Scenario 2027: AI Advances Two Months’ Worth In A Year—Then Compute Runs Out?
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Scenario 2027: AI Advances Two Months’ Worth In A Year—Then Compute Runs Out? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get movie nights delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Thorsten Meyer’s November 18 article proposes a possible 2027 scenario in which AI helps develop more capable AI, accelerating software progress before limits on chips, buildings and power slow it. The article presents this as a scenario, not a forecast established by the data; METR’s task measurements show rapid gains but have limits, and the claimed acceleration is much larger than METR’s current estimate.

Thorsten Meyer published a 2027 AI scenario on November 18 arguing that AI systems could help develop their successors fast enough to compress a year of progress into two months, before shortages of computing capacity slow further gains. The article is explicitly a scenario, not a confirmed forecast; its evidence points to improving AI performance but does not establish that a sixfold acceleration will occur.

The scenario has a software feedback loop at its center: AI systems take on more coding, debugging and experimentation involved in building subsequent AI models. Meyer argues that if this loop tightens, progress could arrive in weeks rather than at a steady annual pace. His estimate of a year’s progress in two months implies roughly six times the pace, a much steeper acceleration than the evidence cited from METR.

METR, a nonprofit that evaluates frontier AI agents, measures the length of software tasks an agent can complete at a specified reliability. Its 80% time horizon is the task duration at which a system is predicted to succeed four times out of five. In the selected record-setting results cited by Meyer, GPT-4 measured 0.89 minutes in March 2023, while an early Claude Mythos Preview measured 185.9 minutes in April 2026. These are observed task results, not projections of when future thresholds will be reached.

The article also reports that METR estimated Anthropic was improving models on well-specified tasks at about twice its earlier rate, while saying it would not be surprised by a rate between 1.3 and three times. Separately, Meyer says OpenAI’s GPT-6.2 Astra is the company’s first externally released model to meet its “High” threshold for AI self-improvement. Those claims do not, by themselves, show that AI development has reached the sixfold acceleration proposed in the scenario.

At a glance
analysisWhen: Published November 18, 2026; the propos…
The developmentThorsten Meyer published a scenario examining whether AI self-improvement could sharply accelerate progress in 2027 before physical compute constraints slow it.

The Gap Between Software and Deployment

If AI systems become substantially better at helping build other AI systems, development timelines and the demand for computing infrastructure could change quickly. But the scenario’s effects would depend on more than software capability: researchers would need access to chips, data-center space and reliable power, and the resulting systems would need to be adopted in real workplaces.

Meyer’s article cites US 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 estimates that 18% of US firms used AI in a business function in late 2025 and early 2026, while 2% reported AI-related job cuts. These figures describe different parts of the picture: rising infrastructure investment does not prove that new capacity is available where needed, and reported business use does not establish broad economic transformation.

Bose QuietComfort Wireless Noise-Cancelling Headphones, Moonlight Grey

Bose QuietComfort Wireless Noise-Cancelling Headphones, Moonlight Grey

Ninja Creami Scoop & Swirl Ice Cream Maker – Bone & Champagne (Amazon Exclusive)

Ninja Creami Scoop & Swirl Ice Cream Maker – Bone & Champagne (Amazon Exclusive)

Bose QuietComfort Ultra Wireless Noise Cancelling Earbuds, 2nd Gen, with 20W 2-Port Wall Charger

Bose QuietComfort Ultra Wireless Noise Cancelling Earbuds, 2nd Gen, with 20W 2-Port Wall Charger

As an affiliate, we earn on qualifying purchases.

What METR’s Measure Captures

METR’s time horizon is a measure of an agent’s performance on a particular set of tasks, not a general measure of intelligence or economic impact. The tasks mainly involve software engineering, machine learning and cybersecurity, and METR says they are cleaner and more clearly defined than many economically valuable jobs. A task lasting two hours is closer, in its description, to work assigned to a new hire or freelance contractor than to the work of an experienced colleague.

