machine-learning A field note by Vikrant Sharma
Nvidia just bought Hugging Face for thirteen billion dollars
The GPU monopoly now owns the model zoo. This changes who controls open-weight AI.
Nvidia agreed to acquire Hugging Face for thirteen billion dollars. Not a partnership. Not an investment round. Full acquisition. Hugging Face is where every open-weight model lives. LLAMA, Mistral, Flux, Stable Diffusion, the weird Chinese vision models that beat GPT-4V on benchmarks nobody asked for. If you have ever downloaded a safetensors file, you downloaded it from Hugging Face. They host the artefacts, the datasets, the inference endpoints. They are the PyPI of machine learning. Nvidia already sells the shovels. Now they own the mine. Every researcher who fine-tunes a model, every startup that serves open weights because they cannot afford OpenAI pricing, every compliance team that needs local inference, they all touch Hugging Face infrastructure. Nvidia gets telemetry on what models people actually run, what datasets they use, what quantisation formats work. That is not public information today. The interesting bit is timing. Thirteen billion is expensive for a company that gives away most of its product. Hugging Face makes revenue from enterprise inference and storage, but that is not thirteen billion in present value. Nvidia is paying for position. If open weights win the next three years, Hugging Face is the distribution layer. If closed wins, Nvidia still sold the H100s to OpenAI. The acquisition also kills the idea that Hugging Face was neutral infrastructure. They took VC money, they were always going to exit. But the framing was community-first, researcher-friendly, the good guys. That credibility does not survive an Nvidia acquisition. Expect forks. Expect someone to spin up a nonprofit mirror within six months. What I am watching: whether Nvidia keeps the platform open or starts gating model uploads behind GPU quotas. If you want to host your fine-tune, you rent our A100s first. That would be the move.
