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

🔍 Read the full analysis: Could Compute Shortages Follow AI’s Rapid Progress In 2027? on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

TL;DR

A November 18, 2026, analysis presents a possible 2027 AI-compute crunch as a scenario, not a forecast. METR’s software-agent measurements show rapid gains but are no longer actively updated, and the projected timing changes sharply depending on the trend line. Data-center construction is rising, but power, deployment and the pace of AI-assisted research remain uncertain.

Thorsten Meyer published a scenario analysis on November 18, 2026, arguing that accelerating AI development could run into limits on chips, data-center capacity and electricity in 2027. The proposed crunch is not a confirmed forecast: the article says the outcome depends on whether AI systems keep speeding up AI research, and on how quickly physical infrastructure can expand.

The scenario begins with AI systems taking on more of the coding, testing and experimentation involved in building later systems. If that feedback loop tightens, Meyer argues, progress could arrive much faster than it has in the past. His specific proposition is that the pace might compress a year of progress into about two months, before growth in available computing capacity becomes a constraint. That is a hypothesis about a possible sequence, not a measured result.

One public measure cited is METR’s 80% time horizon: the length of a task, expressed in human expert work time, that an AI agent is predicted to complete successfully four times in five. METR’s reported results rose from 0.89 minutes for GPT-4 in March 2023 to 185.9 minutes for an early Claude Mythos Preview in April 2026. But METR marked its public page no longer actively updated on September 8, 2026, and the task set focuses mainly on software engineering, machine learning and cybersecurity—not the full range of economic work.

The projected date changes with the statistical fit. Meyer’s analysis says an exponential fit to METR measurements since 2024 reaches a 16-hour horizon in January 2027, while a fit beginning in 2019 reaches it in June 2028. The article says its hyperbolic fit, which reaches infinity in November 2027, performed worst in its backtests: a version fitted to data before July 2025 had already reached infinity by January 2, 2026. These are calculations based on selected data, not METR forecasts.

At a glance
analysisWhen: Published November 18, 2026; the propos…
The developmentThorsten Meyer published a scenario analysis on November 18 arguing that rapid AI progress could meet physical compute limits in 2027, while acknowledging substantial uncertainty in the evidence and timing.

Why Compute Could Set the Pace

If AI systems become more capable at contributing to AI research, their progress could affect how quickly new models are developed. But faster software progress does not automatically produce more usable AI: training and operating systems require chips, data centers and connected electricity. A mismatch between rapidly growing demand and infrastructure could slow expansion, raise costs or limit where and how new systems are deployed.

The infrastructure figures cited by Meyer point to a large buildout, but do not establish that a shortage will occur. US data-center construction was running at an annual rate of $96.5 billion in September 2026, compared with $50.7 billion a year earlier. The article attributes to Epoch AI the view that power is likely to be the first scaling constraint and that new power supply can take three to five years from a decision. Those estimates make grid access and construction timelines relevant to the scenario, while leaving the size and timing of any shortfall unsettled.

The distinction matters to businesses and workers, too. 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 figures do not establish how future adoption or employment will change. They do show why a forecast of rapid technical progress should not be treated as evidence that deployment across the economy will happen at the same speed.

Ninja Foodi Smart 6-in-1 Indoor Grill and Air Fryer, 4 qt, FG551

Ninja Foodi Smart 6-in-1 Indoor Grill and Air Fryer, 4 qt, FG551

As an affiliate, we earn on qualifying purchases.

What the Progress Measure Captures

METR describes its time 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 is a demanding success threshold, but it applies to tasks in METR’s evaluation suite. The organization says these tasks are cleaner and more clearly specified than real economically valuable work. It also cautions that a two-hour task is more comparable to work a new hire or freelance contractor might do in two hours than to the work of an experienced colleague.

METR has also warned that its current task suite does not support reliable measurements above 16 hours. Of 31 tasks estimated to take people at least eight hours, the source says human completion times were measured for only five; the rest were estimates. These limits matter because the scenario relies on extrapolating a rising curve beyond the range where the benchmark is most reliable. A benchmark trend can inform a discussion of AI capability, but it cannot by itself determine the date of a compute bottleneck.

