🔍 Read the full analysis: A Fast Year For AI? The 2027 Scenario Of Progress And Compute on ThorstenMeyerAI.com
Get hardware and tech essentials 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 scenario from ThorstenMeyerAI.com argues that AI could compress a year of progress into roughly two months in 2027 as systems help develop their successors. The article presents this as a conditional scenario, not a forecast: METR’s task data, AI labs’ self-improvement claims, physical compute limits and slow business adoption all leave major uncertainties.
Thorsten Meyer published a scenario analysis on November 18 arguing that AI progress could accelerate enough in 2027 for a year of development to arrive in about two months, as AI systems take on more work building later systems. The article calls this a conditional scenario, not a forecast established by current evidence, and says the process could slow when chip, building and power capacity limit available computing.
The case starts with a measure from METR, a nonprofit that evaluates frontier AI agents. Its 80% time horizon estimates the length of a task, measured by human completion time, that an agent can complete reliably four times out of five. In METR results cited by Meyer, the measure 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 in April 2026.
Those observations do not establish that the same pace will continue. Meyer’s own fits to the measured points produce different timelines: an exponential fit using data since 2024 doubles about every 3.4 months and reaches a 16-hour horizon in January 2027, while a fit using data from 2019 doubles about every 6.4 months and reaches that level in June 2028. The article says a hyperbolic fit that reaches infinity in November 2027 performed worst in its backtests. METR’s public page was marked no longer actively updated on September 8, 2026.
The proposed acceleration would also be faster than current evidence Meyer cites for AI-assisted model development. METR estimates that Anthropic is improving models on well-specified tasks at around twice the earlier rate; it says it would not be surprised by a rate between 1.3 and 3 times as fast. By comparison, compressing a year of progress into two months implies a sixfold pace. OpenAI’s GPT-6.2 Astra, the article says, is its first externally released model to meet the company’s “High” threshold for AI self-improvement. That is a company threshold, not proof that an autonomous development loop has produced the scenario’s proposed acceleration.
The Gap Between Capability and Deployment
If the development loop described in the scenario strengthens, it could change how quickly AI capabilities improve and how much work AI systems can perform. Meyer’s proposed next stage, which he calls embodied self-improvement, would extend the loop beyond software into robots, factories and automated laboratories. That stage depends on physical systems and infrastructure, so it would operate on a different timeline from software changes.
The article also distinguishes technical progress from effects on jobs and businesses. It cites estimates 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. These figures describe adoption and reported employment effects, not a measure of AI capability. A fast improvement in benchmark performance would not by itself show that businesses have adopted the systems widely or that labor-market impacts have followed.
Products Worth Considering

Logitech MX Brio Ultra HD 4K Webcam

Apple MacBook Pro 16.3-inch with PROLED XDR Display, Silver

LG 27G610A-B 27-inch UltraGear QHD (2560 x 1440) IPS Gaming Monitor, 200Hz, 1ms, AMD FreeSync Premium, VESA DisplayHDR 400, HDMI, DisplayPor
As an affiliate, we earn on qualifying purchases.
What METR’s Task Measure Captures
METR’s time horizon focuses mainly on software engineering, machine learning and cybersecurity tasks. The 80% figure is the estimated task duration at which an agent succeeds four times out of five; it is not a measure of general intelligence or of all work an AI system can perform. METR says its tasks are cleaner and more specified than economically valuable work in real settings, and likens a two-hour task more closely to work a new hire or freelancer might do than to an experienced colleague’s work.
The measure has limits at longer durations. METR says results above 16 hours are unreliable with its current task suite. Of 31 tasks that take human experts at least eight hours, it has measured human completion times for only five; the other times are estimates. The data therefore show growth on a defined set of tasks, but do not provide a reliable direct timeline for broad workplace automation.
The physical side of Meyer’s scenario has its own constraints. The article cites U.S. data-center construction running at an annual rate of $96.5 billion in September 2026, up from $50.7 billion a year earlier. It also cites Epoch AI’s view that power could become the first constraint on scaling, with new power supply taking three to five years after a decision. Those figures point to investment and lead times, but do not show exactly when any particular AI project will run short of compute.
“Measurements above 16 hrs are unreliable with our current task suite.”
— METR, as quoted in Meyer’s article
The Evidence Leaves Key Steps Open
The scenario requires several developments to occur in sequence: AI systems must speed up the work of developing later systems; that improvement must reach the scale implied by a sixfold acceleration; progress must encounter a physical compute ceiling; and automation must then help expand physical capacity. Each link could fail or proceed at a different pace.
It is not yet clear whether METR’s benchmark trend will continue, particularly because its public page was no longer actively updated and its task suite has limits at longer durations. The source also does not establish that OpenAI’s self-improvement threshold corresponds to a measured increase in research productivity, or that model gains on specified tasks will translate into broad economic effects. Meyer discloses that the article was researched and drafted with AI assistance, including Anthropic’s Claude, and says Anthropic, OpenAI and METR neither reviewed nor sponsored it.
Signals to Watch During 2027
Meyer identifies future measurements and infrastructure as the main tests of the scenario. A continuing rise in independently measured task horizons would support the claim that AI agents are taking on longer software work, while results from a more capable evaluation suite would help address METR’s stated limits. Evidence that AI systems are speeding up model research by more than the roughly twofold rate cited by METR would be needed to support the proposed compression more directly.
Data-center construction, chip availability and power supply will show whether compute capacity is expanding quickly enough to sustain further gains. The article does not provide a confirmed date for a compute ceiling or a specific prediction for when broader deployment will occur. Its 2027 scenario remains a proposition to test against those signals, not a timetable readers can treat as certain.
Key Questions
Is the article predicting that artificial superintelligence will arrive in 2027?
No. Meyer presents 2027 or shortly after as a scenario he is testing, not a confirmed forecast. The article also says artificial superintelligence has no agreed definition.
What does METR’s 80% time horizon measure?
It estimates the duration of a task, based on human expert completion time, at which an AI agent is predicted to succeed four times out of five. METR’s task set focuses mainly on software, machine learning and cybersecurity work.
Does the cited data show that AI progress is already six times faster?
No. Meyer says METR estimates Anthropic is improving models on well-specified tasks at around twice the earlier rate. A sixfold acceleration is the pace implied by the scenario’s claim that a year of progress could arrive in two months.
What could limit further AI development?
The scenario points to chips, data-center buildings and electricity as physical constraints. Meyer cites Epoch AI’s view that power could be the first scaling constraint and that new power supply may take three to five years after a decision.
Source: ThorstenMeyerAI.com
