Prime Intellect, a San Francisco startup that sells compute and tooling for training custom AI agents, has raised a $130 million Series A at a $1 billion valuation. The round was led by Radical Ventures, with participation from Nvidia Ventures, Intel Capital, Dell Technologies Capital, and Iconiq, plus a roster of founder-angels including Perplexity's Aravind Srinivas, Box's Aaron Levie, Harvey's Winston Weinberg, Cognition's Jeff Wang, and Mercor's Brendan Foody, according to TechCrunch and PYMNTS.

From decentralized training to enterprise agents

Prime Intellect was founded in late 2023 by CEO Vincent Weisser and CTO Johannes Hagemann, initially raising a $5.5 million seed round in April 2024 to build a decentralized, crowdsourced compute platform for training open models. The company has since pivoted its pitch toward reinforcement learning (RL): giving enterprises the infrastructure to iteratively train and refine their own agentic models on proprietary data and tasks, rather than fine-tuning or prompting a general-purpose model from OpenAI or Anthropic.

The thesis, per the company's Series A announcement, is that companies increasingly don't want to hand proprietary workflows and data to a frontier lab, and that RL techniques — which reward a model for successfully completing a task and penalize failures — let a business become its "own AI lab" for narrow, high-value tasks.

Customers and traction

Prime Intellect says it has reached an annualized revenue run rate of roughly $100 million, with customers including fintech company Ramp and automation platform Zapier paying for a hosted version of its training stack. In one cited case, Ramp used Prime Intellect's tools to train a 35-billion-parameter model for spreadsheet search that outperformed Anthropic's Claude Opus on accuracy while running faster and cheaper than Claude Haiku — a proof point Prime Intellect is using to argue that smaller, task-specific agents trained with RL can beat general frontier models on narrow enterprise work.

Why it matters for the agent market

The round lands amid a broader shift in agentic AI spending: enterprises that spent 2025 experimenting with off-the-shelf agent demos are now being pushed toward tools that let them own and control the model powering a specific workflow. Prime Intellect's $1 billion valuation — up sharply from its 2024 seed stage — signals investor appetite for infrastructure plays that sit underneath, rather than compete directly with, the major model providers.