The article also points to signs that AI is being used in AI development, but the scale is disputed or uncertain. It says METR estimates Anthropic is improving models on well-specified tasks at about twice the earlier rate, while saying it would not be surprised by a range from 1.3 to 3 times. A sixfold acceleration would be needed for a year of progress to arrive in two months. The source also says OpenAI’s GPT-6.2 Astra is the first model the company has released externally to meet its “High” threshold for AI self-improvement. That is a company threshold, not independent proof that the full feedback loop in Meyer’s scenario is operating at the required speed.

“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

Where the 2027 Scenario Could Break

The timing is not established. The different extrapolations in the source point to dates from January 2027 to June 2028, and the most dramatic curve performed poorly in the reported backtest. METR’s page was no longer actively updated as of September 8, 2026, so the cited public series does not settle whether the trend has continued or changed.

It is also unclear whether AI-assisted research is improving quickly enough to generate the proposed acceleration. METR’s reported estimate is about twice the earlier rate, with a wider possible range; Meyer’s two-month compression would imply roughly sixfold acceleration. The source does not establish that this gap has been closed. Nor does it quantify how much compute will be available in 2027, how much demand new systems will create, or whether the limiting factor will be chips, electricity, data-center construction or another constraint.

Finally, capability benchmarks are not measures of broad adoption. The cited figures on firm use and job cuts describe late 2025 and early 2026, not the effects of a future capability jump. The analysis itself was researched and drafted with AI assistance, including Anthropic’s Claude. It says Anthropic, OpenAI and METR neither reviewed nor sponsored the article; its claims should be read as Meyer’s analysis of cited material, not as conclusions endorsed by those organizations.

Signals to Watch Before 2027

The scenario can be tested as new evidence arrives. Relevant signals include whether METR or other evaluators publish updated, comparable agent task results; whether AI systems demonstrably shorten the research and engineering work used to develop later models; and whether independent evidence supports a faster improvement rate than the roughly twofold estimate cited in the article. Better measurements of longer tasks would also help clarify whether capability gains extend beyond METR’s present benchmark limits.

On the infrastructure side, readers can track data-center construction, chip supply, grid connections and the time required to bring new electricity generation online. The source’s three-to-five-year estimate for new power supply suggests decisions made now may shape capacity later, but it does not show that a shortage is inevitable. Until more benchmark and infrastructure data are available, the 2027 crunch remains a conditional scenario rather than a confirmed outcome.

Key Questions

Is an AI compute shortage in 2027 confirmed?

No. The source presents a scenario, not a confirmed forecast. It argues that rapid AI progress could meet limits in chips, data centers or power, but the timing and likelihood are uncertain.

What does METR’s 80% time horizon measure?

It estimates the length of tasks, measured by human expert completion time, that an AI agent is predicted to complete successfully four times in five. METR’s cited tasks mainly cover software engineering, machine learning and cybersecurity.

Why do the projections give different dates?

They use different periods of historical data and different curve fits. In Meyer’s calculations, an exponential fit from 2024 reaches a 16-hour task horizon in January 2027, while a fit from 2019 reaches it in June 2028. These are author calculations, not METR forecasts.

What could constrain AI growth?

The source identifies chips, data-center buildings and electricity as possible physical constraints. It cites Epoch AI’s view that power may be the first limit on scaling, but does not establish that any particular shortage will occur in 2027.

Source: ThorstenMeyerAI.com

You May Also Like

Why AI Hardware Efficiency Is Now a Big Deal

Just as AI advances rapidly, understanding why hardware efficiency matters is crucial for shaping a sustainable, high-performance future—discover how next-generation innovations are changing the game.

How To Keep Claude Code, Codex, And More In One Workspace

T3 Code is an open-source app for running coding agents in one interface, with separate Git worktrees and review tools.

9 Best AI Meeting Transcription Tools For 2026: Which Features Matter?

Plaud Note Pro leads the 2026 AI meeting transcription field; here is how all nine tools compare on summaries, battery, and subscription costs.