Recursive Superintelligence Banks $500M From GV and Nvidia at $4B Valuation, Four Months In

Recursive Superintelligence, a stealth-mode AI lab founded just four months ago, has raised at least $500 million at a $4 billion pre-money valuation in a round led by Google's venture arm GV with participation from Nvidia. The round was reportedly oversubscribed and is part of a broader effort to raise up to $1 billion, making it one of the fastest-moving early-stage AI deals on record.

The founding team is unusually deep for a four-month-old company. The lab was co-founded by Richard Socher, the former chief scientist at Salesforce and founder of search startup You.com, alongside Tim Rocktäschel, a UCL professor and recent director at Google DeepMind, and former OpenAI researchers Josh Tobin, Jeff Clune, and Tim Shi. The pedigree alone explains how a pre-product company convinced two of the largest strategic investors in AI to write nine-figure checks.

The technical thesis is the most ambitious version of the autonomous-research bet currently in the market: build AI systems that improve themselves without human intervention, automating the entire frontier development pipeline — evaluation, data selection, training, post-training, and even research direction. If it works, the company believes it can compress the cycle that currently has thousands of researchers and engineers tuning each new generation of frontier models. If it does not, it joins a long list of self-improving-AI projects that ran into the wall of compounding errors.

The deal is the second European-flavored mega-seed in days, after AlphaZero creator David Silver's Ineffable Intelligence closed $1.1 billion. Together, the two rounds signal that researcher-founder labs targeting superintelligence — the layer above today's frontier models — are now a recognized investment category, with GV and Nvidia underwriting both bets. A public launch from Recursive Superintelligence is expected around mid-May 2026.