Recursive Superintelligence Raises $500M to Automate AI Research

The AI industry's appetite for early-stage capital has produced a new benchmark: a 20-person team, four months old, valued at $4 billion.

Recursive Superintelligence, a London lab founded on December 31, 2025, announced this week it had raised $500 million in a round led by Google Ventures and NVIDIA, with the round reportedly oversubscribed and total interest approaching $1 billion. The founders include Richard Socher, formerly Salesforce's chief scientist and an early AI researcher at Stanford, alongside veterans from DeepMind and OpenAI.

The lab's stated goal is to remove humans from the frontier AI development pipeline. Where current AI labs rely on researchers to guide which experiments to run, what training data to select, how to evaluate results, and which research directions to pursue, Recursive Superintelligence wants AI systems to handle all of those decisions autonomously. It is building, in short, an AI researcher that designs and trains future AI researchers.

The idea of recursive self-improvement has circulated in AI research for years, mostly in theory. What is different here is that a credible team is attempting to build a working version on a commercial timeline. The lab has a public launch planned for mid-May 2026.

The round reflects something broader happening in AI investment. Valuations are decoupling from conventional metrics: revenue, users, a shipping product. Recursive Superintelligence has none of those yet. What it does have is a founding team with serious track records, backing from two of the most strategically positioned investors in the sector, and a thesis that, if it works, would change how frontier models are built.

Whether fully automated AI research is achievable at the pace the company is projecting, and whether it would be safe if achieved, are questions the lab has not yet addressed publicly. In the current funding climate, the ambition itself appears to be enough.