AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Thorsten Meyer’s article presents a possible 2027 scenario in which AI advances quickly through AI-assisted development, then slows as chips, data centers and power become constraints. METR benchmark results show longer software tasks completed by AI agents, but the projected pace depends heavily on the data fit, and the cited evidence does not establish a sixfold acceleration or confirm that the scenario will occur.

Thorsten Meyer’s November 18 analysis outlines a possible 2027 scenario in which AI systems help build more capable successors, accelerating progress before chips, data-center construction and power limit further gains. The article treats this as a scenario rather than a forecast: current benchmark and infrastructure data offer some support for its mechanisms, but do not confirm the predicted pace or timing.The article uses METR’s 80% time horizon as a measure of how long a software task an AI agent can complete successfully four times out of five. It reports that the measured horizon rose from under a minute for GPT-4 in March 2023 to 185.9 minutes—about three hours and six minutes—for an early Claude Mythos Preview in April 2026. METR marked its public page as no longer actively updated on September 8, 2026, according to the source article. Meyer’s calculations show how much the projection depends on which historical data points are used. An exponential fit beginning in 2024 doubles about every 3.4 months and crosses 16 hours in January 2027; a fit beginning in 2019 doubles every 6.4 months and crosses that threshold in June 2028. The article says a hyperbolic fit reaches infinity in November 2027, but performed worst in its backtests. METR’s benchmark suite does not measure tasks above 16 hours, so projections beyond that point extend past the reported measurement range. The article also distinguishes an estimated improvement rate from the faster scenario. It says METR estimates Anthropic is improving its models on well-specified tasks at around twice the earlier rate, while describing 1.3 times or three times as plausible possibilities. Compressing a year of progress into two months would require about a sixfold acceleration, which the cited estimate does not establish. The source also reports that GPT-6.2 Astra is the first externally released OpenAI model to meet the company’s “High” threshold for AI self-improvement.
At a glance
analysisWhen: Published November 18, 2026; the scenar…
The developmentThorsten Meyer published a scenario examining whether AI could compress a year of progress into two months in 2027 before hitting compute constraints.
Log-scale chart titled Measured to April 2026, then five possible futures. Dark points show METR's measured 80 percent time horizons of record-setting models from 2019 to April 2026, ending in an orange point at 3 h 6 min for Claude Mythos Preview (early). Five fitted curves run from the last measurement towards 2035: exponential since 2024 doubling every 3.4 months, exponential since 2023 doubling every 4.2 months, exponential 2019 to 2026 doubling every 6.4 months, a quadratic-in-log curve and a hyperbolic curve that becomes infinite in November 2027. A dashed line marks 16 h; above it the chart is labelled beyond METR's task suite
Measured points and our five fits. Source: METR, Time Horizon 1.1 results file, retrieved November 18, 2026; fits and projections: our calculation.
Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

The Gap Between Capability and Scale

If AI systems take on more of the work involved in developing and testing future systems, the pace of capability gains could affect how quickly companies adopt AI and how much labor is exposed to automation. But faster benchmark progress would not by itself show that businesses are deploying AI at the same rate. The article 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. Those figures point to a gap between technical capability and observed business effects. The proposed physical ceiling also matters for planning. The source reports U.S. data-center construction at a September 2026 annual rate of $96.5 billion, up from $50.7 billion a year earlier. It cites Epoch AI’s view that power may become the first constraint on scaling, with new power supply taking three to five years from a decision. These figures describe investment and infrastructure timelines; they do not establish when a specific AI system or industry will reach a limit.

Products Worth Considering

Bose QuietComfort Ultra Wireless Earbuds (2nd Gen) with Deep Plum Finish and 20W Dual-Port Charger

Bose QuietComfort Ultra Wireless Earbuds (2nd Gen) with Deep Plum Finish and 20W Dual-Port Charger

Logitech MX Master 4 Ergonomic Wireless Mouse with Haptics – Graphite

Logitech MX Master 4 Ergonomic Wireless Mouse with Haptics – Graphite

As an affiliate, we earn on qualifying purchases.

