Nvidia doesn’t need to block rival chips on Hugging Face. It just needs the defaults.
The New Stack Amanda Caswell ● Covered by 2 sources
Rumor — unconfirmed reporting.
Nvidia is reportedly buying Hugging Face for $12.9 billion. The real power move isn’t blocking rivals; it’s making Nvidia the easy default.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Nvidia has reportedly agreed to buy Hugging Face for $12.9 billion, a deal that would put a dominant chipmaker in charge of a platform many developers use to find and run open models. The Information first reported the agreement on Wednesday, citing a person familiar with it. Neither company had confirmed it publicly at the time of publication.
What makes this deal interesting is that Hugging Face has never been a simple Nvidia shop. Its Optimum libraries work with Nvidia’s TensorRT-LLM, but they also support AMD, Intel, and AWS. There are even projects such as Optimum AMD and Optimum Intel that help developers run Transformers and Diffusers models on non-Nvidia hardware. So the platform already sits in the middle, helping developers get a model from idea to actual deployment.
That middle layer matters a lot. Hugging Face is more than file hosting now. Through Inference Endpoints, developers can deploy a model from the Hub while Hugging Face handles the infrastructure underneath. Those hosted deployments can run on AWS, Microsoft Azure, or Google Cloud, but the GPU list Hugging Face shows is heavily skewed toward Nvidia, including the T4, L4, and A100. The open-source tooling offers broader hardware choice than the hosted services do.
Nvidia already has a foot in the door through NIM, which works with models hosted on Hugging Face. A developer can point NIM at an hf:// repository path and pull a model straight from the Hub. If Nvidia owned Hugging Face, it would have more room to fold NIM and CUDA-optimized containers into the deployment flow itself. Nvidia has not said it plans to make NIM the default, and support for AMD and Intel could stay exactly where it is. But defaults are the quiet part of platform power.
That’s the bigger story here. Nvidia is also pushing its own Nemotron open models while its biggest customers keep building ways to rely less on Nvidia hardware. Google has TPUs, AWS has Trainium, Microsoft has Maia, and OpenAI and Anthropic are building their own AI server chips. So if Nvidia buys Hugging Face, it isn’t just buying revenue — The Information puts annualized revenue at about $150 million — it’s buying the place developers already go when they want open models to work across different systems.
My take — AI-written commentary, not fact-checked reporting
This is the old platform trick with a new coat of paint: don’t kill the rival path, just make your path smoother. That’s how “open” software gets slowly squeezed without anyone having to ban anything. The joke is that everyone calls it flexibility right up until the defaults start doing the political work.
Read more about this at: The New Stack