🔍 Read the full analysis: AI In 2027: Fast Advances Raise Questions About Compute on ThorstenMeyerAI.com
Get audio and creator gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A November 18, 2026 scenario from ThorstenMeyerAI.com argues that AI could accelerate its own development in 2027 before running into limits on chips, buildings and power. The article presents this as a scenario, not a forecast: METR’s public benchmark shows rapid progress but is no longer actively updated, and the outcome depends heavily on how its past trend is extrapolated.
ThorstenMeyerAI.com published a scenario on November 18 arguing that AI systems could help accelerate the development of later systems in 2027, before growth meets limits in chips, data-centre construction and electricity. The article presents the idea as a scenario, not a factual forecast, and says the available evidence does not establish that progress will compress to a year’s worth every two months.
The scenario’s central mechanism is a software feedback loop: AI systems take on more coding, testing and research work that contributes to building subsequent AI systems. If that loop tightens, the author argues, progress could arrive in shorter intervals. The piece then posits a physical bottleneck: additional capability cannot be translated into more computing capacity without chips, buildings and connected power. A later phase of improvement through automated labs and factories is described as slower and dependent on expanding that infrastructure.
The article uses METR’s 80% time horizon as one public measure of AI agents’ ability to complete software tasks. In the cited results, the measure rose from 0.89 minutes for GPT-4 in March 2023 to 185.9 minutes for an early Claude Mythos Preview in April 2026. METR defines the horizon as the task duration, measured by human expert completion time, at which an agent is predicted to succeed with a specified reliability. The 80% figure means success four times out of five.
That record does not settle when systems might reach much longer tasks. The source says METR marked its public page as no longer actively updated on September 8, 2026, and warns that measurements above 16 hours are unreliable with its current task suite. Its tasks largely concern software engineering, machine learning and cybersecurity, and METR says they are cleaner than economically valuable work in real settings.
Compute Could Set the Pace
The scenario matters because it links two issues that are often discussed separately: AI capability growth and the physical infrastructure needed to run and develop increasingly powerful systems. If AI speeds up parts of its own research and engineering, progress could affect companies and workers faster than ordinary annual product cycles suggest. But the article’s evidence does not establish that such a cycle will produce economy-wide change on that schedule.
The proposed bottleneck also has consequences beyond AI labs. The source cites US data-centre construction running at a $96.5 billion annual rate in September 2026, compared with $50.7 billion a year earlier. It says Epoch AI expects power to be the first constraint on scaling and that new power supply takes three to five years from a decision. Those figures suggest why physical capacity may shape development, but they do not show how much capacity future AI systems will require or when a shortage would occur.
The distinction between capability and adoption is also material to readers assessing possible effects on work. 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 are separate measures, not proof that future job impacts will be limited or that rapid capability gains will translate directly into widespread deployment.
Products Worth Considering

Samsung 32 Odyssey G5 QHD Gaming Monitor

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G50F)

Logitech MX Master 4 Ergonomic Wireless Mouse with Haptics – Graphite
As an affiliate, we earn on qualifying purchases.
What the Benchmarks Measure
METR’s time horizon is a benchmark for how long an AI agent can work on certain tasks at a specified success rate; it is not a direct measure of general intelligence, economic value or the pace of all AI research. The source emphasizes that a two-hour task is closer to work a new hire or freelance contractor could complete than to the work of an experienced colleague. It also reports that human completion times are directly measured for only five of the 31 tasks taking eight hours or more, with the rest estimated.
The scenario’s projected date changes with the historical data selected. The author’s own exponential fit to METR points since 2024 doubles every 3.4 months and reaches a 16-hour horizon in January 2027. A fit beginning in 2019 doubles every 6.4 months and reaches that point in June 2028. These are the author’s extrapolations, not METR forecasts. The article also tests a hyperbolic curve that reaches infinity in November 2027, but says that approach performed worst in its backtests and had already projected an infinite horizon by January 2, 2026, when fitted to earlier models.
The source 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 of 1.3 or 3 times. The scenario’s proposed “a year of progress in two months” would require roughly sixfold acceleration, a substantially stronger claim than the estimate attributed to METR. The article also says OpenAI’s GPT-6.2 Astra is the first externally released model from the company to meet its “High” threshold for AI self-improvement; this is a company-defined threshold, not by itself evidence that the full feedback loop is operating at the proposed speed.
The 2027 Timing Is Unsettled
Whether the feedback loop will produce a sixfold acceleration is unknown. The source’s cited estimate for Anthropic’s improvement rate is about twice the earlier rate, with a wider range METR would not rule out; that is not the same as the sixfold acceleration required for a year of progress in two months. The source does not provide evidence that the full chain—from AI-assisted research to rapid capability gains, then infrastructure expansion and broad adoption—will occur on the proposed timeline.
The benchmark itself has important limits: it covers selected software tasks, has few directly measured long-duration human task times, and is unreliable above 16 hours. Its public page is also reported to have stopped active updates. The article’s January 2027, June 2028 and November 2027 dates come from different curve fits, not observed future performance. It is also unclear how quickly new electricity supply, chips and data centres could expand in response to demand, or how fast better-performing systems would be adopted by businesses.
The source identifies a conflict of interest relevant to its process: the article was researched and drafted with AI assistance, including Anthropic’s Claude. It says Anthropic, OpenAI and METR did not review or sponsor the article. Artificial superintelligence also has no agreed definition, so the piece focuses on the proposed mechanism rather than claiming a settled threshold for that label.
Signals to Watch in 2027
The scenario will be tested by new benchmark results and whether independent measurements show agents completing longer, reliably measured tasks. METR’s public page is no longer actively updated, according to the source, so future comparisons may depend on other evaluations or on a resumption of updates. Any new figures will need to be read alongside the task types, reliability level and measurement method used.
Other indicators include the rate at which AI systems contribute to model research and engineering, and whether that produces a sustained increase in improvement rates rather than isolated gains. On the infrastructure side, data-centre construction, chip availability and power connections will show whether physical capacity is keeping pace with demand. Business-use and employment data will help distinguish stronger models from actual deployment and workplace effects.
The article offers no confirmed milestone establishing that the proposed 2027 acceleration will occur. Its stated test is whether these signals support each link in the chain; if they do not, the scenario’s timing or underlying mechanism would need to be reconsidered.
Key Questions
Is the article predicting that AI will reach superintelligence in 2027?
No. It presents 2027 or shortly after as Thorsten Meyer’s scenario, not as a forecast established by the evidence. The source says artificial superintelligence has no agreed definition.
What does METR’s 80% time horizon measure?
It estimates the length of a task, measured by human expert completion time, 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.
Why do the article’s projected dates differ?
The author fits different trend lines to different spans of METR data. The exponential fit since 2024 reaches 16 hours in January 2027, while a fit since 2019 reaches it in June 2028. These are modelled extrapolations, not measured outcomes or official METR projections.
What could limit faster AI development?
The scenario identifies chips, data-centre buildings and electricity as possible physical limits. The source cites an estimate that new power supply takes three to five years from a decision, but does not establish when or where a binding constraint will arise.
Does stronger AI capability mean widespread job losses?
Not on its own. The source reports that 18% of US firms used AI in a business function in late 2025 and early 2026, and 2% reported AI-related job cuts. Those figures describe different measures and do not determine future employment effects.
Source: ThorstenMeyerAI.com
