Prime Intellect, a startup founded in 2024 that supplies compute and specialized software tools to help companies build AI agents, has closed a $130 million Series A round at a $1 billion valuation.
The round was led by Radical Ventures. Other participants included Nvidia Ventures, Intel Capital, Dell Technologies Capital, Iconiq, and a group of angel investors who are founders of notable companies: Aravind Srinivas (Perplexity), Aaron Levie (Box), Winston Weinberg (Harvey), Jeff Wang (Cognition), and Brendan Foody (Mercor).
Mission and product
Prime Intellect’s stated goal is to enable organizations to train their own agentic systems without depending on closed frontier AI labs. The company points to reinforcement learning techniques—which iteratively reward successful task completion and penalize errors—as a way for businesses to become effectively their own AI labs by refining models for specific workflows.
Despite advances that make it possible to bypass closed labs, the underlying infrastructure remains complex, and most companies lack the expertise to assemble those components into production-ready systems. Prime Intellect addresses that gap.
The startup offers what it calls a "full stack" for AI agent development: access to compute, a reinforcement-learning framework, and evaluation tools. The platform operates like a marketplace with modular access, so customers can pick and choose components rather than adopting an all-or-nothing product.
David Katz, a partner at Radical Ventures, said Prime Intellect has stitched these pieces together in a way that operates at the frontier while remaining affordable. He contrasted Prime Intellect with providers that offer only parts of the stack, and described the startup as a "one-stop shop" that delivers capabilities similar to a top-tier AI lab.
Customers, revenue and outcomes
Prime Intellect has already attracted customers including Ramp, Zapier, and Flapping Airplanes, which pay for hosted versions of the startup’s tools. According to the report, that adoption has driven the company to an annualized revenue run rate of $100 million.
Concrete customer outcomes helped spur growth: Ramp used Prime Intellect to build an agent that finds answers inside spreadsheets. Karim Atiyeh, Ramp’s co-founder and co-CEO, said the resulting system beat frontier models on accuracy while running faster and costing a fraction of the alternatives.
Why companies are interested
Another factor behind Prime Intellect’s traction is growing corporate concern about the risks of building on top of frontier labs. Many organizations are reluctant to provide proprietary information to providers such as OpenAI and Anthropic because of data-control concerns. They also worry about dependence on models that might be shut off or materially changed—an issue highlighted by the report’s reference to Anthropic’s Fable last month.
Katz summarized the unease: companies ask how they can be sure they are not working with a partner that could later replace them or generalize their work. "All of these things are causing people to think, 'How do I own my own enterprise intelligence and not have these risks'", he said.
Prime Intellect co-founder and CEO Vincent Weisser told TechCrunch that enterprises are moving away from closed-source frontier models, and that Prime Intellect provides the infrastructure to facilitate that transition. He argued that the ability to train AI models should not be limited to a few researchers in a glass tower in San Francisco, but should be available to every enterprise and nation-state.
Conclusion
The $130 million Series A and $1 billion valuation underscore investor interest in a modular, hosted stack that lets companies build and own agentic AI for business tasks. With a growing customer base, cited performance gains, and a reported $100 million annualized revenue run rate, Prime Intellect is positioning itself as an infrastructure provider for enterprises that want to control their AI development outside of frontier labs.
