The archive · AI & Models · Strategic decision · 2024–2026
Prime Intellect bets enterprises become their own AI labs: $130M at $1B
Decentralized-training startup Prime Intellect pivoted to selling enterprises a full agent stack, hitting $100M run rate.
Prime Intellect
What the business is
Prime Intellect sells a modular "full stack" — compute access, a reinforcement learning framework, and evaluation tools — for companies to build their own AI agents.
Starting capital：$5.5M seed (2024), $15M led by Founders Fund (2025), $130M Series A at $1B led by Radical Ventures (July 2026)
How it started
Vincent Weisser and Johannes Hagemann founded Prime Intellect in 2024 as a decentralized AI company, raising $5.5M from CoinFund, Distributed Global, and Protocol Labs to aggregate distributed GPU supply for training models.
What happened
The company evolved its early decentralized-training platform into a hosted marketplace selling compute, a reinforcement learning framework, and evaluation tools to enterprises. Customers such as Ramp and Zapier paid for hosted tools; Ramp said an agent it built with Prime Intellect beat frontier models on spreadsheet-answer accuracy at a fraction of the cost.
How it ended up
By July 2026 the startup had reached a $100M annualized run rate and raised a $130M Series A led by Radical Ventures at a $1B valuation, with Nvidia, Intel, and Dell capital participating. It is still scaling.
Background
Prime Intellect started in early 2024 as a decentralized AI research company. Co-founders Vincent Weisser and Johannes Hagemann raised $5.5M from CoinFund, Distributed Global, and Protocol Labs to train models over distributed GPUs and build an open protocol for collective AI ownership.
Two years later it was a different business. Rather than only crowdsourcing model training, Prime Intellect began selling enterprises a hosted "full stack": compute access, a reinforcement learning framework, and evaluation tools for building agents on their own data.
The shift caught a wave of enterprise distrust of frontier labs. Customers did not want to hand proprietary data to OpenAI or Anthropic or depend on models that could be withdrawn, so they used Prime Intellect to build agents in-house — Ramp reported an agent that beat frontier models on spreadsheet-answer accuracy at lower cost.
By July 2026 Prime Intellect had customers including Ramp, Zapier, and Flapping Airplanes, a $100M annualized run rate, and a $130M Series A led by Radical Ventures at a $1B valuation, with Nvidia, Intel, and Dell among the backers.
What has to be true
- Frontier labs scared their own customers: data-control worries and sudden model shutdowns pushed enterprises to seek independent AI stacks.
- Reinforcement learning matured enough that fine-tuning for a specific business task no longer required a frontier-lab-sized team.
- Its decentralized-training origin gave Prime Intellect rare credibility for running its own training at far below big-lab cost.
- A modular marketplace avoided the all-or-nothing lock-in that enterprises resisted, easing adoption.
What can be applied
A research project can monetize by handing its hardest infrastructure to customers who fear depending on the biggest labs.
Aftermath
As of the July 2026 TechCrunch report, Prime Intellect was generating about $100M in annualized revenue, serving Ramp, Zapier, and Flapping Airplanes through hosted tools, and expanding its agent stack. The Series A at a $1B valuation positioned it as a leading independent alternative for enterprises that want to own their AI, with Weisser framing the mission as putting model-building capability beyond a few San Francisco labs.
Sources
- Prime Intellect raises $130M Series A to help enterprises build their own AI agents
- CoinFund and Distributed Global lead $5.5 million seed round for decentralized AI firm Prime Intellect
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card