Proximal emerged from stealth on September 29 with a striking set of metrics: roughly ten months after launching it has reached a $200 million annualized revenue run rate and is now raising a $15 million seed round led by General Catalyst at a $300 million valuation.
Proximal’s business and customers
Proximal builds reinforcement learning environments and post-training data for AI coding agents — the systems that write, test and ship software with decreasing human intervention. Its product creates realistic practice problems and answer keys grounded in real codebases, and its customers are the frontier labs developing those agents. According to the company, enough labs were paying for its services to drive the $200 million annualized revenue figure in under a year.
Founders, funding and early terms
The founding team includes Justus Mattern, who led RL research and data work at Prime Intellect and co‑founded Revideo (a Y Combinator Summer 2023 company); Calvin Chen, who serves as CEO; and Navid Pour, who joined from Cursor. The $15 million seed round is led by General Catalyst, with participation from Scribble Ventures, SV Angel, Chemistry and Go Global Ventures. The company said it spent nothing on a launch party when it came out of stealth.
Expansion beyond software engineering
In its stealth announcement, Proximal said it is expanding beyond software engineering into domains such as targeted drug discovery, chip design and rewriting legacy software. The company framed this expansion as applying the same reinforcement learning approach — build realistic environments for models to practice in — to other technical domains.
Valuation, market position and risks
A $300 million valuation on a $15 million seed is relatively modest given a $200 million run rate, the company noted, a gap that the report suggests could reflect investor concerns about customer concentration or simply that the founders did not need to push for a higher price. Proximal’s traction highlights a broader dynamic in AI infrastructure: while compute and chips have dominated headlines, high-quality training environments and domain-specific data for reinforcement learning are scarce inputs that some specialist vendors now sell to frontier labs.
The company faces the common risk for training-data vendors: customers may eventually choose to build the same capability in‑house. Proximal’s current revenue run rate gives it time and leverage to diversify beyond coding, but whether drug design and chip design turn into material revenue lines or remain aspirational is the key question going forward.
