Recursive Superintelligence
Category: Foundation Models / LLMs
Building recursive self-improving superintelligence — AI systems that autonomously improve their own code, architecture, and research capabilities through open-ended algorithms. Recursive Superintelligence was founded in 2025. The company is led by Richard Socher. Based in London, United Kingdom; San Francisco, United States. Team size: 25-30. Total funding raised: $650M. Latest round: Series A. Key investors include GV (Google Ventures); Greycroft; NVIDIA; AMD Ventures.
- Founded
- 2025
- Headquarters
- London, United Kingdom; San Francisco, United States
- Team size
- 25-30
- Total funding
- $650M
Value proposition
Recursive Superintelligence is building AI that can improve itself — discovering its own limitations, designing experiments, writing benchmarks, and rewriting its own codebase to become more capable, without human intervention. The company intervention. The goal is to automate the entire AI research and development pipeline, starting with AI research itself and expanding to all scientific disciplines.
Products and solutions
Recursive self-improving AI system (in research phase); Level 1 autonomous training system (targeting mid-2026 launch); Open-ended architecture for automated scientific discovery; AI that improves AI codebase, harness, training, and inference infrastructure
Unique value
The only AI company pursuing recursive self-improvement as the next scaling law, with a founding team of 8 world-class researchers who collectively pioneered Vision Transformers, open-ended evolution algorithms, rainbow teaming for AI safety, and self-improving coding agents — assembled at a density unmatched in the industry.
Target customer
AI research labs, scientific research institutions, pharmaceutical/biotech companies, materials science organizations, and any enterprise needing accelerated scientific discovery
Industries served
AI Research">AI research; Scientific discovery (physics, chemistry, pre-clinical biology); Drug discovery; Battery materials; Fusion physics
Technology advantage
Open-ended algorithm architecture inspired by biological and cultural evolution; AI systems that conduct experiments to safely improve their own capabilities; automated red teaming and capability discovery; self-improving coding agents; foundational world models; quality diversity algorithms; AI-generating algorithms; retrieval-augmented generation; AI scientists
How they differentiate
Unlike competitors focused on specific techniques (RL for Ineffable Intelligence, world models for AMI Labs, safety for SSI), Recursive is attempting to automate the entire AI development pipeline itself — building systems that improve their own code, architecture, training methods, and research direction without continuous human oversight. The company's open-ended evolution approach is inspired by biological and cultural evolution rather than scaling laws.
Main competitors
Ineffable Intelligence (David Silver's startup, $1.1B seed at $5.1B valuation, RL-based approach); AMI Labs (Yann LeCun's startup, $1B raised, world models approach); Safe Superintelligence / SSI (Ilya Sutskever's startup, safety-first superintelligence)
Key partnerships
GV (Google Ventures) — lead investor; Greycroft — co-lead investor; NVIDIA — strategic investor and compute partner; AMD Ventures — strategic investor
Notable customers
Not yet publicly disclosed (company still in stealth/pre-launch phase; public launch targeted mid-2026)
Major milestones
Incorporated December 31, 2025 in London; Raised $500M+ at $4B valuation in April 2026 (first reported); Emerged from stealth May 13, 2026 with $650M at $4.65B valuation; 8 world-class co-founders publicly founders announced; Targeting mid-2026 public launch with Level 1 autonomous training system
Growth metrics
Grew from 7 to 25-30 employees in ~5 months; valuation increased from $4B (pre-money April 2026) to $4.65B (May 2026)
Market positioning
Premium-tier AI research lab positioned as the most ambitious bet on recursive self-improvement as the next scaling law after pre-training. Valued at $4.65B at launch — among the highest-valued early-stage AI startups ever. Part of a structural wave of top AI scientists leaving Big Tech to found their own labs.
Geographic focus
Global — offices in London (HQ) and San Francisco. Competitors span US (Ineffable Intelligence, SSI, AMI Labs), Japan (Sakana AI), and China.
About Richard Socher
Ex-Chief Scientist at Salesforce; Founder/CEO of You.com ($1.5B valuation); Founder of MetaMind (acquired by Salesforce); PhD in Computer Science from Stanford under Andrew Ng; Co-author of ImageNet and GloVe word vectors; 240K+ Google Scholar citations; Time100 AI honoree; WEF Young Global Leader
Official website: https://recursive.com