🔍 Read the full analysis: Could AI’s 2027 Surge Stall When Compute Runs Out? on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A November 18 scenario from ThorstenMeyerAI.com examines whether AI systems could speed up AI development enough to compress a year of progress into two months in 2027, then slow when compute capacity runs short. Its evidence shows rising software-task performance and rapid data-center construction, but also measurement limits, a much smaller acceleration estimate from METR, and modest business adoption. The proposed timeline is a scenario, not an established forecast.
ThorstenMeyerAI.com published a scenario on November 18 arguing that AI development could accelerate sharply in 2027 as systems help build subsequent systems, then slow when chips, data-center space or power become limiting. The article presents that sequence as a possibility to test, not a forecast established by the evidence; public measurements do not yet demonstrate the proposed sixfold acceleration or show when infrastructure would bind.
The scenario’s proposed chain has four steps: AI systems take on more coding and experimentation in AI development; that work speeds the next generation; progress reaches a physical compute ceiling; and later expansion in hardware and power allows automated systems to spread into more work. Thorsten Meyer describes the proposed physical phase as “embodied self-improvement”. Each step is conditional, and the article says the chain could break at any of them.
One measure cited is METR’s 80% time horizon: the length of a task, measured by human completion time, at which an agent is predicted to succeed four times in five. METR’s reported results rose from 0.89 minutes for GPT-4 in March 2023 to 185.9 minutes, or about three hours, for an early Claude Mythos Preview result in April 2026. These are measured software-task results, not projections of general intelligence or economy-wide work performance.
The article says its own projections vary substantially with the historical period used to fit the data. An exponential fit beginning in 2024 reaches a 16-hour horizon in January 2027; a fit beginning in 2019 reaches it in June 2028. The article’s hyperbolic fit implies a mathematical singularity in November 2027, but its backtests performed worst among five approaches and had already projected an infinite value by January 2, 2026. METR’s public page was marked no longer actively updated on September 8, 2026, adding a limitation to interpreting the trend.
The Gap Between AI Progress and Deployment
The scenario matters because it links AI capability to the pace of AI research and, later, to infrastructure and employment. If systems materially accelerate the work needed to build more capable systems, development timelines could shorten. But the source’s evidence does not establish that a rapid feedback loop is already operating at the scale or speed required for a 2027 surge.
There is also a difference between a benchmark result and practical economic change. The source 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. Those figures do not settle future effects, but they show why capability claims should not be treated as proof of broad deployment or labor-market change.
Physical capacity could shape the timing even if software improves quickly. The article cites U.S. data-center construction running at a $96.5 billion annual rate in September 2026, compared with $50.7 billion a year earlier. It also cites Epoch AI’s view that power may become the first scaling constraint and that new supply can take three to five years after a decision. These figures indicate infrastructure activity and lead times; they do not establish when a shortage will occur.
Products Worth Considering

Apple MacBook Air 15-inch Laptop with M5 Chip, 24GB Memory, 1TB SSD, Midnight

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

FlexiSpot E6 Electric Standing Desk, 55 × 28 in, Bamboo Top, White Frame
As an affiliate, we earn on qualifying purchases.
What METR’s Benchmark Can Show
METR evaluates frontier AI agents using tasks that are largely in software engineering, machine learning and cybersecurity. Its time horizon estimates the human-expert duration of tasks an agent can complete at a specified reliability. The 80% measure is not interchangeable with the 50% horizon, which is longer for the same model.
METR cautions that its tasks are cleaner and lower-context than economically valuable work in real organizations. It says that a two-hour benchmark task is closer to work a new hire or freelance contractor might complete in two hours than to the work of an experienced colleague. It also warns that measurements above 16 hours are unreliable with its current task suite; among 31 tasks estimated to take humans at least eight hours, human completion times were measured for only five.
The article also distinguishes the observed benchmark trend from its interpretation of AI-assisted AI research. It cites METR’s estimate that Anthropic is improving models on well-specified tasks at about twice its earlier rate, while METR said it would not be surprised by a rate between 1.3 and 3 times as fast. A year of progress in two months would imply about sixfold acceleration, a substantially stronger claim than that estimate.
“the task duration (measured by human expert completion time) at which an AI agent is predicted to succeed with a given level of reliability.”
— METR, describing its time-horizon measure
Which 2027 Signals Will Hold Up
The timing is unresolved. The source’s fitted dates range from January 2027 to June 2028, depending on which measurements are included, while the hyperbolic curve is explicitly described as the least reliable approach. The benchmark itself has limits on task scope and duration, and its public page was no longer actively updated as of September 8, 2026.
It is also unclear whether AI-assisted research is accelerating model development enough to create the proposed feedback loop. The source gives a METR estimate of roughly twofold improvement for a specific class of well-specified tasks, not evidence that all AI research is accelerating at that rate. Whether infrastructure becomes a binding constraint, when additional electricity and data centers would be available, and how quickly capable systems would be adopted by businesses are also unsettled.
The source describes OpenAI’s GPT-6.2 Astra as its first externally released model to meet the company’s “High” threshold for AI self-improvement. That is an attributed company classification; the material does not provide independent validation or enough detail to establish what the threshold predicts about development speed. The scenario also discloses that it was researched and drafted with AI assistance, including Anthropic’s Claude, and says Anthropic, OpenAI and METR did not review or sponsor it.
Evidence to Watch Through 2027
The most useful next evidence would be updated, comparable agent evaluations with clear task definitions and reliability levels, especially results that extend beyond the current benchmark’s stated 16-hour limit. Readers can also watch whether independent measurements support sustained acceleration rather than a curve selected from a short or unusually fast period.
Infrastructure indicators offer a separate test: data-center construction, access to chips, electricity supply and the time between power decisions and new capacity coming online. Even if software benchmarks improve quickly, those measures will help show whether compute limits are binding or whether new capacity is arriving in time.
Finally, evidence of broader adoption would help distinguish capability from impact: how many firms use AI in real business functions, what tasks systems perform, and whether measurable changes in staffing or productivity follow. Until such evidence clarifies the links in the chain, the proposed 2027 surge remains a scenario rather than a confirmed development.
Key Questions
Is the article predicting that AI will reach superintelligence in 2027?
No. It tests a scenario based on Thorsten Meyer’s stated view that ASI could arrive in 2027 or shortly afterward. The source does not present that timing as established fact and says ASI has no agreed definition.
What does METR’s 80% time horizon measure?
It estimates the duration of tasks, measured by human expert completion time, that an AI agent can complete with 80% reliability. The cited tasks are mainly in software engineering, machine learning and cybersecurity, so the measure is not a direct assessment of all workplace tasks.
Do current measurements show a sixfold acceleration?
No. The source cites METR’s estimate of about twofold improvement for Anthropic models on well-specified tasks, with a possible range METR would not be surprised to see of 1.3 to 3 times. The sixfold figure is what the scenario would require for a year of progress to arrive in two months.
What could stop a rapid AI-development loop?
The scenario points to chips, data-center buildings and electricity as possible constraints. It cites a three-to-five-year lead time for new power supply after a decision, but the source does not establish when a specific shortage will occur.
Source: ThorstenMeyerAI.com
