Recursive Superintelligence Raises $500M at $4B Valuation — Just Four Months After Founding

Summary

London-based AI startup Recursive Superintelligence has closed at least $500 million in funding at a $4 billion pre-money valuation, with reports suggesting the oversubscribed round could reach $1 billion. The round was led by Google’s venture arm GV, with significant participation from Nvidia.

What makes this remarkable: the company was incorporated just four months ago, in December 2025, and has a team of roughly 20 people. It was founded by Richard Socher (former Salesforce chief scientist) alongside researchers from DeepMind and OpenAI, including UCL professor Tim Rocktäschel, Josh Tobin, Jeff Clune, and Tim Shi. A public launch is expected around mid-May 2026.

The company’s stated mission is to build self-improving AI systems that can automate the entire AI development pipeline — from evaluation and data selection through training, post-training, and research direction — without human intervention. Socher describes this as the “third and perhaps final stage of neural networks.”

Source

The Decoder — Recursive Superintelligence raises $500M
MLQ.ai — Funding Details

Commentary

A $4 billion valuation for a four-month-old company with 20 employees building “self-improving AI” is either the most exciting or the most alarming sentence in AI right now, depending on where you sit. The team’s pedigree is undeniably stacked — Socher, Clune, and Rocktäschel are serious researchers, not hype merchants — but the mission itself is audacious to the point of raising safety eyebrows.

This also reflects the broader AI funding frenzy: Q1 2026 saw $297 billion in global startup funding, with 81% flowing to AI. At some point the question shifts from “can we build self-improving AI?” to “should we, and who’s watching?” The fact that Google and Nvidia are backing this bet tells you where the industry’s center of gravity is moving. Keep an eye on the mid-May launch.