Prime Intellect, an open AI lab, said it has closed a $130 million Series A round, taking its total funding to more than $150 million. The company said the capital will be used to build an open superintelligence technology stack and push AI development toward a decentralized model.
Radical Ventures led the round
The Series A was led by Radical Ventures. Participants included NVIDIA Ventures, Intel Capital, Dell Technologies Capital, and several existing investors.
The investor group also included angel backers focused on frontier AI. Prime Intellect named Thinking Machines co-founder John Schulman, Ramp co-founder Karim Atiyeh, Box CEO Aaron Levie, and Cloudflare CEO Matthew Prince, along with founders and senior executives from Tesla, Harvey, Zapier, and LangChain.
The company said support from technical experts and venture arms tied to major hardware companies points to strong interest in its open AI model architecture.
Focus on reinforcement learning infrastructure
Prime Intellect said pretraining for frontier AI models has long been concentrated in a small number of major research institutions. As reinforcement learning becomes more widely used, companies are gaining the ability to steer model optimization themselves and train systems directly on their own product data for specific workflows.
The company’s stack covers compute, large-scale reinforcement learning, environment construction, sandbox testing, performance evaluation, and model deployment. Prime Intellect said the platform has attracted more than 6,000 customers and that annual recurring revenue surpassed $100 million within a year.
It also pointed to a case involving fintech company Ramp, which used the platform’s post-training technology to train a 35 billion-parameter model. According to the company, that model showed better performance and lower cost than closed models in certain real-world use cases.
Work expands to Recursive Language Models
Prime Intellect is also building out training capabilities for Recursive Language Models, or RLMs. The company said current language models tend to lose performance in long-context settings, while RLMs offer a scalable structure for long-duration tasks by managing context and coordinating sub-agents.
The development team is also building infrastructure around pretraining, autonomous NanoGPT, and continual learning for general agents. The aim is to integrate training and inference more tightly so models can improve automatically in production environments.
Open-source stance is central to the company’s pitch
In its technical manifesto, Prime Intellect said open-source systems can help stop AI from being controlled by a small group of institutions. The company argued that businesses and individuals should retain ownership and control over their data and models so the number and variety of models can better reflect real-world demand.
Prime Intellect said broader use of open technology stacks and self-improving agents could make AI development more transparent and more competitive through a distributed architecture.