indirect effect is the one to watch. AMD, Intel and AWS have to decide how much engineering they want to contribute to integrations maintained inside a competitor-owned platform.

Why it is worth this much

At roughly $150 million annualised revenue, per The Information, $12.9 billion is a multiple of about 86. Hugging Face chief executive Clément Delangue told The Information in June that paying subscribers doubled in the first half of 2026 and later said the company was close to profitability. None of that explains 86 times.

So the price pays for market position, and that position protects Nvidia against rival chips. Google has TPUs, AWS has Trainium, Microsoft has Maia, and OpenAI and Anthropic are both developing server chips. Five European companies committed this month to buying compute built on non-Nvidia hardware that has not been manufactured yet. OpenAI published results for its Jalapeño accelerator showing 1.5 to 1.9 times more work per watt and up to 3.6 times lower end-to-end latency on large open-weight models, though it is not deployed at anything near Nvidia's scale.

Open models run in many environments, and Hugging Face is what makes that practical. Owning it protects Nvidia against every one of those chips.

If your model downloads, your fine-tuning and your deployment path all run through one platform, this is worth an hour of thought before either deal closes.

The repository where open models are stored, the marketplace where they are rented and the licences they arrive under can all change under one owner. Knowing which parts of your stack would have to move, and how quickly, is cheap to work out now.

Source

• Nvidia's $12.9B Hugging Face deal has an open-source problem, The New Stack, 27 August 2026, reporting on a story first published by The Information