Artificial superintelligence
AI that vastly exceeds the best humans in every cognitive domain, including science, strategy and social reasoning, and can run civilization-scale infrastructure.
State of the art
Superhuman performance only in bounded domains (games, protein structure, some algorithm discovery); general agents remain below expert-human reliability on long tasks (2026).
- Requires AGI plus sustained self-improvement, neither yet demonstrated
- Unknown whether capability keeps scaling with compute or hits data and algorithmic limits
- Energy and chip supply for orders of magnitude more compute
Who’s working on it
- Safe Superintelligence Inc.Startup
Ilya Sutskever's lab aimed only at superintelligence; Nvidia committed $5B to it in July 2026
- Ilya SutskeverResearcherSafe Superintelligence Inc.
OpenAI co-founder who left in 2024 to found SSI with a 'straight shot' superintelligence plan
- Meta Superintelligence LabsCompany
Formed 2025 under Alexandr Wang; in Sept 2026 said most Meta compute would serve users rather than racing on recursive self-improvement
- OpenAI ↗Startup
Says publicly that it is working toward superintelligence after AGI; Stargate is its compute base
- Google DeepMind ↗Company
AlphaFold, AlphaGo and AlphaEvolve-style systems are the main examples of superhuman performance within a bounded domain
- METR ↗Academic lab
Tracks whether AI R&D automation and long-horizon autonomy are accelerating, a leading indicator of recursive self-improvement
What to invest in
| Kind | Investment | Scale |
|---|---|---|
| research | Scaling-law and capability-forecasting science to predict when general superhuman performance emerges | $$ $10M – $100M |
| infrastructure | Multi-gigawatt compute with secure, monitored training environments | $$$$ > $1B |
Depends on
Required by
No other capability builds on this directly.