Prime Intellect Raises $130M at a $1B Valuation to Let Companies Train Their Own Agents

Radical Ventures led the round for the startup whose reinforcement-learning stack turns enterprises into model trainers — with Nvidia, Intel and Dell joining and revenue already running past $100 million a year.

Jul 9, 2026Source: TechCrunch

Prime Intellect Raises $130M at a $1B Valuation to Let Companies Train Their Own Agents

The company that made its name training AI models across volunteers' scattered GPUs is now selling that capability to the enterprise. Prime Intellect has raised a $130 million Series A at a $1 billion valuation, led by Radical Ventures, the startup said on July 8. Nvidia Ventures, Intel Capital, Dell Technologies Capital and Iconiq joined — a strikingly strategic, hardware-heavy investor list.

The pitch

Prime Intellect frames its product as an "open superintelligence stack": a training platform called Lab, a reinforcement-learning framework (prime-rl) that spreads training across thousands of GPUs, an Environments Hub of user-built RL sandboxes, evaluation tools, and a compute marketplace that auctions clusters across 50-plus data centers. The modular idea: enterprises assemble the pieces to reinforcement-train their own agent models rather than renting intelligence from a frontier lab. "It shouldn't just be a few nerds in a glass tower in San Francisco that have the capability to train AI models," said co-founder and CEO Vincent Weisser.

Real revenue, fast

The traction is what stands out. Prime Intellect is running at more than $100 million in annualized revenue in under a year, with about 6,000 customers including Ramp and Zapier. Ramp used the stack to RL-train a "Fast Ask" spreadsheet subagent that its co-CEO Karim Atiyeh said "beat the frontier models on accuracy while running at faster speeds and a fraction of the cost." Radical's David Katz said the team is "operating at the frontier in a way that's affordable."

From decentralized to sovereign

Founded in 2024 by Weisser and Johannes Hagemann, Prime Intellect rose to prominence with decentralized training runs — INTELLECT-1 (10B, trained across three continents in late 2024) and INTELLECT-2 (32B, decentralized RL, 2025). Notably, its most recent open model, the 106B INTELLECT-3, was trained not on a permissionless network but on a conventional cluster of 512 Nvidia H200s. Paired with a Series A led by chipmakers, the shift signals a pivot from the crypto-tinged "decentralized AI" story toward "sovereign enterprise AI." Earlier funding included a $15M round led by Founders Fund and a $5.5M seed.

The bet

New money goes toward more compute, longer-horizon agents and continual-learning infrastructure — the plumbing for companies that want to own their models outright.

Why It Matters

Prime Intellect is monetizing a thesis that cuts against the frontier labs: that the durable enterprise value is in the training stack, not any single model, letting companies build cheaper task-specific agents that beat general models on their own workflows. That a chipmaker syndicate — Nvidia, Intel, Dell — is funding it signals hardware vendors want to seed demand for training capacity they can sell into, independent of OpenAI and Anthropic.

Filed underAI Agents AI Research #AI funding #Prime Intellect #Radical Ventures #reinforcement learning