Open-weight AI companies become the Valley's hottest acquisition targets

Everyone is waiting for Nvidia to confirm what would be this week’s most interesting tech deal: a reported $13 billion acquisition of Hugging Face, a platform for sharing open-weight AI models and benchmarks. Hugging Face sits at the center of the ecosystem of developers building and deploying large language models not owned by frontier labs — something like a GitHub for the AI era.

The rumor follows Nvidia’s $6 billion agreement with Poolside, an open-weight model builder, under which most of Poolside’s employees will move to the chip maker. Two weeks ago, Stripe acquired OpenRouter, the top provider of open-weight models to businesses, for more than $7 billion. That is a lot of capital pouring into a sector built on giving software away, reflecting current AI-industry dynamics. For Nvidia, the appeal is reducing dependence on deals with hyperscalers and frontier labs, especially as model builders like OpenAI and Google develop their own inference chips — OpenAI’s Jalapeño chip had its capabilities announced this week. Nvidia already builds its own Nemotron family of open-weight models, though uptake has not been large; acquiring the largest U.S. open-model developer hub would give it a mass of users to steer toward its chips and standards.

Rising AI inference costs are also pushing some companies toward cheaper models from Chinese developers like Moonshot, DeepSeek and Alibaba, though adoption remains small: just 6% of companies use open-weight models, according to spending-data research from Ramp, and just 2% of software engineers do, per a Jellyfish survey. Jellyfish’s AI product lead, Nik Albarran, told TechCrunch that open-weight models are mainly used by companies with high-volume, repeated inference workloads, such as customer-service chat. For coding and agentic tasks, proprietary frontier models still tend to win out, partly because of easier access and, in some cases, token subsidies from their makers — though Albarran expects that to shift as frontier-lab prices rise and companies’ AI workflows mature.

Lin Qiao, CEO of Fireworks, an open-weight model router and host that itself is often discussed as an acquisition target, said her company processes 40 trillion tokens a day — more than either Gemini’s or OpenAI’s APIs. “The future is actually specialized intelligence,” Qiao said. “Literally, every single company should have their own model per use case, and that will happen automatically.”

Why it matters

The scale of these deals shows major tech companies no longer see OpenAI and Anthropic’s dominance as inevitable, and are instead paying billions to control the infrastructure and talent behind open alternatives.