Prime Intellect · They are building the part of the stack most people cannot see — where a model is trained, given an environment to act in, and graded — and they publish the engineering behind it in the open, which makes them a primary source for us whether or not the relationship goes anywhere.

The whole loop for training a model on your own workload and keeping it improving — compute (on-demand GPUs and clusters), large-scale reinforcement learning, environments and sandboxes for agents to act in, evaluations, and inference to serve the result. They train open frontier models themselves (INTELLECT-1/2/3, SYNTHETIC-1/2, Prime Agent) and sell the same stack. Their own phrase for it is "the open superintelligence stack", and the product nav is four numbered rungs: TRAINING 01 · INFERENCE 02 · COMPUTE 03 · RESEARCH 04.

Sector

GPU compute

Stage

Series A

On the roster since

15 Sept 2026