Update: NVIDIA Reportedly Makes A $12.9 Billion Power Play To Buy Hugging Face

If you've been playing with (or seriously using) local AI models in a home server environment, there's a very high chance that the tutorial you followed included creating a Hugging Face account and generating an OAuth token. The reason for that is it's the largest repository of open weights on the internet, and every big AI model developer has a presence there. Whether you're serving local AI models like Qwen in your basement, or building out a datacenter to serve open frontier models like Kimi K3 or Deepseek 4 Pro, high-speed downloads for enormous files generally come from Hugging Face. 

However, for all the purpose Hugging Face serves, a business model for the AI-hosting startup is hard to pin down without sponsorships from the same companies that use it. It does sell private hosting space on a subscription basis to companies who want to host their own private finely tuned model variants. However, it does not currently offer subscriptions to actually run those models on its own hardware in the cloud. It has chosen to offer its services as a support structure for inference companies, not compete with them. 

NVIDIA is a very natural fit to acquire Hugging Face, considering how much the green team's hardware business is driven by rapid growth in artificial intelligence. And considering the company just announced another quarter of astronomical year-over-year growth, it seems NVIDIA has plenty of cash on hand to cover it. If the $12.9B figure reported by The Information is correct, CEO Jensen Huang might even have it in his couch cushions. 

This is also pretty consistent behavior for NVIDIA. Just last month, NVIDIA and many other companies banded together to create the Open Secure AI Alliance, which seeks to keep tools open-source and trustworthy. Open weight models hosted by Hugging Face are definitely a big part of that. Beyond just models, Hugging Face also hosts training datasets used by researchers to train or at least fine tune their own models.

And of course, all of the open playbooks hosted by NVIDIA's developer portal for use on GPUs and AI workstations like the DGX Spark rely on open tools and hosted repos. In other words, avoiding a big shake-up in model distribution is in the company's financial interests as well. We're sure the service has seen a big uptick in traffic over the last couple of weeks, as mid-sized models like Alibaba's Qwen 3.8, Meta's Muse Glimmer, and NVIDIA's own Nemotron 3.5 Lightning all launched in August 2026. 

So far, both companies have been tight-lipped on the reported acquisition, and at the moment the deal is being reported is rumored. While The Information's article is paywalled, we think the headline pretty much says it all. 

UPDATE 2026-08-28 10:27 AM EDT: In February of this year, Georgei Gerganov and the GGML team that develops the open-source AI framework llama.cpp joined Hugging Face, which likely means ownership of the project would transfer to NVIDIA or any other potential suitor. While llama.cpp has a very generous MIT license that cannot be easily changed, it's in the open source AI community's best interest that a new owner would be a good steward of the platform. It's consistently been the most friendly towards non-NVIDIA hardware with its own Vulkan and Metal backends to support AMD, Intel, and Apple hardware, and we're hopeful it would remain that way whether Hugging Face is acquired by NVIDIA or another company.