Prime Intellect just landed $130 million to prove that enterprises can build their own AI agents without renting one from OpenAI or Anthropic.
The San Francisco startup closed its Series A this week at a $1 billion valuation, according to TechCrunch, which first reported the deal. Radical Ventures led the round. Nvidia Ventures, Intel Capital, Dell Technologies Capital and Iconiq all joined in, and so did a bench of operator angels: Perplexity's Aravind Srinivas, Box's Aaron Levie, Harvey's Winston Weinberg, Cognition's Jeff Wang, Mercor's Brendan Foody. That's not a typical venture syndicate. It's closer to a customer list with checkbooks.
Prime Intellect was built by CEO Vincent Weisser to do one thing: give companies the full stack needed to train, evaluate and deploy their own AI agents, rather than leasing a finished one from a foundation model lab. The platform bundles compute access, a reinforcement learning framework and evaluation tooling into what the company describes as a modular marketplace. Customers pick the pieces they want and skip the rest, with no requirement to commit to a single vendor's stack end to end.
That pitch already has revenue behind it. Prime Intellect says it's running at $100 million in annualized revenue, a figure that would make it one of the faster-growing infrastructure plays in the current AI cycle. "It shouldn't just be a few nerds in a glass tower in San Francisco that have the capability to train AI models," Weisser said, framing the company's mission as opening frontier grade training tools to businesses with no interest in building their own research lab.
Ramp is one of the customers making that case concrete. Karim Atiyeh, the fintech company's co-founder, told TechCrunch that models built on Prime Intellect's stack "beat the frontier models on accuracy while running at faster speeds and a fraction of the cost." Zapier is also a customer, according to the report. For a company two years removed from its seed round, landing production use at two well known operators is the kind of proof point investors chase harder than any pitch deck.
Frankly, the timing is the real story here. Enterprise agent infrastructure has kept pulling nine and ten figure rounds through 2026 even as public markets grow twitchy about whether AI capital spending is outrunning actual returns. Investors aren't slowing down on the picks and shovels layer, they're getting choosier about which shovel they back. Radical Ventures didn't frame its bet as a hedge. A partner at the firm told TechCrunch that Prime Intellect has "stitched this together and built it in such a way that they're operating at the frontier in a way that's affordable," which is a specific claim about cost structure, not a vague nod to potential.
The harder question is whether Prime Intellect's approach survives contact with OpenAI and Anthropic, both of which are pushing their own agent tooling directly at enterprise customers. Prime Intellect's answer is architectural. It doesn't own a single frontier model that it needs you to rent. It sells the rails, compute, reinforcement learning, evaluation, and lets a company plug in whichever base model fits, then train and tune agents against it. That's a genuinely different bet than a model lab selling you its own agent product built on its own model.
Whether that's a durable moat or just good timing is still an open question. Nvidia Ventures and Intel Capital didn't write checks here purely as financial investors. Both have a stake in AI training staying distributed across many compute providers rather than consolidating inside two or three closed labs. That alignment, chipmakers backing an open agent training layer, tells you as much about where this money is headed as anything in Prime Intellect's pitch deck.
$130 million and a $1 billion valuation buy Prime Intellect time. What they don't buy is certainty that enterprises keep choosing to build over renting once OpenAI and Anthropic sharpen their own enterprise agent offerings. That fight is just getting started.
Also read: China Will Let Its Top AI Firms Buy Nvidia's H200 Chips Again • Metal Turns Fundraising From a Networking Problem Into a Data Problem • Iluvatar CoreX seeks up to $850 million just as its IPO lockup expires
