What is Recursive Superintelligence?

On May 14, 2026, Recursive Superintelligence officially launched — a startup founded by Meta FAIR former Research Director Tian Yuandong and "Prompt Engineering之父" Richard Socher, joined by former researchers from OpenAI, Google DeepMind, and Meta [citation:2].

The company raised $6.5 billion at a $46.5 billion valuation from investors including Google Ventures, Greycroft, AMD Ventures, and NVIDIA Capital [citation:2].

The Mission

Recursive targets "Recursive Self-Improvement" — building AI systems capable of discovering knowledge and optimizing themselves autonomously. Tian Yuandong states: "We are building a system that can automatically discover knowledge and recursively improve itself" [citation:2].

Founding Team

  • Richard Socher (CEO): "Prompt Engineering之父," former Stanford AI researcher, 240,000+ citations, founder of MetaMind and http://You.com [citation:2]
  • Tian Yuandong: Meta FAIR former Research Director (nearly 10 years), Shanghai-born, CMU PhD, NeurIPS/ICML Area Chair, deep involvement in Llama series [citation:2]
  • Tim Rocktäschel: DeepMind Open-Ended Intelligence Director, UCL Professor
  • Josh Tobin (CTO): Former OpenAI Agent researcher
  • Alexey Dosovitskiy: Vision Transformer (ViT) researcher
  • Peter Norvig (Advisor): Google former Research Director, author of "Artificial Intelligence: A Modern Approach"
  • Shi Tianlin (Tim Shi): Tsinghua Yao Class, former OpenAI researcher, Cresta co-founder
  • Caiming Xiong: Salesforce AI former SVP (nearly 10 years)

Technical Direction

Recursive focuses on "Open-endedness" and "Recursive Self-Improvement" — AI that can:

  • Generate new ideas continuously
  • Verify results autonomously
  • Adjust its own capabilities
  • Iterate over extended timeframes [citation:2]

As Socher notes: "AI is itself code, and now AI can write code." The team believes the industry is approaching AGI, but remains significantly distant from "superintelligence."

Initial Focus

Recursive first targets "AI improving AI" — automating the research and development process for AI itself. Future applications include scientific discovery and drug development. Safety is prioritized with systems designed to enhance capabilities while reducing risks [citation:2].

Pricing

Not yet applicable — company is in stealth/research phase. Offices in San Francisco and London, team currently 25+ people [citation:2].

Pros

  • Dream team of AI researchers from OpenAI, DeepMind, Meta, Google
  • $6.5B funding — one of largest AI startup launches in history
  • Focus on recursive self-improvement is unique and ambitious
  • Strong safety commitment
  • Industry-leading technical talent

Cons

  • No product yet — purely research stage
  • Recursive self-improvement is theoretical, unproven
  • Massive funding creates high expectations
  • Competition from DeepMind, OpenAI on similar goals

Who Should Watch?

Perfect for: AI researchers, investors, and technologists tracking the frontier of AI development toward superintelligence.

Verdict

Recursive Superintelligence has assembled the most impressive AI founding team since OpenAI. Whether recursive self-improvement works is an open question — but this is the team most likely to figure it out [citation:2].

Rating: N/A (Pre-product) - The one to watch.