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

🔍 Read the full analysis: What If AI Makes A Year’s Progress In Two Months—and Compute Runs Out? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

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

TL;DR

Thorsten Meyer’s Nov. 18 article presents a possible 2027 scenario in which AI helps accelerate AI development, compressing a year of progress into two months before compute constraints slow it. The piece is an opinion scenario, not a forecast established by evidence: METR’s public benchmark has limits, its trend varies with the data chosen, and power, chips and deployment remain constraints.

Thorsten Meyer published a 2027 AI scenario on Nov. 18 arguing that AI systems could help build better AI quickly enough to compress a year of progress into about two months, before shortages of computing capacity slow development. The article labels this a scenario, not a forecast established as fact; its evidence also includes substantial limits and counterarguments.

The proposed mechanism is a feedback loop: AI systems take on more of the coding, debugging and experimentation involved in developing subsequent systems, which could make each development cycle faster. Meyer’s scenario then has progress meeting a physical ceiling in chips, data-center buildings and electricity. A slower phase of automation in factories and laboratories would follow, potentially expanding the amount of compute available and the range of work AI can perform.

One public measure cited is METR’s 80% time horizon: the estimated length of software task an AI agent can complete with 80% reliability. METR’s recorded figure rose from 0.89 minutes for GPT-4 in March 2023 to 185.9 minutes, roughly three hours, for an early Claude Mythos Preview in April 2026. But METR marked its public page as no longer actively updated on Sept. 8, 2026, and its benchmark does not measure general intelligence or all economically valuable work.

The article’s projections depend heavily on which historical data points are fitted. Meyer says a fit using measurements since 2024 would double the horizon about every 3.4 months and pass 16 hours in January 2027; a fit starting in 2019 would double about every 6.4 months and reach that point in June 2028. Those are model-based extrapolations, not measured outcomes. The article also reports that its hyperbolic fit performed worst in backtests, reaching infinity in the projection by Jan. 2, 2026.

At a glance
analysisWhen: Published November 18, 2026; the scenar…
The developmentThorsten Meyer published a scenario arguing that AI-assisted AI development could sharply accelerate progress by 2027 before physical compute limits intervene.

Why Compute Could Set the Pace

The scenario matters because it links AI capability to the speed at which companies can build and deploy computing infrastructure. If AI tools materially shorten the research and engineering cycle, development could move faster than a yearly product timetable suggests. If the supply of chips, electricity or data-center capacity cannot keep pace, however, faster software development would not automatically produce faster real-world deployment.

The source cites U.S. data-center construction running at an annual rate of $96.5 billion in September 2026, compared with $50.7 billion a year earlier. It also cites Epoch AI’s view that power may be the first scaling constraint and that new power supply can take three to five years after a decision. These figures point to infrastructure pressure, but do not establish when or whether a specific AI compute shortage will occur.

Adoption is a separate issue from capability. 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 do not show that a sudden capability jump would translate immediately into widespread workplace change. The scenario’s effects on jobs and businesses depend on costs, reliability, deployment and organizational choices as well as model performance.

Products Worth Considering

Canon SELPHY CP1500 Wireless Photo Printer Bundle with 108 Sheets of KP-108 Paper, 3 Ink Cartridges, and Cleaning Cloth

Canon SELPHY CP1500 Wireless Photo Printer Bundle with 108 Sheets of KP-108 Paper, 3 Ink Cartridges, and Cleaning Cloth

As an affiliate, we earn on qualifying purchases.

What the Benchmark Measures

METR describes its time horizon as the task duration, measured by human expert completion time, at which an agent is predicted to succeed at a specified reliability. The 80% figure means success on four out of five attempts. The evaluated tasks mainly involve software engineering, machine learning and cybersecurity, rather than a representative sample of all work.

METR cautions that its tasks are cleaner and more clearly specified than many economically valuable jobs. It also says measurements above 16 hours are unreliable with its current task suite. Of 31 tasks estimated to take humans at least eight hours, human completion times were measured for only five; the rest were estimates. These limits make the benchmark useful for tracking a narrow class of agent tasks, but not a direct clock for the arrival of general-purpose AI or the pace of economic change.