The measure also has limits at longer durations. Meyer reports that METR says results above 16 hours are unreliable with its current task suite. Of 31 tasks estimated to take humans eight hours or more, METR measured human completion times for only five; the rest are estimates. METR’s public page was marked no longer actively updated on September 8, 2026, according to the article.

Different statistical fits to the same measurements produce very different dates for reaching a 16-hour horizon. Meyer’s fit using data from 2024 doubles about every 3.4 months and reaches that level in January 2027; a fit starting in 2019 doubles about every 6.4 months and reaches it in June 2028. His most dramatic, hyperbolic fit projected an infinite value in November 2027, but he says it performed worst in backtests and had already projected infinity by January 2, 2026, when fitted to models released before July 2025.

“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, as quoted in his article

How Fast the Feedback Loop Runs

The available measurements do not settle whether AI development will accelerate to the pace Meyer describes. The article’s sixfold figure is a scenario assumption, while the METR estimate it cites is about a twofold improvement rate for a narrower category of work. The different curve fits also yield substantially different timelines, and the METR task suite becomes unreliable above 16 hours.

It remains unclear whether performance on measured software tasks will translate into faster development of frontier models, or whether improvements will generalize to work outside those tasks. The article names power, chips and data-center construction as possible constraints; it does not establish when any one of them will become binding. Meyer also notes that the article was researched and drafted with AI assistance, including Anthropic’s Claude, and that Anthropic, OpenAI and METR did not review or sponsor it.

Signals to Watch Through 2027

The scenario will be easier to evaluate as new, comparable measurements become available. Relevant signals include whether AI agents can reliably complete longer software tasks, whether independent evaluations show that AI is speeding model development, and whether those gains persist across measurement methods rather than depending on one curve fit.

Infrastructure is another test: actual access to computing capacity, data-center construction and power connections will help show whether physical limits are slowing expansion. Business adoption and reported employment effects can indicate whether improved systems are spreading beyond technical development. Meyer’s article does not provide a confirmed timetable for these developments; its central claim remains a possibility to test against future evidence.

Key Questions

Is the article predicting that AI will reach superintelligence in 2027?

No. Meyer describes the article as a scenario, not a forecast presented as fact. He also notes that artificial superintelligence has no agreed definition and focuses on the proposed development mechanism instead of resolving the label.

What does METR’s 80% time horizon measure?

It estimates the task duration at which an AI agent is predicted to succeed four times out of five. The tasks cited are mainly in software engineering, machine learning and cybersecurity, so the measure does not directly represent performance across all jobs.

Does the evidence show a sixfold acceleration?

No. The sixfold pace is what Meyer’s two-month-for-a-year scenario would require. The article cites a METR estimate of about two times the earlier rate for improvements on well-specified tasks, with a possible range of 1.3 to three times.

What could slow AI progress in the scenario?

Meyer points to physical computing constraints, including chips, data-center buildings and power. The article discusses these as possible limits, not as a confirmed bottleneck with a settled date.

Source: ThorstenMeyerAI.com

You May Also Like

AI Market Trends: Prices Drop Because Consumers Are Tight On Cash, Not Fixes

Memory prices are slowing their rise, driven by consumer demand exhaustion rather than supply recovery, indicating a prolonged industry squeeze.

Drone Racing League Introduces Zero‑Gravity Tracks

Keen drone enthusiasts will discover how zero-gravity tracks revolutionize aerial racing, but the full impact on sports and technology remains to be seen.

Understanding DeepSeek-V4-Flash-High: The Ninth Point At $0.25 Per Million In AI

DeepSeek-V4-Flash-High ranks ninth on the Arena leaderboard with a 145-point increase after post-training updates, at a cost of $0.25 per million tokens.

SpaceXAI’s Grok 4.6: AI Breakthrough With Fable 5-Level Power At 80% Off

SpaceXAI announces Grok 4.6, claiming Fable 5-level performance with an 80% discount, but lacks independent verification or detailed technical data.