Prime Intellect raises $130 million to help enterprises train their own AI agents

Prime Intellect raised a $130 million Series A led by Radical Ventures, the company said on July 8, 2026. The company sells enterprises the compute and tooling to train their own AI agents rather than rent them from a model provider.

NVIDIA Ventures, Intel Capital, Dell Technologies Capital and existing backers joined the round, which brings total funding above $150 million, Prime Intellect said in its funding announcement. The chip and hardware venture arms lining up behind a company that helps customers train models in-house is itself a signal of where compute demand is heading.

Prime Intellect said it now has more than 6,000 customers and an annualized revenue run rate above $100 million, up from near-zero a year earlier. Those figures are the company’s own and were not independently audited.

The pitch is that enterprises can build smaller, cheaper models tuned to their own work. Prime Intellect cited customer Ramp as having trained a 35-billion-parameter model on its stack that the company says outperformed Anthropic’s Claude Opus on spreadsheet-search tasks while running faster and cheaper than Claude Haiku. That is a vendor-supplied comparison and has not been independently verified.

The funding rides a broader bet that many companies will train specialized models instead of paying per token for a frontier system. The unverified run-rate and benchmark claims are the main caveats: strong growth from a low base is common, and self-reported model comparisons rarely survive third-party testing unchanged.

Whether the run rate holds above $100 million into 2027 will show if the do-it-yourself training pitch has staying power.