The source also distinguishes a claimed acceleration in AI research from the much faster pace required by the headline scenario. It says METR estimates Anthropic is improving models on well-specified tasks at about twice the earlier rate, while allowing that the rate could be 1.3 or three times. A year of progress in two months would imply about six times the prior pace, so the cited estimate does not itself establish that outcome.

“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 article identifies multiple links that could fail: AI may not take on enough of the work of developing successor systems; benchmark gains may not translate into faster research progress; and physical infrastructure may limit expansion before the proposed feedback loop has a large effect. The METR trend also changes substantially depending on the starting year selected, while its public page is no longer actively updated and the task suite has stated measurement limits.

It is also unclear whether the cited infrastructure spending will translate into usable power and compute quickly enough to change model development. The article does not establish a timeline for a compute ceiling, the scale of any shortage, or the timing and impact of automated factories and labs. Its discussion of OpenAI’s GPT-6.2 Astra meeting the company’s “High” threshold for AI self-improvement is a company classification cited by the source, not independent proof that recursive improvement is already driving rapid capability gains.

Meyer says the article was researched and drafted with AI assistance, including Anthropic’s Claude. He discloses that Anthropic is among the labs discussed and says Anthropic, OpenAI and METR did not review or sponsor the article. The underlying scenario remains the author’s analysis, not a jointly endorsed conclusion by those organizations.

Signals to Watch in 2027

Meyer’s article says the scenario should be tested against future evidence rather than accepted on its timeline alone. Relevant signals include whether independently measured software-agent task horizons continue to rise, whether AI systems take on a larger share of AI research and engineering, and whether those changes produce measurable acceleration in model development. Updates to the benchmark and clear documentation of its methods would help readers distinguish new measurements from extrapolation.

Infrastructure is another test. Data-center construction, access to chips and power, and the time required to bring new electricity supply online could show whether compute capacity is expanding fast enough to support higher workloads. Evidence of constraints would need to be separated from general growth in spending: construction totals alone do not establish that usable compute is scarce or that a specific model-development program has been delayed.

Finally, adoption and labor effects should be tracked separately from benchmark records. The source’s cited business-use and job-cut figures provide a snapshot, not a forecast. Whether more capable systems change work at scale will depend on deployment and practical performance. The article gives no confirmed milestone that would settle the 2027 question; for now, the claim remains a scenario whose central steps are open to testing.

Key Questions

Is a year of AI progress in two months a confirmed forecast?

No. Meyer presents it as a possible 2027 scenario, not an established prediction. The article discusses evidence, counterarguments and signals that could support or weaken it.

What does METR’s 80% time horizon measure?

It estimates the length of tasks an AI agent can complete successfully four times out of five, based on human task-completion time. METR’s cited tasks mainly cover software engineering, machine learning and cybersecurity, not all forms of work.

Why might compute limit AI progress?

More AI workloads require chips, data-center capacity and electricity. The source cites Epoch AI’s view that power may be the first scaling constraint and says new power supply can take three to five years after a decision; neither point establishes a specific shortage date.

Does the benchmark show that AI is already improving itself six times faster?

No. The article cites METR as estimating roughly a twofold improvement rate on certain well-specified tasks, with a possible range of 1.3 to three times. Meyer’s sixfold pace is what the two-month scenario would require, not a measured result reported in the source.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Why China’s Hands-On Approach Is Reshaping The Future Of AI

China is making significant progress in domestic chip manufacturing, signaling a shift in AI development capabilities, with implications for global tech leadership.

OpenAI Splits Its Robotics Team As Runway’s Praxis-1 Opens New Possibilities

OpenAI moved robotics staff into other teams, while Runway released open-weight Praxis-1. Early tests show a trade-off between task success and speed.

AI’s Persistent Radar: A Key Asset For Corporate And Government Resilience

Commercial SAR satellites are transforming resilience strategies for enterprises and governments, offering persistent, weather-independent monitoring capabilities.

The Bubble Is Not in Valuations: It’s in the Productivity Gap

New data shows AI’s impact on productivity remains minimal, challenging market valuations and expectations. The real bubble is in management assumptions, not stock prices.