The Fed blames an AI investment bust for a contraction

Where things stand (15 October 2026)

Nothing qualifies yet, and the economy is not contracting.

• FOMC statement, 16 September 2026. The Fed raised rates to 3¾–4%. The statement said "Economic activity is expanding at a solid pace… capital investment is robust." Under Chair Warsh, statements are very short and have never named AI. (https://www.federalreserve.gov/newsevents/pressreleases/monetary20260916a.htm)
• GDP. Q2 2026 grew 2.2% (third estimate, 30 September). The Atlanta Fed's GDPNow puts Q3 at about 3.6% (8 October).
• July 10, 2026 Monetary Policy Report. It uses explicit AI language throughout:
  • "Capital spending growth has been brisk, reflecting strong demand for investment related to artificial intelligence."
  • "AI-related categories."
  • Equity prices driven by "optimism about artificial intelligence."
  • Q1 GDP growth was "supported by robust growth in high-tech business investment." (https://www.federalreserve.gov/monetarypolicy/files/20260710_mprfullreport.pdf)
• No February 2026 report. The House committee never scheduled Powell's hearing (Reuters, 22 June 2026). So the report usually comes twice a year, but not always.
• September 2026 minutes. They say the AI buildout "continued to surprise to the upside."
• AI stocks. A sell-off is under way: chip stocks fell about 12% over seven sessions after a hot inflation report (CNBC, 14 October). It is a correction, not a bust.
• Hyperscaler spending. Capital spending by the largest cloud companies is about $700bn in 2026 and is guided to about $1.1–1.3tn in 2027. Wall Street expects the slowdown to start in 2027 and to be meaningful in 2028 (Fortune, 7 October 2026).

How the question can resolve YES

Three things must all happen:

1. A real-GDP decline in a period after the opening date.
2. An actual fall in AI-related investment or AI share prices.
3. An explicit causal link between the two in the Monetary Policy Report or an FOMC statement.

The terse Warsh-era statements make the report the realistic route. Its timing matters:

• The February report covers the previous year's Q2–Q4, so most of the hazard falls in Q1 quarter-ends.
• The June/July report covers Q1.
• Contractions after Q1 2031 can only be caught by an FOMC statement, which is unlikely to name AI.

Base rates and evidence

Negative GDP quarters. I used the Philadelphia Fed's real-time first-, second- and third-release GDP file (routput_first_second_third.xlsx). Since 1985:

• 72.5% of 18-quarter windows contained at least one negative print.
• 21% of 3-quarter windows did.
• Isolated negative quarters are common: 2012, 2014, 2015, 2022 and 2025.

Professional forecasters. The Q3 2026 Survey of Professional Forecasters (prob.xlsx) puts the chance of a negative quarter at 20–22% for each quarter from Q4 2026 to Q3 2027.

Prediction markets. Kalshi puts a 2027 recession at 20% (6 October). Polymarket puts a recession by end-2027 at about 34%.

Would the Fed name the cause? The report has a record of naming the parts of GDP that drove a decline:

• February 2002: "deep cuts in outlays for high-technology equipment."
• July 2015: the drop in drilling investment "subtracted more than ½ percentage point," but GDP was called "little changed," which would not count here.
• June 2025: GDP "edged down… mostly due to" an import surge.

So if an AI bust coincides with a decline, an explicit attribution is likely, at roughly 70–80%.

Bust scenarios already on the table. Deloitte's downside (September 2026) has a 35% equity correction and investment falling from Q3 2027, with GDP down 1.2% in 2028. The IMF's October 2026 reports flag an AI correction as a key risk.

Model

I built a simple regime simulation:

• A boom-to-bust hazard of about 1% per quarter now, rising to about 4% per quarter from 2028. That gives roughly a 50% chance of a meaningful fall by 2031.
• About a 50% chance that a bust produces a negative GDP print.
• About a 75% chance of explicit attribution, given both.
• A small chance of attribution in contractions with other causes.

