When will superintelligence arrive?
The forecast board
These are the most-cited public estimates, with sources. Each one uses its own definition, so read the "what" column carefully.
Said in Feb 2026 that it may be a couple of years away, and added "we could be wrong".
Source ↗Described as a country of geniuses in a datacenter, possibly arriving as early as 2026.
Source ↗Said in Oct 2025 that superintelligence may be less than a decade away.
Source ↗Said "maybe within five years" in 2026, and that it likely needs one or two more breakthroughs.
Source ↗25% probability by 2029 (Feb 2026).
Source ↗Says it is at least a decade away, probably much more.
Source ↗How timelines have shifted
The trend is clearer than any single number. Metaculus's community forecast for AGI was around 2070 in 2020. By February 2026 it had a median of 2033 and a 25% chance by 2029. The 2025 AI 2027 scenario imagined automated AI research leading quickly to superintelligence before the end of the decade. Lab leaders have moved from "decades" to "years".
Arguments for sooner
- Scaling keeps working. More compute and data have reliably produced more capable models, and spending on compute keeps growing.
- AI is starting to do AI research. OpenAI has set a public target of an automated AI researcher by 2028. Altman claimed in July 2026 that AI can now improve itself, though this hasn't been verified.
- Frontier results. In 2026, models were credited with research-level mathematics, including disproving a conjecture that had been open for about 80 years.
Arguments for later
- Missing pieces. Critics such as Yann LeCun argue that current architectures lack real world models, persistent memory and planning.
- Physical bottlenecks. Energy, chips and real-world experiments don't speed up as fast as software does.
- Verification. DeepMind's 2026 report highlights that checking superhuman outputs is itself a hard problem.
- Policy. Calls for a prohibition on superintelligence are growing, both from civil society and in Congress.
Track it live
Forecasts change with every model release. The Superintelligence Tracker logs each new claim, milestone and policy move with a source, so you can see the trend instead of a snapshot.
Frequently asked questions
What is the most common prediction for superintelligence?
There isn't a single consensus. AI company leaders cluster in the late 2020s. Forecasting platforms are later, with AGI around 2030 to 2033. Academic surveys are later still. Most forecasters agree that timelines have shortened dramatically since 2020.
Why do forecasts disagree so much?
Mostly because people use different definitions, and because they disagree about whether scaling current methods is enough or new breakthroughs are needed. Forecasts from lab leaders may also be shaped by incentives, while sceptics may underrate how quickly things are changing.
What would signal superintelligence is close?
Watch for automated AI research, meaning AI systems that meaningfully speed up the creation of the next AI system. OpenAI has publicly targeted an automated AI researcher by 2028. Other signals are AI producing verified scientific discoveries and agents working on their own for weeks at a time.
How often is this page updated?
Whenever a major forecast or claim changes. New entries appear first on the tracker homepage.