The open-weight AI ecosystem is suddenly awash in money, and Nvidia appears ready to write the biggest check yet. Reports point to a $13 billion acquisition of Hugging Face, the de facto GitHub for AI models and benchmarks. The move would cap a frantic few weeks: Nvidia already struck a $6 billion deal with Poolside, and Stripe swallowed OpenRouter for more than $7 billion. That’s over $26 billion in capital chasing a sector built on giving models away.
Nvidia’s motivation isn’t hard to parse. The company that prints money selling GPUs to hyperscalers now watches its biggest customers—OpenAI, Google—build their own inference chips. OpenAI’s Jalapeño chip, detailed this week, is the clearest warning shot. Nvidia already has its Nemotron family of open models, but uptake has been modest. Owning Hugging Face would give Nvidia something more valuable than model weights: distribution. Millions of developers, steered toward Nvidia’s chips and standards by default.
The open-model pitch is shifting, too. Adoption remains tiny—Ramp’s spending data puts it at 6% of companies, Jellyfish at just 2% of software engineers. Nik Albarran, AI product lead at Jellyfish, told TechCrunch the sweet spot is high-volume, repetitive inference like customer service chats, where a tuned open model slashes costs. But for coding and agentic work, frontier models still win on reasoning quality and subsidized token access. “If the prices continue to go up from the frontier labs, more and more companies will be forced to at least consider it,” Albarran said.
Fireworks CEO Lin Qiao is betting bigger. Her company processes 40 trillion tokens daily—more than Gemini or OpenAI’s APIs, she claims—and her thesis is specialization: every company eventually training its own model per use case. “Literally, every single company should have their own model per use case,” she said. Whether Nvidia’s rumored Hugging Face deal accelerates that future or just locks developers deeper into one hardware vendor is the question nobody’s answering yet.
