NeoLabsAn index of research-first AI labs
Home / Labs / Recursive Superintelligence
Self-improving AI & automated research · San Francisco, US · founded 2026

Recursive Superintelligence: who founded it, who funds it

Recursive Superintelligence is a San Francisco self-improving AI lab founded in 2026 by Richard Socher, Yuandong Tian, Tim Rocktäschel, Jeff Clune and 4 others, who came from Salesforce, Meta, DeepMind and OpenAI. What it is building: aI that automates model architecture, training, evaluation and research direction: recursive self-improvement. It has raised $650M in disclosed rounds, most recently a Seed of $650M at a $4.65B valuation in May 2026, led by GV (Google Ventures) and Greycroft. It has not released anything outsiders can use yet.

Founders
Richard Socher (ex-Salesforce), Yuandong Tian (ex-Meta), Tim Rocktäschel (ex-Google DeepMind), Jeff Clune (ex-Google DeepMind), Caiming Xiong (ex-Salesforce), Alexey Dosovitskiy (ex-Google DeepMind), Josh Tobin (ex-OpenAI), Tim Shi (ex-Cresta)
Founders from
ex-Salesforce, ex-Meta, ex-Google DeepMind, ex-OpenAI, ex-Cresta
Location
San Francisco, US (North America)
Valuation
$4.65B set by the last priced round
Raised
$650M disclosed
Investors
GV (Google Ventures) lead, Greycroft lead, Nvidia, AMD
What ships
Research only
Notes
Committed $410M of the raise to a multi-year AWS compute deal (Jul 2026).

Rounds

DateRoundAmountValuationLeadSource
May 13, 2026Seed$650M$4.65BGV (Google Ventures), Greycroftsiliconangle.com

Sources

News

Investors who back other labs here

  • Nvidia also: Safe Superintelligence, Reflection AI, Skild AI, Poolside
  • AMD also: Thinking Machines Lab, Hark, World Labs, Luma AI
  • GV (Google Ventures) also: Humans&, Isomorphic Labs

Same research area