The Evidence Behind the Timeline

The scenario links four steps: AI helps develop later AI systems; that feedback loop speeds up progress; available compute limits the software gains; and, later, automated physical systems and facilities help expand production and deployment. Meyer calls this final phase embodied self-improvement. Each step is conditional, and the article says the chain could break at any point. METR’s time horizon is a measure of task length at a specified success rate, not a direct measure of general intelligence or a calendar forecast. The source says the chart plots METR’s measured models from GPT-2 in 2019 through an early Claude Mythos Preview in April 2026. Its five fitted curves are Meyer’s calculations using those data, while the projected scenario curve and compute ceiling are illustrations rather than measurements. The source labels the piece a scenario and opinion article, and says it was researched and drafted with AI assistance, including Anthropic’s Claude. It discloses that Anthropic is among the labs whose evidence is discussed and says Anthropic, OpenAI and METR did not review or sponsor the article.

“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

What the Data Cannot Establish

The source material does not show that AI progress will accelerate sixfold, that artificial superintelligence will arrive in 2027, or that a compute ceiling will be reached on a particular date. METR’s public page is described as no longer actively updated, and the benchmark suite stops measuring task horizons above 16 hours. The extrapolations therefore extend beyond both the observed data and the benchmark’s stated range. It is also unclear whether AI-driven development will keep speeding up at the assumed rate, how quickly chip and data-center capacity can expand, and where power constraints will bind. The article’s figures on AI use and job cuts do not determine how adoption or employment will change during 2027. No timing or scale for the proposed embodied self-improvement phase is confirmed in the source.

Signals to Watch in 2027

The scenario’s claims can be assessed as new benchmark results, model-development evidence and infrastructure data appear. Relevant signals include whether measured task horizons continue to rise, whether independent evidence shows AI systems materially accelerating AI research, and whether data-center construction and power supply keep pace with demand. Readers should also watch for updated business-use and employment figures, since technical capability does not establish broad deployment. The source does not identify a scheduled milestone that will settle the scenario; its central test is whether the proposed feedback loop and physical constraints become visible in the coming year.

Key Questions

Is the article predicting that artificial superintelligence will arrive in 2027?

No. It presents a scenario, not a confirmed forecast. It also says artificial superintelligence has no agreed definition.

What does METR’s 80% time horizon measure?

It measures the length of software tasks an AI agent can complete successfully four times out of five. It is a benchmark measure, not a direct forecast of general intelligence.

Does the cited evidence show AI progress is accelerating sixfold?

No. The article reports METR’s estimate of about a twofold improvement rate for Anthropic on well-specified tasks. The sixfold rate is what the two-month scenario would require, not a measured result.

Why might compute limit AI progress?

AI systems run on physical infrastructure, including chips, data centers and power. The article cites a view that power may become a constraint, but it does not establish when or where that will happen.

What evidence could test the scenario?

Updated task-horizon measurements, evidence about AI’s role in developing later systems, and data on compute capacity, power supply and business adoption could show whether its proposed links are emerging.

Source: Thorsten Meyer AI

You May Also Like

The Real Cost Of A Local-Inference Rig In 2026

An in-depth analysis of the hardware costs for local AI inference in 2026, highlighting VRAM constraints, hardware choices, and value considerations.

Muse Spark Open

Meta has released Muse Spark Open, an 816-billion-parameter open-weight model for coding and agentic tasks, with desktop support on two M5 Ultra Macs.

Apple Wants Blacklisted Chinese RAM — And That Tells You How Bad The Squeeze Got

Apple is lobbying US officials to buy Chinese-made memory chips from CXMT, raising concerns over supply security and national security implications.

A Study Of Microsoft’s Early 2026 Rollout Of Claude Code And GitHub Copilot CLI

Microsoft is confirmed to launch Claude Code and GitHub Copilot CLI in early 2026, aiming to enhance developer tools with AI integration.