Nvidia agreed on September 3 to buy Hugging Face for $12.93 billion, putting the developer’s default model hub under a chip maker’s roof

  • The deal and what it covers: Nvidia confirmed on September 3, in a blog post from Jensen Huang, that it is acquiring Hugging Face for $12.93 billion. Hugging Face hosts 3 million models, 1 million applications and 500,000 datasets, and over 18 million developers use it. TechCrunch · NVIDIA
  • The neutrality pledge: Huang says Hugging Face “will remain an open platform for the entire AI ecosystem,” letting developers pick their own models, frameworks, clouds, inference providers and compute. “Nvidia compute will not be required.” Hugging Face CEO Clément Delangue approached Huang weeks ahead of the deal. CNBC · The Guardian

The owner of your default model hub just changed hands, and “neutrality” now governs your whole stack

  • What touches your build: Hugging Face is the front door for model downloads, deployment, datasets and inference APIs. Handing that door to the company that sells GPUs puts even products that never touch an Nvidia chip on a path that pulls toward those chips. For now you are trusting a promise.
  • What open source just bought: A chip giant buying the open hub reads as a win for open models, but it is also the biggest platform forcibly buying a chokepoint of the open ecosystem. Indie builders will find out which reading is true inside their own products.

Do not chain your deployment pipeline to one hub; build your exit path now

  • Experiment 1, keep two homes for assets: Stop storing models and datasets only on Hugging Face. Mirror them to another repo or your own storage, and decouple your deploy layer from the model repo so a policy shift does not move your product.
  • Experiment 2, keep your inference vendor swappable: Ride on a standard interface and keep vendor swaps cheap. If Nvidia’s pull widens, your structure still survives because swapping is a few lines away.

The deal still has to clear antitrust review, and whether neutrality actually holds is open

  • What to hold off on: The acquisition faces mandatory antitrust review in the US and Europe. Until it clears, do not bank on Hugging Face’s ownership staying as it is or treat the hub as a permanent business asset.
  • What to keep watching: Whether Nvidia compute stays a no-strings option or quietly becomes the favored one, and whether the multi-accelerator pledge shows up in real APIs and pricing. If neutrality turns out to be a tagline, your exit path is the answer.

The rest of today’s news

  • Google DeepMind released WeatherNext 3: An hourly-updated global weather model at 5km resolution, rolling into Search, Maps and Gemini. Google blog
  • Qualcomm Ventures put $70M into Ultrahuman: A $365M valuation round to turn the smart ring into an on-device AI computer. TechCrunch
  • Wafer raised a $40M Series A for inference optimization: Investment is flowing to platforms that make AI inference faster and cheaper. AI Market Watch