Prime Intellect raises $130 million so enterprises can train their own agents
Nvidia, Intel and Dell just backed the counter-narrative: big companies may rather own their agents than rent them from the labs.
Photo: Risk and runner blocks on a blue surface. // GaaS News- Prime Intellect raised a $130 million Series A at a $1 billion valuation, announced July 8. Radical Ventures led the round.
- The strategic names are the story: Nvidia Ventures, Intel Capital, Dell Technologies Capital and Iconiq all participated.
- The company sells the full stack for training agentic models in-house: compute, a reinforcement learning framework, and evaluations.
- Prime Intellect says it is already at a $100 million annualized revenue run rate, with Ramp and Zapier among its customers.
Prime Intellect, a two-year-old startup that helps enterprises train their own AI agents, has raised a $130 million Series A at a $1 billion valuation. TechCrunch reported the round on July 8. Radical Ventures led, and the participant list reads like a hardware industry caucus: Nvidia Ventures, Intel Capital, Dell Technologies Capital and Iconiq, plus angels including Perplexity's Aravind Srinivas, Box's Aaron Levie, Harvey's Winston Weinberg and Cognition's Jeff Wang.
What Prime Intellect actually sells
The pitch is independence. Prime Intellect packages the compute, the reinforcement learning framework and the evaluation tooling a company needs to train agentic models on its own data, without leaning on OpenAI or Anthropic for the finished intelligence. CEO and co-founder Vincent Weisser put it bluntly to TechCrunch: "It shouldn't just be a few nerds in a glass tower in San Francisco that have the capability to train AI models."
The company told TechCrunch it has reached a $100 million annualized revenue run rate, a figure worth treating as a company claim until financials surface. Its customer list is verifiable and interesting: Ramp and Zapier both train agents on the platform. Ramp co-CEO Karim Atiyeh said the resulting model "beat the frontier models on accuracy while running at faster speeds and a fraction of the cost."
Why the hardware money matters
Nvidia, Intel and Dell writing checks into the same Series A is not a coincidence. Every enterprise that trains its own agents buys GPUs, networking and servers instead of routing that spend through a frontier lab's API. The chipmakers are funding demand for their own silicon, one training run at a time.
The rent-or-own question
Most of the agent economy this publication covers runs on rented intelligence: you buy an outcome from a vendor, and the vendor pays a lab per token. Prime Intellect is the other path. If a company can train an agent that beats the frontier on its own narrow workload, the margin that would have gone to the model provider stays home. David Katz of Radical Ventures told TechCrunch the team is "operating at the frontier" on exactly that stitched-together stack.
A $1 billion valuation at Series A says investors believe the own-your-agent market is real, and it lands the same week the cost of rented intelligence split sharply in two. For a primer on why that unit of sale matters, start with what GaaS is.
Sources: TechCrunch, PYMNTS, Seeking Alpha.