NVIDIA acquires Hugging Face for $12.93 billion and commits to not locking the platform
NVIDIA announced on 3 September 2026 the acquisition of Hugging Face for $12.93 billion, promising to keep the platform open and multi-cloud. If you build on open-weight models, it is Jensen Huang’s written commitments — not the announcement — that you should watch over time.
3 September 2026. NVIDIA announces it is acquiring Hugging Face for $12,930,300,000. 18 million developers, 3 million models, 500,000 datasets, 200,000 companies. That is the weight of the reference hub for open-source AI, now passing under the control of the world’s largest GPU maker. Why it matters: the question is not the price tag, but whether the one neutral point of the open-weights ecosystem will stay neutral.
Announced by Jensen Huang himself
The announcement is signed by Jensen Huang, who published the post on the NVIDIA Blog that same morning. After days of rumors, the statement confirms the exact figure — $12.93 billion — and the stated intent: to “scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide.”
The numbers cited give the scale of the target. More than 18 million developers, researchers and creators share more than 3 million models, 500,000 datasets and 1 million applications on Hugging Face. More than 200,000 companies use the platform to discover, evaluate, customize and deploy AI. Huang salutes Clem, Julien, Thomas and the team who built “a vibrant home for the open model developer community.”
The deal is not a complete surprise. NVIDIA was already the largest contributor of models and data on the platform, with more than 500 models and 250 open datasets published. The transaction extends an openness strategy the company has pushed for years.
Hugging Face, the de facto registry of open weights
To grasp what is at stake, measure what Hugging Face has become since its founding in 2016 by Clément Delangue, Julien Chaumond and Thomas Wolf. Born as a natural-language-processing library — the transformers package — it grew into a four-sided platform: the Hub, where models and datasets are stored and downloaded; Spaces, for hosting demos; Inference Endpoints, for serving a model without infrastructure; and the open-source libraries that irrigate the ecosystem.
It is this registry position that gives the target its strategic value. When an open-weight model ships — from Meta, Mistral, DeepSeek or a university lab — it almost always lands on the Hub within hours. The weights are versioned, signed, and downloaded by millions of machines. Controlling the registry is, in theory, controlling the choke point: which is exactly why Huang’s neutrality commitment matters more than the price tag.
The comparison situates the deal. Where Microsoft owns GitHub and Google fields Kaggle and Vertex AI Model Garden, NVIDIA had no direct contact point with the model-developer community — only the hardware. Buying Hugging Face gives it the channel that was missing between its GPUs and developers, without having to build one from scratch.
The written commitment: no NVIDIA lock-in
The part an infrastructure operator cares about is the lock-in — or its absence. Huang writes, in black and white, what the community feared it would not read: “Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want.”
Two sentences go further. “NVIDIA compute will not be required to build on or deploy through Hugging Face.” And: “It will continue to support multi-cloud and multi-accelerator development and deployment.” In plain terms, the platform does not become a forced GPU sales channel, and a model hosted there will not have to run on NVIDIA hardware.
These commitments did not come out of nowhere. Huang notes he coauthored an open letter on the importance of open weights to the AI economy, joined by leaders across the industry. The argument is economic as much as ideological: open models let startups, universities and public institutions build on advanced capabilities without training every model from scratch, and “match the right model to the right job.”
Why a GPU maker buys a neutral hub
The strategic question is the same as in every consolidation of shared infrastructure: who does neutrality serve? Hugging Face is not just a model-sharing site. It is the de facto registry for open weights, the place where models get downloaded before running on third-party hardware, including accelerators that compete with NVIDIA.
The risk is not theoretical. An operator deploying open-weight models on AMD, Intel or Cerebras hardware needs a hub that does not quietly favor the CUDA stack. The “multi-accelerator” commitment answers that fear directly — but a press release is not a governance guarantee.
On top of the governance question sits the regulator one. A $12.93 billion acquisition by a company that already controls the overwhelming majority of AI accelerators will not pass unnoticed by competition authorities on both sides of the Atlantic. The Microsoft–Activision precedent showed that even a deal eventually cleared can drag on for months and force concessions. If independence safeguards are required, they become the strongest neutrality guarantee the community could get — stronger than a press release.
Recent context argues for a cautious read. Days earlier, the Open Secure AI Alliance, initiated by NVIDIA, moved to the Linux Foundation — an institutional gesture showing the company understands the value of governance outside its direct control for reassuring an ecosystem. Whether Hugging Face receives equivalent treatment, or remains a wholly-owned subsidiary, is the open question.
Verdict
If you build on open weights, change nothing today: the announcement alters neither the APIs, nor the downloads, nor the licenses of hosted models. Do, however, record the two precise commitments — no NVIDIA compute required and multi-cloud / multi-accelerator — and track their translation into concrete action: independent governance, maintained open licenses, unchanged pricing.
If you are a Hugging Face customer or partner, the consolidation has an immediate upside: NVIDIA’s infrastructure, engineering and global reach can improve reliability, model evaluation and inference capabilities — historically weak points of the hub relative to the clouds.
If you are evaluating the openness of your AI stack, treat this acquisition as a long-term test rather than a done deal. The real guarantee is not the 3 September statement, but what the platform looks like in two years: still neutral, or progressively optimized for a single accelerator.