Getting to $100 million in annual revenue before your second birthday is rare. Doing it while building infrastructure most enterprises are still figuring out how to use—rarer still.
Prime Intellect, the San Francisco startup that wants to become the default platform for companies training their own AI agents, announced a $130 million Series A on July 8 that values it at $1 billion post-money. Founded in 2024 by Vincent Weisser and Johannes Hagemann, the company has bet its future on a thesis that sounds obvious in hindsight but remains unproven at scale: Organizations will increasingly want to own the entire loop of training, deploying, and continuously improving their AI models, rather than relying on OpenAI or Anthropic's black boxes.
Whether that thesis holds may determine if Prime Intellect becomes essential infrastructure or an expensive detour.
Radical Ventures led the round, joined by the venture arms of NVIDIA, Intel, and Dell Technologies, along with Iconiq. More telling, perhaps, is the collection of operator-angels who participated—over a dozen, including John Schulman of Thinking Machines, Ramp's Karim Atiyeh, Box CEO Aaron Levie, and Cloudflare's Matthew Prince. When founders writing checks spans fintech, devtools, security, and frontier AI research, something's resonating.
The company had previously raised $15 million in February 2025 from Founders Fund and a $5.5 million seed in April 2024 co-led by Distributed Global and CoinFund, bringing total capital raised north of $150 million.
The Stack, Explained (Sort Of)
What Prime Intellect actually does requires unpacking several layers of AI infrastructure jargon. The company describes its offering as "the open superintelligence stack"—compute infrastructure, reinforcement learning post-training, environments for testing, evaluation frameworks, sandboxes, and inference capabilities all rolled together. Think of it as the picks-and-shovels play for the current wave of agentic AI, where models don't just respond to prompts but take actions, improve themselves, and require substantially more complex infrastructure to train and deploy safely.
In May 2026, Prime Intellect launched its Lab platform, which offers hosted reinforcement learning training—the technique behind models that learn by trial and error rather than memorizing patterns. The company also maintains open-source libraries including prime-rl for distributed RL training and tools for managing evaluation environments. It's a hybrid business model: open-source to build community and credibility, paid platform for companies that want someone else handling the complexity.
Dell Technologies Capital, in its investment memo, pointed to "strong customer signal during diligence" and emphasized growing demand for companies to control their own model optimization. Prime Intellect reports more than 6,000 customers as of July 2026, including Ramp, Zapier, and Perplexity, among others.
Who's Building This
Weisser's path to AI infrastructure came through decentralized science, including work with VitaDAO, which funds longevity research through blockchain-based governance. It's an unconventional résumé for someone now building systems to train reinforcement learning models at scale, but perhaps that outsider perspective helped. His co-founder Hagemann has more traditional credentials: distributed training frameworks at Aleph Alpha, the German AI company that's been positioning itself as Europe's answer to OpenAI.
The founding team's blend of decentralized-tech idealism and hardcore distributed systems experience shows up in the product philosophy. Prime Intellect isn't just selling cloud credits for model training—it's betting that openness and customizability will matter more than convenience as AI systems become more capable and consequential.
The Broader Context
David Katz, a partner at Radical Ventures, framed Prime Intellect as a "one-stop shop" for achieving frontier-level AI capability without frontier-level budgets. Intel Capital struck a different note, describing it as "the default open stack for training and deploying self-improving agents" and pointing to reinforcement learning infrastructure as a core enterprise need heading into 2027.
Both framings sidestep an uncomfortable reality: We don't yet know which architectures, training methods, or deployment patterns will dominate the next phase of AI development. Prime Intellect is making a calculated bet that RL-trained agents represent a sufficiently large and durable market to build a billion-dollar business around. The customer list suggests early traction. The company reports reaching a $100 million annualized revenue run-rate in under a year—a figure that, while unaudited, implies real willingness to pay if accurate.
But hyperscale cloud providers aren't sitting still, and frontier labs are racing to make their own agent platforms. Prime Intellect's window to establish itself as the default open alternative may be narrower than the funding round suggests.
The company says it will use the capital to continue building across its platform—compute, RL post-training, sandboxes, inference, the works—and is hiring across reinforcement learning, inference, distributed systems, and infrastructure roles. Translation: They're staffing up for a land grab while the category is still being defined.
Whether "open superintelligence" becomes a category or a footnote depends on execution from here. The capital and backing give Prime Intellect room to find out.
