Image: NVIDIA
Summary
- NVIDIA has agreed to acquire open-model sharing platform Hugging Face for $12.9303 billion
- Hugging Face is an open-source AI marketplace connecting more than 18 million developers and over 3 million models
- NVIDIA said it will keep Hugging Face as an open platform after the acquisition and will not require use of its own compute
NVIDIA has agreed to acquire open-model sharing platform Hugging Face for $12.9303 billion. NVIDIA announced the deal on its official blog on the 3rd, saying it would "expand Hugging Face's platform and strengthen its infrastructure to broaden AI access for developers and institutions around the world."
To put it plainly: NVIDIA has so far sold AI chips to companies like OpenAI and Google without directly competing in model development itself. This time, though, it bought the marketplace where those very models get uploaded and downloaded. A company that sells chips now also controls the storefront where models circulate.
What Hugging Face is
Hugging Face isn't a company that builds AI models itself — it's a marketplace where publicly available models built by other companies gather and get distributed. It was founded in New York in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf, who originally built a teen-focused chatbot app before open-sourcing the underlying technology, which eventually became the platform it is today. Developers often call it "GitHub for models." According to NVIDIA's announcement, the platform brings together more than 18 million developers, researchers, and creators exchanging over 3 million models, more than 500,000 datasets, and over 1 million applications, with more than 200,000 companies using it to find, evaluate, and deploy models. Whether it's Meta's open models, Alibaba's Qwen, or DeepSeek, regardless of who built them, they all funnel through Hugging Face as the distribution point.
"You won't need NVIDIA compute"
One notable part of the post-acquisition operating plan is that openness will stay intact. NVIDIA said Hugging Face will remain an open platform for the entire AI ecosystem, and that developers will still be free to choose whatever models, frameworks, cloud and inference providers, and computing platforms they want. It explicitly stated that NVIDIA compute will not be required to build or deploy models on Hugging Face. The company also said it will keep supporting multi-cloud and multiple accelerator options as before.
NVIDIA's history with open source
This isn't NVIDIA's first move into the open-model ecosystem. Last July, NVIDIA joined more than 200 other companies and institutions in signing an open letter emphasizing the importance of open weights, and it has contributed hundreds of open models and datasets to Hugging Face over time. In August, it released Nemotron 3.5 Lightning — an open model that delivers 30B-class performance with just 3B active parameters — through Together AI. Around the same time, it also open-sourced Cosmos 3, an open physical AI model that combines vision reasoning, world generation, and behavior prediction, under the Linux Foundation's OpenMDW 1.1 license. NVIDIA said in its announcement that "AI advances faster when people build together."
Editor's take
Up to now, NVIDIA has occupied something like a referee's seat in the AI race — OpenAI, Google, Meta, all of them ultimately buy NVIDIA chips to train their models. This acquisition pushes that position a step further. Beyond the hardware used to train models, NVIDIA now also controls the storefront where those models actually gather and spread. As the open-source camp grows, where models get hosted and how they're distributed becomes a form of power in itself — and NVIDIA now directly manages that layer.
NVIDIA's pledge not to force its own compute is an obvious defensive move. Hugging Face's value comes from its neutrality — not being tied to any particular hardware. If NVIDIA had insisted on its own GPUs right after the acquisition, developers would likely have migrated to alternative platforms. That said, there's still room for a subtler shift over time, where inference optimizations or new features quietly get attached to NVIDIA hardware first. Whether that actually happens will show up in Hugging Face's future deployment options and how models get recommended.
For companies and development teams outside the US, there's likely no immediate change to the models they're already using or how they access them. Still, the fact that a chip maker now owns the distribution platform for the open-model ecosystem is a variable worth factoring into longer-term decisions about model choice and deployment strategy. Over the coming months, the real test of this deal will be whether Hugging Face actually sticks to its multi-cloud, multi-vendor principles in practice.
Comments