Nvidia is paying $12.9 billion to keep open models on its chips
The New Stack Matthew Burns ● Covered by 4 sources
Nvidia reportedly agreed to pay $12.9 billion for Hugging Face. It’s a huge bet on the place developers already use to find open models.
Based on reporting by The New Stack, Matthew Burns — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Nvidia has reportedly agreed to pay $12.9 billion for Hugging Face, according to The Information. That’s a striking sum for a company the source pegs at about $150 million in revenue. But the logic is pretty clear: Nvidia isn’t buying weights, it’s buying the front door developers already use.
The comparison to Microsoft and GitHub fits because the target isn’t just a storehouse. Hugging Face is where open model releases land, where fine-tunes get shared, and where a big chunk of the industry already goes to figure out how to run them. Nvidia already gives away its own open models. This move is about owning the place where everyone else’s models show up.
The timing matters too. Open models are getting easier to run. Ollama released version 0.33.0 on August 21, and it lets Claude Desktop pick models like Qwen, DeepSeek, and Kimi through a local proxy. Apple also released new Mac Mini and Mac Studio configurations this week that appear aimed at running larger models locally. And Alibaba’s Qwen3.8-27B, as Frederic Lardinois noted, can run on a Mac with 32GB of unified memory in 4-bit form.
The bigger shift is cost. Janakirm MSV wrote in May that OpenCode had passed Claude Code on GitHub stars, and the real question for developers was whether their environment can tolerate a single-vendor harness at all. Jason Calacanis put the business version bluntly on X: open source covers most startup use cases, and CFOs start tightening up when the bill gets big. That’s the pressure Nvidia is stepping into.
Nvidia’s own scale makes the price look smaller than it sounds. Aakash Gupta calculated that $12.9 billion is about 12 days of Nvidia’s quarterly revenue, which was $96.2 billion. Meanwhile, some of Nvidia’s biggest customers are working on escape routes: OpenAI with Broadcom, Anthropic on Amazon’s Trainium, Google on its TPUs. The open-model bet is that even when buyers want out, they still end up running on Nvidia CUDA unless they work hard not to.
My take — AI-written commentary, not fact-checked reporting
This is classic platform behavior: if people are going to route around you, buy the toll booth. Nvidia clearly understands that open models are not just ideology; they’re distribution. The funny part is that the more the industry talks about escape routes, the more it keeps shipping on CUDA anyway.
Read more about this at: The New Stack