When Will Superintelligence Arrive? Every Major Prediction Compared

When will machines become smarter than us? It may be the most consequential forecasting question of the century. The people closest to the technology disagree by decades. This page collects the most cited predictions, with dates and sources, so you can compare them directly.

A note on terms: this page is about superintelligence (ASI), meaning hypothetical systems far beyond human ability. It is not about the U.S. government’s new name for AI, “Super Intelligence” (SI). See SI vs superintelligence. Many forecasts are framed around AGI, human-level ability, which most people expect would come first.

The predictions at a glance

Who	What they predicted	When they said it
Sam Altman, CEO, OpenAI	“It is possible that we will have superintelligence in a few thousand days (!); it may take longer”	September 2024
Anthropic (company submission)	“Powerful AI systems will emerge in late 2026 or early 2027”	March 2025
Dario Amodei, CEO, Anthropic	Powerful AI “could come as early as 2026”	October 2024
Elon Musk, xAI	AI smarter than people by the end of 2026, as reported by Windows Central	Repeated several times; see sources
Demis Hassabis, CEO, Google DeepMind	AGI in “five to 10 years”	March 2025
Ray Kurzweil, futurist	Human-level AI by 2029; “the Singularity” by 2045	Restated in 2024
AI 2027 scenario, AI Futures Project	A detailed scenario in which superhuman AI researchers arrive around 2027	April 2025
2,778 AI researchers, survey	50% chance of “high-level machine intelligence” by 2047	Published January 2024
Skeptics (e.g., Yann LeCun, Gary Marcus)	Current methods alone won’t produce human-level intelligence; many years or decades away	Ongoing

The optimists: lab leaders

The most aggressive timelines come from the people running frontier labs.

Sam Altman wrote in his 2024 essay The Intelligence Age that superintelligence might arrive “in a few thousand days”. That would put it roughly between 2030 and the mid-2030s. In June 2025’s The Gentle Singularity he went further: “We are past the event horizon; the takeoff has started.”

Anthropic has been unusually specific. In formal recommendations to the White House in March 2025, it said it expected “powerful AI systems” to “emerge in late 2026 or early 2027”. It defined these as systems with “intellectual capabilities matching or exceeding that of Nobel Prize winners across most disciplines.” CEO Dario Amodei had written in Machines of Loving Grace that such AI “could come as early as 2026”, while stressing uncertainty. As of October 2026, that window is open now, so this prediction will be tested soon.

Demis Hassabis of Google DeepMind has been more measured, saying in 2025 that AI matching humans at any task would arrive in “five to 10 years”.

The long-run forecasters

Ray Kurzweil has predicted since 1999 that computers would pass a demanding Turing test by 2029. He reaffirmed it in The Singularity Is Nearer (2024), with a merger of human and machine intelligence, “the Singularity”, by 2045.

The AI 2027 scenario, written by former OpenAI researcher Daniel Kokotajlo and colleagues, sketched one concrete path in which AI systems automate AI research around 2027 and progress accelerates sharply after that. Its authors present it as a plausible scenario rather than a confident forecast.

What researchers say

The largest survey, published in January 2024 by Katja Grace and colleagues, asked 2,778 AI researchers when unaided machines would outperform humans at every task. The aggregate forecast gave a 50% chance by 2047, 13 years earlier than the same survey’s estimate a year before. The jump showed how much the arrival of ChatGPT-era systems shifted expectations.

The skeptics

Not everyone thinks current approaches lead to superintelligence at all:

• Yann LeCun, Meta’s former chief AI scientist, who announced in November 2025 that he would leave to start his own company, has long argued that large language models lack world models, memory and planning, and won’t reach human-level intelligence by scaling alone.
• Gary Marcus, cognitive scientist and frequent critic, argues that persistent problems like hallucination and unreliable reasoning point to fundamental limits.
• Others point to physical bottlenecks: chips, electricity and high-quality data. See scaling laws and compute.

Why it matters for policy

Timelines shape the debate. If superintelligence is years away, binding safety rules look urgent. That is the case made by the Future of Life Institute’s 2025 Statement on Superintelligence and by Senator Sanders’ September 2026 bill to ban it. If it is decades away or impossible with current methods, lighter, sector-specific regulation looks more proportionate, closer to the current U.S. federal approach. Our policy tracker follows both camps.

How to read any prediction

1. Check the definition. “AGI”, “powerful AI”, “human-level” and “superintelligence” mean different things to different people.
2. Check the incentives. Lab leaders raise money on the promise of transformative AI; critics build audiences on skepticism.
3. Check the date. Forecasts made before 2023 often look dated now. Track whether a forecaster updates when their deadlines pass.
4. Watch the evidence. Benchmarks, real-world autonomy and especially whether SI starts to meaningfully speed up its own research (an intelligence explosion) are better signals than any single quote.

We’ll update this page as new forecasts are made and old ones come due.