This gives about 15–27% by the end of 2031, centred near 20%. I shaded slightly down for Warsh's push toward leaner communications (a communications task force and "shorter statements," per Goldman Sachs).

Strongest case against my estimate

For a higher number. The economy is unusually dependent on AI investment and AI-driven wealth gains, the Fed is hiking into the boom (as in 1999–2000), and the report already speaks in AI terms. Any recession in 2027–2030 would probably involve AI, which could justify 30–35%.

For a lower number. AI lab revenues are growing very fast, spending is constrained by supply, and much AI equipment is imported, so falls in that spending are partly offset in GDP. Spending may plateau rather than fall, and the question requires a strict three-part match.

On balance I land at about 21% by the end of 2031. The near term is close to zero, because spending is still accelerating and Q3 GDP is strong.

Question. By the last day of each quarter, has the Federal Reserve, in an FOMC statement or its Monetary Policy Report, named a fall in AI-related investment or in AI-related share prices as a cause of a contraction in US output?

Definitions.

• FOMC statement: the Federal Open Market Committee's official monetary-policy decision statement, whether issued following a scheduled meeting or an unscheduled decision.
• Monetary Policy Report: the Federal Reserve Board's semiannual report to Congress pursuant to section 2B of the Federal Reserve Act, including its summary, chapters, boxes, footnotes, and appendices.
• AI-related investment: spending or financing that the qualifying document explicitly connects to artificial intelligence, including AI research, software, chips, computing infrastructure, or data centers. Generic technology investment or data-center spending does not count without that connection in the document.
• AI-related share prices: equity prices or equity valuations of companies or groups of companies that the document explicitly identifies as AI-related. A broad-market decline counts only if the document identifies its AI-related component as contributing to the output contraction.
• Fall: an actual decrease in investment spending or financing, or in equity prices or valuations, relative to an earlier level. Slower growth, a reduced forecast, or a level below a hypothetical counterfactual alone is insufficient.
• Contraction in US output: an actual decline in aggregate US real output, identified in the document as a decline in real GDP or as a contraction of overall US output or economic activity. One quarter of declining real GDP is sufficient; neither two consecutive quarters nor an NBER recession determination is required. A sectoral contraction, decline in nominal output alone, or slowing but still positive aggregate growth is insufficient.
• As a cause: the document adopts an affirmative causal attribution, such as “contributed to,” “partly reflected,” or “was driven by.” The AI-related fall need not be the sole or largest cause, and indirect channels such as reduced wealth, consumption, or credit count. An attribution described as likely or probable counts; a mere possibility, correlation, or quoted outside opinion not adopted by the document does not.

Resolves YES if a qualifying document published by the quarter-end identifies an actual US output contraction and explicitly attributes at least part of it to either a fall in AI-related investment or a fall in AI-related share prices. The causal chain may span multiple passages within the same document but must not require an inference supplied by the resolver. Apply the Resolution window and Observation period conventions; any measured contraction used to qualify must concern a period ending after the opening date. A qualifying attribution based on preliminary or revised data counts, and later revisions or retractions do not reverse YES.

Resolves NO if no qualifying publication exists by the quarter-end. Exclude forecasts of future contractions, hypothetical scenarios, warnings about downside risks, and statements that AI-related declines merely weighed on otherwise positive growth. Exclude FOMC minutes, press conferences, the Summary of Economic Projections, congressional testimony separate from the report, speeches, interviews, Financial Stability Reports, Beige Books, FEDS Notes, staff papers, and publications of individual Federal Reserve Banks, unless the qualifying attribution itself also appears in one of the two specified document types.

Resolving source. Only the official FOMC statements published on federalreserve.gov, indexed through the FOMC calendars and information page, and the Board's Monetary Policy Report archive, in official HTML or PDF form. This is a narrower source requirement than Public confirmation (A2); third-party reporting cannot substitute. Use the public publication date under Resolution date (A3), not the observation period, report coverage date, or later testimony date. An explicitly dated correction adding a qualifying attribution counts on the correction's publication date.