Prime Intellect raises $130M Series A for decentralized training and RL environments

The round closed July 8 at a $1 billion valuation, with NVIDIA Ventures, Intel Capital, Dell Technologies Capital and ICONIQ also participating. Total funding now exceeds $150M. The company serves over 6,000 enterprises and startups, including Ramp and Zapier, and reports annualized revenue above $100M. Proceeds will expand compute clusters and reinforcement learning work on long-horizon agents, recursive language models and continual learning. The product is a modular stack covering compute access, environments, sandboxes, evaluations, deployment and large-scale reinforcement learning. Ramp trained a 35-billion-parameter spreadsheet agent that beat a leading frontier model on accuracy, ran 27% faster and cost less. Rivals include frontier labs' own agent-building tools and infrastructure startups such as Together AI. For founders, the pitch is that small models post-trained on a company's own workflows can beat general-purpose APIs on cost and speed. That would reduce dependence on any one lab. Three hardware vendors backing the round suggests they expect post-training to become a major source of compute demand. The revenue figure is self-reported and has not been independently verified.