Find this research useful? Add aibubblequestion.com to your Google Preferred Sources, one click, affects only your own results.

The AI Bubble Question
All pages

The Evidence

Outside Shocks

Historically, shocks don't cause bubbles to pop on their own. They pull triggers that leverage and fragility have already loaded. Here is what's sitting in the chamber, mid-2026.

Updated

The historical record is consistent: wars, embargoes, and supply crises accelerate collapses; they almost never cause them without pre-existing leverage (NBER). The 1997 Asian crisis needed years of foreign-currency borrowing before a currency attack could detonate it. So the right question is not "will a shock happen" but "which shocks could interact with the fragilities already on the board." The research grades five, in descending order of trigger potential.

1. Taiwan and the chip chokepoint (high)

Nearly all cutting-edge AI chips are made by TSMC, mostly in Taiwan, and the binding constraints have tightened: advanced packaging (CoWoS) capacity was described as sold out through 2025 into 2026, though tightness is reported narrowing from about 20% to about 10% by the end of 2026 as TSMC's own capacity expands. SK Group's chairman told reporters of a 20%+ wafer shortfall for AI memory (HBM) that could persist to 2030; that is a company chairman's oral forecast, not a measured shortage, and no underlying transcript or filing was located to support it (Reuters, 16 Mar and 2 Jun 2026; TrendForce, 15 Jun 2026, counterpoint on CoWoS). Near term, scarcity props up chip prices. The tail risk is a Taiwan crisis, blockade, or disaster that interrupts TSMC: a single event that would reprice the entire AI complex at once. The research rates this among the most serious hard fragilities in the cycle.

2. Oil and the Strait of Hormuz (medium-high)

Since the US-Iran conflict began in February 2026, the Strait of Hormuz, normally carrying about a fifth of the world's oil, has been repeatedly disrupted; flows fell to an average of 2.7 million barrels a day in March through May 2026, down from roughly 20 million before the conflict (International Energy Agency, 22 Jun 2026). Brent crude spiked above $110-120 in the spring. The US Energy Information Administration forecasts Brent averaging around $85 a barrel in the third quarter of 2026; settlement-adjacent pricing for 28 Aug 2026 showed WTI near $83.40 and Brent near $88.10, via a financial data aggregator rather than a primary exchange feed (EIA, Short-Term Energy Outlook, 11 Aug 2026). Oil sustained above roughly $100-120 during leveraged booms has historically helped tip markets over (1973-74, 1979-80, 2007-08). Current levels are elevated but not crippling; a durable blockade would change that.

3. The power grid (medium-high, sector-specific)

Electricity has become the binding constraint on the data-center build-out. Texas's grid operator (ERCOT) had a large-load interconnection queue of approximately 474 gigawatts as of late July 2026, about 90% of it data centers, roughly double the earlier 226-gigawatt reading and still growing fast; it stood at about 410 GW just three months earlier. This is a request queue, not built or operating capacity; only 5.9 GW was observed energized plus 3.2 GW approved-not-operational as of April 2026 (ERCOT). The mid-Atlantic grid (PJM) projects its own summer peak load climbing to 220 GW by 2040, a fifteen-years-out demand forecast, not an interconnection queue, and it should not be read alongside ERCOT's live request figure as though the two measure the same thing (PJM Inside Lines). The national generator-and-storage interconnection queue, a different and supply-side figure, stood at 2,061 GW at year-end 2025, down 10% from 2024 after more than 750 GW of unviable requests were withdrawn; that is the opposite direction from continued growth past 2.2 TW, and the two queues, ERCOT's rising load requests and this contracting national supply queue, should be read as separate trends rather than one number (Lawrence Berkeley National Laboratory, 1 Jul 2026). If AI valuations assume capacity arriving faster than the grid can deliver, revenue simply arrives late, the same capacity-timing mismatch that sank the telecom carriers. This is also the core of Ed Zitron's announced-versus-built critique: about 114 GW of AI data centers announced worldwide, his own figure, reproduced accurately here. The paired "under construction" comparator, previously given as roughly 15 GW, has no reproducible methodology behind it in his post or anywhere else located across three independent search passes, so it is marked unverifiable here rather than treated as a confirmed figure; see the Ed Zitron record for whose figures these are.

4. US fiscal capacity (medium, an amplifier)

The US enters any downturn with a deficit of 5.8% of GDP and net interest costs of 3.3% of GDP in FY2026 (Congressional Budget Office, "The Budget and Economic Outlook: 2026 to 2036," February 2026). Debt held by the public is 101% of GDP in FY2026, not near 120%: 120% is CBO's own projection for FY2036, a decade out, and the current-year and decade-out figures should not be read as one "now" number. Fiscal overhangs have historically been amplifiers rather than triggers, but they constrain the rescue: if an AI unwind coincided with limited room for stimulus and a bond market demanding higher yields, a correction would have less cushion beneath it than in 2008 or 2020.

5. Fertilizer and food prices (background)

Hormuz disruption has also pushed fertilizer costs sharply higher: the World Bank's fertilizer index rose over 12% in Q1 2026 to its highest since October 2022, and is projected to rise more than 30% in 2026 before easing; urea prices climbed 80% from February to April 2026 and are projected up nearly 60% for the year, both higher than this page previously stated (World Bank, "Fertilizer prices surge as Strait of Hormuz disruptions tighten supplies," 14 May 2026). US farm incomes were already weakening, with USDA cutting its 2025 net farm income forecast by $25 billion (Farm Bureau). Food-price pressure feeds inflation and squeezes households; it worsens downturns rather than starting them. Semiconductor gases (helium, neon), a genuine constraint in 2022, have largely normalized and rank as background risk too.

How a trigger actually works

None of these forecasts a crash. The historical mechanism runs: a shock raises costs or blocks revenue; a leveraged structure somewhere, a neocloud with GPU-backed debt, an SPV with one tenant, misses expectations; credit spreadsThe extra interest a risky borrower pays compared to the safest borrower (the US government). Widening spreads mean lenders are getting nervous.Full definition in the glossary widen; funding closes; and selling begets selling in whatever was most crowded, which today means the most concentrated index in modern US history. Remove the leverage and the same shock is just a bad quarter. That is why this site spends more pages on the financing structure than on the headlines.