Recursive Superintelligence emerged from stealth on 13 May with a $650 million Series A round at a $4.65 billion valuation — one of the largest debut funding rounds in AI history. The round was led by Alphabet's GV fund and Greycroft, with additional backing from Nvidia and AMD's venture capital arm. The company operates from offices in San Francisco and London.
The startup was founded earlier in 2026 by Richard Socher, the former Chief Scientist at Salesforce, alongside a team drawn from OpenAI, Google DeepMind, Meta AI, and Uber AI. Starting with Socher and six staff members, the company has expanded to over 25 researchers and engineers, with a public launch targeted for mid-2026.
The core thesis is provocative: rather than improving AI models through human-guided training and reinforcement learning — the approach used by every major AI lab — Recursive Superintelligence is building AI systems that can autonomously improve their own codebase. The system will conduct automated scientific discovery through simulations, developing experiment ideas, testing them, and validating the results without human intervention. The model searches for improvements across multiple dimensions simultaneously: code, auxiliary programmes, and training and inference infrastructure.
Socher's vision extends beyond AI research. He envisions the self-improving systems eventually expanding into physics, chemistry, and particularly pre-clinical biology — domains where the pace of experimental discovery is currently bottlenecked by human researchers' capacity to design, run, and interpret experiments.
The company has not disclosed what specific machine learning methods will power its self-improving AI, maintaining technical secrecy around its approach. It is not alone in this pursuit — competitor Ineffable Intelligence is using reinforcement learning for similar recursive improvement goals — but the scale of Recursive's funding and the calibre of its founding team set it apart.
For context engineers, Recursive Superintelligence represents the most ambitious and well-funded attempt at a goal that has been discussed theoretically for decades: AI systems that can improve themselves without human guidance. The safety implications are significant — recursive self-improvement is precisely the capability that AI safety researchers have long identified as a potential inflection point. Whether Recursive's approach produces genuine self-improvement or hits the same diminishing-returns ceiling that has limited previous attempts will be one of the most consequential technical questions in AI over the coming year. The $4.65 billion valuation on zero revenue suggests that investors are betting the answer will reshape the industry.
