Prime Intellect: Building the Open Super Intelligence Stack

Building the Open Superintelligence Stack

We first heard of Prime Intellect through the AI researcher community; researchers loved their full-stack and open approach to AI infrastructure and research. After spending some time with Vincent Weisser, the founder, at NeurIPS we realized that they were seeing the future well before the rest of the market and executing towards it – they believed that intelligence should be owned, not rented and that frontier AI should be built in the open. They saw post-training and reinforcement learning as the first pillars in that vision and were building primitives to democratize access to it.

Today, Vincent, co-founder Johannes Hagemann, and the company announced a $130 million Series A funding to support the open superintelligence stack they’re building. Here's why we invested alongside friends at Radical Ventures, NVIDIA Ventures, Intel Capital, and Iconiq.

Why we’re investing

The ability to own the AI model development loop, to train on real product signal, iterate fast, and compound performance with every run, has until recently been the exclusive domain of a handful of frontier labs. Everyone else rented access to models they couldn't improve and infrastructure they didn't control. Vincent and Johannes are ending that arrangement with Prime Intellect. To do so, they built a strong team spanning frontier research and infrastructure, a combination rare to see outside of the handful of closed model labs.

The shift that made this possible

Pre-training rewarded scale, and scale rewarded incumbents. That's changing. Advances in RL, combined with growing demand for data sovereignty, mean it now makes sense for companies to own their model optimization loop, train on their own data, and build agents that improve directly against their own outcomes. The company that owns its model, its data, and its results holds a durable advantage. We already saw this direction for select AI-native startups – Cursor realized they could no longer build on closed models and built their own Composer stack, Cognition went the same route. We learned from developers that there is a lot of appetite to do this, but it’s still very hard to do. Prime Intellect is the infrastructure to make intelligence ownership possible easily and at scale.

Strong, repeatable customer signal

During diligence, we met several technically sophisticated teams that chose Prime Intellect. They ran demanding workloads with PI and in short order, came back to run more. That repeat behavior, combined with more than $100 million in annualized revenue reached in under a year, reflects real utility. Sophisticated customers like Perplexity and Ramp were using the stack to do frontier AI research. Add that to the emphatic product love we heard from the researchers and developers, and it’s clear that something durable is being built here.

That signal is being repeated at the market level as well. Two years ago, this market was effectively zero. It’s now into the several billion dollars a year range. We anticipate it growing 10x over the next three to five years as every company becomes an AI company, and the ones that own their learning loop will pull ahead.

Compute and talent shortages that aren’t going away

Compute supply is still nowhere near demand, and the people who know how to run frontier-scale post-training and RL are just as scarce. Most teams don't have a research organization in-house or a bench of ML specialists to draw from. Prime Intellect addresses both: large-scale clusters on demand, plus a full-stack abstraction layer spanning reinforcement learning, environments, sandboxes, evaluations, and deployment that handles the hardest parts of model optimization. Importantly, the company is building this infrastructure in the open, giving customers (initially neolabs and startups, but now increasingly enterprises too) an alternative to closed platforms at a time when more teams want to own critical AI capabilities rather than rent them.

Why Prime Intellect

Prime Intellect is racing the most well-funded labs in the world and doing it in the open.

We believe the next wave of AI leaders will be the companies that own their models, their data, and the systems that continuously improve from real-world outcomes. Making that possible requires a new infrastructure layer—one that combines compute, post-training, evaluation, and deployment into a single stack while remaining open and accessible to builders.

That's what Prime Intellect is building. We're proud to partner with Vincent, Johannes, and the entire team as they help make frontier AI ownership available to far more organizations around the world.

‍