Prime Intellect just raised $130 million to convince companies they don't need OpenAI or Anthropic to own real AI capability, and investors from those very labs' orbit are betting it works.
The San Francisco startup announced a $130 million Series A on July 8, pushing its valuation to $1 billion, according to TechCrunch. Radical Ventures led the round. Nvidia Ventures, Intel Capital, Dell Technologies Capital and Iconiq joined in. That's a serious lineup of strategic money, the kind that shows up when the chip and cloud makers want a stake in whoever ends up training tomorrow's models on their hardware.
But the angel list is the part worth reading twice. Aravind Srinivas of Perplexity is in. So is Box's Aaron Levie, Cloudflare's Matthew Prince, and John Schulman, the OpenAI co-founder who now runs research at Thinking Machines. Tesla's Milan Kovac put money in. So did the founders of Harvey, Cognition and Mercor. You don't usually see this many people who already run frontier AI companies personally back a startup whose entire premise is that you shouldn't need a frontier AI company at all.
Prime Intellect was founded in 2024 by CEO Vincent Weisser, who came out of decentralized science, and CTO Johannes Hagemann, who scaled LLM training at Aleph Alpha. Their pitch to enterprises is blunt. Stop renting intelligence through an API and start owning it. The company sells the compute a business needs to build and keep its own models - plus the reinforcement learning environments and training tooling - rather than sending every query to GPT or Claude and paying per token forever.
A billion-dollar mark is easy to write off as a slow news week's headline. The revenue figure is not. Prime Intellect is now at a $100 million annualized revenue run rate, per TechCrunch's reporting, with Ramp, Zapier and Flapping Airplanes paying for a hosted version of its stack. Ramp's co-founder and co-CEO, Karim Atiyeh, said the agent his team built with Prime Intellect to search spreadsheets beat frontier models on accuracy while running faster and cheaper. That's not a marketing claim. That's a fintech company's own CEO comparing its homegrown agent directly against the labs Prime Intellect is trying to unseat, and finding it comes out ahead.
Total funding now sits above $150 million. The company's earlier $15 million round, led by Founders Fund and Menlo Ventures, carried its own notable backers: Andrej Karpathy, Hugging Face's Clem Delangue, SemiAnalysis's Dylan Patel, Together AI's Tri Dao and Stability AI's Emad Mostaque. The pattern across both rounds is consistent. People deep inside frontier AI keep writing personal checks to a company betting that frontier AI labs won't be the only ones capable of training good models.
What Prime Intellect actually ships is the Environments Hub, an open platform where more than 2,500 community-built reinforcement learning environments sit alongside its Verifiers library and prime-rl training framework. That environments data has already fed INTELLECT-3, the company's own open agentic model. It's the same playbook Meta ran with Llama, except Prime Intellect is selling the tooling rather than just giving away the weights.
A bet against API dependency #
Here's the real tension. OpenAI and Anthropic have spent years arguing that model quality compounds with scale, and that renting their API is cheaper than any enterprise building its own stack - Google makes the same case. Prime Intellect's investors are betting the opposite, that once a company like Ramp can train a narrower model that beats GPT on its own task, at a fraction of the cost, the rental math stops working for a lot of use cases. Frankly, that's the more interesting fight in AI right now, more than any single model release. It's not about who has the smartest general model. It's about whether cheaper and owned, even if only smart enough, beats smartest and rented.
Nvidia Ventures and Dell Technologies Capital backing a company whose customers might buy less API access from Nvidia's own biggest partners is its own small irony. But chipmakers don't really care whose logo sits on the model, only that the training runs happen on their silicon. Whether Prime Intellect can keep growing that $100 million run rate while OpenAI and Anthropic chase enterprise deals of their own is the question the next year will answer.
Also read: Ollama Just Raised $65 Million to Become AI's Quiet Infrastructure Layer • Hachette, Elsevier and Scott Turow Sue Google Over Gemini's Book Training • IBM Suffered Its Worst Trading Day Since 1987 As Clients Chased AI Hardware