It sounds paradoxical at first: the hottest acquisition targets in the Valley right now are the companies that give their AI models away. TechCrunch has pulled together the deals of the past few weeks, and the numbers are no joke. Nvidia is reportedly close to buying Hugging Face – for around 13 billion dollars. Just before that came a 6-billion-dollar deal with Poolside, an open-weight builder, with most of its staff moving to Nvidia. And two weeks ago, Stripe bought OpenRouter for over 7 billion.
Why is so much money flowing into a space built on giving stuff away? For Nvidia it’s about independence. Big labs like OpenAI and Google are now building their own inference chips – see OpenAI’s Jalapeño. So if the model builders are becoming chip makers, Nvidia wants a slice of the model business in return. With Hugging Face it would gain the largest U.S. developer space for open models, and with it a mass of users it can steer toward its own chips.
What’s striking is how small the actual market still is. According to a Ramp analysis, just 6 percent of companies use open-weight models; among developers, Jellyfish puts it at only 2 percent. The typical use is high-volume, repetitive work – customer service chats, say, where a cheaply tuned model answers the same questions over and over. For coding and agent tasks, the frontier models still win. Today the main reason to reach for open models isn’t price, but control and configurability.
That there’s more here is clear from Fireworks. CEO Lin Qiao says her platform processes 40 trillion tokens a day – more than the APIs of Gemini or OpenAI. Her bet: the more models exist, the easier it gets for any company to train its own for its purpose. “Every single app company should consider hiring an in-house researcher,” she says. The future, in her view, is specialized intelligence, one model per use case.
For us as Claude users, this is a fascinating side note. Day to day, most of us run on the strong, closed models – and that holds for now, because they simply lead on coding and agents. But the message underneath is clear: the dominance of OpenAI and Anthropic isn’t set in stone. As long as the giants pour this much money into open alternatives, they’re deliberately keeping a back door open. And if the frontier labs’ prices keep climbing, that door gets more tempting for more and more companies.
Sources: Open-weight AI companies are the Valley’s hottest acquisition targets (TechCrunch) · Nvidia closes in on Hugging Face acquisition (TechCrunch)