TL;DR
A November 18, 2026 scenario article argues that AI progress could accelerate sharply in 2027 before running into limits on chips, data centers and power. Public measurements show software task horizons rising, but the article’s projections depend on which historical trend is fitted, and current evidence does not establish a sixfold acceleration.
Thorsten Meyer published a 2027 AI acceleration scenario on November 18, arguing that AI systems could help build their successors quickly enough to compress a year of progress into two months, then slow when chips, data centers and power become limiting factors. The article presents this as a scenario to test, not a forecast established by current measurements.
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The scenario rests on a feedback loop: AI systems take on more coding, debugging and experimentation involved in developing later systems, potentially speeding up each generation. Meyer’s proposed pace—a year of progress in two months—would amount to about six times the earlier rate. The article says the loop could eventually meet a physical ceiling, after which improvements to factories, laboratories and robotics might expand computing capacity more slowly.
One public measure cited is METR’s 80% time horizon, which tracks the length of software tasks an AI agent completes successfully four times out of five. METR’s cited results rose from under a minute for GPT-4 in March 2023 to 185.9 minutes for an early Claude Mythos Preview in April 2026. METR marked its public page as no longer actively updated on September 8, 2026, limiting the availability of newer comparable measurements.
Meyer’s calculations fit several curves to METR’s measured points. An exponential fit beginning in 2024 doubles about every 3.4 months and reaches a 16-hour task horizon in January 2027; a fit beginning in 2019 doubles every 6.4 months and reaches that mark in June 2028. The article says the more dramatic hyperbolic curve performed worst in its backtests, reaching infinity by January 2, 2026, when fitted to models available before July 2025. METR’s suite stops measuring above 16 hours, so these projections extend beyond its measured range.

The Gap Between Models and Compute
The scenario matters because it links AI capability gains to constraints that take years to relieve. If systems become substantially better at helping develop AI, research cycles could shorten; if computing infrastructure cannot grow at the same pace, hardware and electricity could limit how quickly those gains translate into larger or more capable systems.
The article cites US data-center construction running at an annual rate of $96.5 billion in September 2026, compared with $50.7 billion a year earlier. It also cites Epoch AI’s view that power may be the first scaling constraint and that new power supply can take three to five years from a decision. These figures point to infrastructure pressures, but they do not establish when or whether a compute ceiling will halt AI progress.
Capability gains also do not automatically translate into broad workplace adoption. The source reports that 18% of US firms used AI in a business function in late 2025 and early 2026, while 2% reported AI-related job cuts. Those figures describe different measures of adoption and employment effects; they do not show that rapid technical progress would produce an equally rapid economic shift.
Products Worth Considering

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight

Bose SoundTouch 30 Wireless Speaker, Black
As an affiliate, we earn on qualifying purchases.
What the Benchmarks Can Show
METR’s time horizon is a measure of how long a software task can be while an agent still has an 80% chance of completing it. It is one yardstick for progress on a defined set of tasks, not a direct measure of general intelligence, all workplace activity or the pace of AI research overall. The article notes that the benchmark’s public update status changed in September 2026.
The projected dates differ because they depend on which portion of the historical data is used and what curve is fitted. The 2024 exponential fit and the longer 2019 fit yield different doubling times and different dates for reaching 16 hours. Meyer’s article also says METR estimates Anthropic’s improvement on well-specified tasks at around twice the earlier rate, while stating it would not be surprised by a rate between 1.3 and 3 times. That estimate is below the scenario’s proposed sixfold acceleration.
The article discusses OpenAI’s GPT-6.2 Astra as the first externally released OpenAI model to meet the company’s “High” threshold for AI self-improvement. This is a company-defined threshold cited by Meyer; it is not, by itself, evidence that AI systems are already driving a sixfold increase in overall AI progress. Meyer disclosed that the article was researched and drafted with AI assistance, including Anthropic’s Claude, and said Anthropic, OpenAI and METR did not review or sponsor it.
“We reach ASI either in 2027 or shortly after. The effect: in two months we see the progress of a whole year, and then further progress until we hit a ceiling in available compute.”
— Thorsten Meyer
Where the Projection Could Break
The article does not establish that the feedback loop will reach sixfold acceleration, or that it will do so in 2027. The measured task-horizon series cited ends with an April 2026 result, while the public page was no longer actively updated from September 8. Projections to 2027 and beyond therefore rely on curve fitting rather than new measurements at those horizons.
It also remains unclear when infrastructure constraints would bind, which resource would become limiting first, and how quickly additional power or data-center capacity could come online. The source describes the chart’s compute ceiling as an illustration, not a forecast value. The timing and scale of any broader effects on businesses, workers or robotics are likewise not established by the figures presented.
Signals to Watch in 2027
The scenario can be assessed as new evidence arrives: updated, comparable task-horizon measurements; clearer evidence about how much AI systems contribute to AI research and development; and changes in data-center construction and available power. Those indicators could show whether measured capability trends are tracking the faster or slower curves discussed in the article.
Evidence of widespread deployment would also help test the proposed path from capability to economic impact. The article does not give a specific date for a future METR update or another milestone, so whether new benchmark results will arrive in time to evaluate the 2027 projections remains unclear.
Key Questions
Is the article predicting that artificial superintelligence will arrive in 2027?
No. Meyer presents 2027 as a scenario to test, not as a confirmed forecast. The article also notes that artificial superintelligence has no agreed definition.
What does METR’s 80% time horizon measure?
It estimates the length of software tasks an AI agent completes successfully four times out of five. It measures performance on a task suite, not general intelligence or every kind of work.
Why do the projections reach 16 hours at different times?
The dates depend on which historical measurements are included and which trend line is fitted. Meyer’s 2024 exponential fit reaches 16 hours in January 2027, while the fit beginning in 2019 reaches it in June 2028.
What could limit AI progress under this scenario?
The proposed limits are chips, data-center buildings and power. Meyer cites Epoch AI’s view that power may be the first constraint and that new supply can take three to five years from the decision.
Is there evidence that AI is already accelerating AI development sixfold?
The source does not establish a sixfold acceleration. It reports METR’s estimate of roughly twice the earlier rate for Anthropic’s improvement on well-specified tasks, with a stated plausible range of 1.3 to 3 times.
Source: Thorsten Meyer AI
