Proximal Exits Stealth With $15M Seed at a $300M Valuation

Proximal, a New York City-based AI coding data startup founded in 2025, is emerging from stealth with $15 million in new funding at a $300 million valuation.

The round is a seed, led by General Catalyst. Additional investors participating include SV Angel, Go Global Ventures and Chemistry, alongside individual investors Liam Fedus, Kevin Weil and Erik Bernhardsson. The company lists its funding date as September 30, 2026, and categorizes itself under AI infrastructure software.

At the center of the raise is a narrow but increasingly contested layer of the AI stack: the post-training feedback loop. Proximal's platform is built to automate that loop across complex operational domains. It processes unstructured real-world data and AI agent execution traces to construct evaluation environments, which the company says are used to identify model reasoning failures and generate targeted datasets for performance optimization.

In plain terms, the company is selling the apparatus that sits between a model's raw output and its next round of improvement — evaluation environments and the data that comes out of them. That places Proximal in the same conversation as a growing field of startups selling data, evaluation and post-training infrastructure to AI developers, rather than attempting to build frontier models themselves.

The new capital will go toward expanding Proximal's engineering and research teams and scaling its post-training and data research infrastructure. The company also intends to accelerate expansion into new industry areas, including computational drug discovery, custom semiconductor design, and legacy software refactoring for critical public infrastructure.

Those three target areas share a common trait: high-stakes, technically dense work where correctness matters and where unstructured operational data is plentiful but rarely organized into something a model can learn from. Extending from coding into drug discovery and chip design suggests Proximal sees its evaluation-and-dataset machinery as domain-agnostic, rather than tied to software development alone. Legacy software refactoring for public infrastructure pushes in the opposite direction — deeper into the coding world, but aimed at systems where failure carries outsized consequences.

The $300 million valuation against a $15 million seed is a notable ratio for a company that only emerged from stealth this month, and it reflects how aggressively investors are currently pricing picks-and-shovels infrastructure for AI training and evaluation. General Catalyst's lead position, paired with a list of well-known individual backers, gives the young company both institutional capital and access to operators in the AI ecosystem.