Nvidia is reportedly in talks to acquire Hugging Face for $13 billion, according to reports this week, in a move that would give the chipmaker control of a leading platform for sharing and deploying open-weight AI models. The reported pursuit follows Nvidia’s agreement with Poolside, valued at $6 billion in the latest account, and Stripe’s $7 billion acquisition of OpenRouter, which was finalized two weeks ago. The companies have not publicly confirmed a Hugging Face deal, and the latest account differs from an earlier report that described an approximately $12.9 billion agreed acquisition. Open-weight models remain a small part of enterprise AI adoption: Ramp found that 6% of companies use them, while Jellyfish said 2% of software engineers do. Their main advantage is the ability to fine-tune and self-host models for high-volume, repetitive inference, potentially lowering costs and dependence on major AI labs and cloud providers. Nvidia could also use Hugging Face’s developer community and distribution platform to encourage adoption of its hardware and software standards, although ownership would raise questions about neutrality and competition with AMD, Intel and custom cloud accelerators. OpenAI’s development of the Jalapeño inference chip adds to the pressure on Nvidia to defend its position beyond GPUs. Fireworks CEO Lin Qiao said specialized models built around individual companies and use cases could become the next phase of AI deployment, while Fireworks processes 40 trillion tokens daily, more than Gemini or OpenAI’s APIs, according to her account.