🔍 Read the full analysis: 2027 AI Scenario: Rapid Progress And A Possible Compute Shortage on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer’s November 18, 2026 article presents 2027 rapid AI progress as a scenario, not a confirmed forecast. METR’s public task-horizon measurements show substantial gains through April 2026, but their scope and limits do not establish a sixfold acceleration; power, chips and deployment could also constrain change.
Thorsten Meyer published a 2027 AI scenario on November 18, arguing that AI systems could help build more capable successors quickly enough to compress a year of progress into two months, before growth encounters limits in chips, data-centre capacity and electricity. The article presents this as a possibility to test, not a forecast established by current evidence.
The scenario depends on a chain of developments: AI systems take on more work in developing later systems; that work speeds up progress; available computing capacity becomes a bottleneck; and, later, automation in factories and laboratories helps expand capacity. Meyer calls that physical phase embodied self-improvement. Each link could fail, and the article says the term artificial superintelligence has no agreed definition, so it focuses on the proposed mechanism rather than the label.
As evidence of improving software-task performance, the article cites METR’s 80% time horizon: an estimate of how long a task an AI agent can complete with 80% reliability, using human completion time as the measure. In METR’s data, the figure rose from 0.89 minutes for GPT-4 in March 2023 to 185.9 minutes for an early Claude Mythos Preview in April 2026. Those are measured results, not projections. METR marked its public page as no longer actively updated on September 8, 2026.
Meyer’s own curve fits produce sharply different dates. Fitting measurements from 2024 gives a doubling time of 3.4 months and reaches a 16-hour horizon in January 2027; fitting from 2019 gives a 6.4-month doubling time and reaches it in June 2028. The article says its hyperbolic fit performed worst in backtests, reaching infinity by January 2, 2026, when fitted to models before July 2025. The projections depend heavily on the selected trend and model, and METR cautions that measurements above 16 hours are unreliable with its current task suite.
Why Compute Could Set the Pace
The scenario matters because it links AI capability growth to both the technology sector and the physical infrastructure needed to run it. If AI systems materially speed up research and engineering, development cycles could shorten. But faster software progress would not automatically mean rapid adoption across the economy: a model’s ability to complete a measured task is not the same as its ability to perform a full job reliably, affordably and within an organization’s rules.
The article points to physical limits as a possible brake. It reports U.S. data-centre 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 could become the first scaling constraint and that new electricity supply can take three to five years from a decision. These figures describe investment and infrastructure timelines, not proof that a shortage has already halted AI progress.
Workplace adoption is another separate question. The article reports that 18% of U.S. firms used AI in a business function in late 2025 and early 2026, while 2% reported AI-related job cuts. That gap cautions against treating advances in benchmark performance as evidence of immediate, economy-wide labour effects.
Products Worth Considering

Logitech MX Master 4 Ergonomic Wireless Mouse with Haptics – Graphite

Apple Watch SE 3 GPS 40mm Smartwatch

DJI Mini 5 Pro Fly More Combo with DJI RC 2, 1-Inch CMOS
As an affiliate, we earn on qualifying purchases.
What METR’s Benchmark Measures
METR is a nonprofit that evaluates frontier AI agents. Its time horizon estimates the duration of tasks, based on human expert completion time, at which an agent is predicted to succeed with a specified reliability. The 80% measure cited by Meyer means success four times out of five. The tasks are primarily in software engineering, machine learning and cybersecurity, rather than a representative sample of all paid work.
METR also describes its tasks as cleaner than real economically valuable labour and says a two-hour task is closer to work assigned to a new hire or freelance contractor than to an experienced colleague’s work. For tasks taking at least eight hours for humans, the article says METR measured human completion times for only five of 31 tasks; times for the rest are estimates. These limits make the benchmark useful for tracking a particular capability, but not a direct measure of general economic productivity.
The article further reports that METR estimates Anthropic is improving its models on well-specified tasks at about twice its earlier rate, while saying it would not be surprised by a rate between 1.3 and 3 times. Meyer’s proposed pace—one year of progress in two months—would imply about a sixfold acceleration. The cited estimate does not establish that scale of change.
“Measurements above 16 hrs are unreliable with our current task suite.”
— METR, as quoted in Meyer’s article
Where the 2027 Case Could Break
No evidence in the article establishes that a 2027 acceleration will occur. The METR series cited ends with an April 2026 result, and its public page was no longer actively updated after September 8, 2026. Its software-focused tasks, estimated human timings for many long tasks and stated reliability limits restrict what can be inferred from the curve.
The article’s extrapolations vary according to the period used, while the fastest-fitting method failed in backtests described by Meyer. It is also unclear whether AI-assisted research will improve quickly enough to produce the proposed sixfold pace, or whether infrastructure additions can keep up with demand. The source does not provide a confirmed date for a compute shortage, a measured slowdown caused by one, or evidence that factory and laboratory automation will create the later self-improvement loop.
Meyer discloses that the article was researched and drafted with AI assistance, including Anthropic’s Claude. He says Anthropic, OpenAI and METR did not review or sponsor the article. Those disclosures do not validate or invalidate the scenario, but they are relevant to how readers assess its sourcing and framing.
Signals to Watch Through 2027
The scenario will be tested by future measurements of AI agents’ performance on longer tasks, particularly if METR or other researchers publish comparable, updated results with clear methods and limitations. Readers should distinguish new benchmark observations from extrapolations, and check whether claimed acceleration rates name a baseline, measurement window and task type.
Infrastructure is another signal: data-centre construction, chip availability, electricity connections and timelines for new power supply could show whether computing capacity is becoming a practical constraint. Evidence of a bottleneck would require more than rising construction spending; it would need to show that constrained resources are limiting deployment or development.
Finally, reports about AI use in businesses and employment can help test the scenario’s rollout stage, but adoption and job effects should be measured separately from model capability. For now, the 2027 outcome remains uncertain; the next meaningful assessment depends on updated performance data and observable changes in compute supply and real-world use.
Key Questions
Is rapid AI progress in 2027 confirmed?
No. Meyer describes it as a scenario, not a forecast established as fact. The article sets out evidence, counterarguments and signals that could test the idea.
What does METR’s 80% time horizon measure?
It estimates the human-duration task length at which an AI agent is predicted to succeed four times out of five. The cited tasks are mainly in software engineering, machine learning and cybersecurity, not a broad sample of jobs.
Why do the article’s projected dates differ?
Meyer fits the METR measurements over different time periods. The fit beginning in 2024 implies faster growth than the fit beginning in 2019, producing different dates for reaching a 16-hour horizon. They are extrapolations, not measured future results.
What could limit AI development even if models improve?
The scenario identifies chips, data-centre buildings and electricity as possible physical constraints. The article cites long power-supply timelines, but does not establish that a shortage has already stopped progress.
Source: ThorstenMeyerAI.com
