# How Much Would an AI Crash Destroy?

> Economists put an AI valuation correction at 20 to 40 trillion dollars, and value and small-cap funds have quietly filled up with the same AI exposure.

Published: 2026-08-25
URL: https://daniliants.com/insights/how-much-would-an-ai-crash-destroy/
Tags: ai-bubble, market-risk, diversification, private-credit

---

## Summary

AI-related stocks have become so dominant that two chip companies (Micron and SK Hynix) produced 17% of the entire global stock market's return in a single month, and "safe" diversification strategies like small-cap or value indexes have quietly filled up with the same AI exposure. Economists' estimates of what a valuation correction (not even a full crash) would erase range from 20 to 40 trillion dollars, five to six times the size of the 2000 dot-com bust, and because stocks recently overtook real estate as the largest component of US household wealth, the fallout would hit the middle of the economy, not just the wealthy.

## Key Insight

- **Concentration is now cross-sector, not just cross-stock**: AI exposure has spread beyond chipmakers into utilities (data center power demand), real estate (server warehouses), and construction (data center building booms). A utility "keeping the lights on in Ohio" is now priced partly as an AI stock.
- **Index rebalancing accidentally hid AI risk in "safe" categories**: the Russell 1000 value index's June rebalance moved overheated chip stocks (Micron, AMD, Western Digital) into the growth index right before they rolled over, and pulled Amazon, Apple, and Microsoft into value right as they bottomed. Investors who thought they'd dodged the AI trade via value or small-cap funds (16 of the Russell 2000's 50 best 2026 performers are semiconductor or chip-equipment firms) were still fully exposed.
- **Three independent loss estimates converge on the same order of magnitude**: economist Dean Baker (roughly $40T if P/E ratios just revert to long-run average, no crash needed), former IMF chief economist Gita Gopinath (roughly $20T US plus $15T foreign wealth in a dot-com-style correction), and Oliver Wyman (roughly $33T), all far above the roughly $6T actually destroyed in the 2000 dot-com crash.
- **Off-balance-sheet AI commitments dwarf reported capex**: the WSJ found roughly $3 trillion in AI-related spending commitments (long-term data center leases, locked-in chip and power purchases) sitting in footnotes rather than the roughly $600B/year of capex big tech actually reports. Alphabet alone carries over $800B of this "iceberg" below the visible number.
- **Private credit is the quiet stress point**: a lightly-regulated $2-3T market that has funded much of the AI and software buildout. Troubled loans at the 20 largest listed private credit funds hit their highest level since 2017, Fitch recorded record private-credit defaults in July, and one large fund reported 7% of its loan book in trouble, echoing pre-2008 opacity risk rather than a 2008-style bank failure (banks themselves are much better capitalized now).
- **Being right about the technology doesn't protect you from the price you paid**: Amazon was the correct pick in the dot-com bubble and still fell roughly 90% from its 1999 peak, taking until 2009 to break even. Most 1999-era search leaders (Infoseek, Lycos, AltaVista, Excite) that looked like the obvious AI-equivalent winners are now trivia answers; Google barely existed until after that bubble burst.
- **Bears calling AI a bubble have a long track record of being early and wrong** (Steve Blank 2011, Marc Andreessen 2014, Mark Cuban 2015, Jim Breyer 2016, Jeremy Grantham 2021 all called tops that didn't arrive on schedule). The lesson isn't "sell everything," it's that timing an exit and a re-entry is nearly impossible even for people who are eventually proven right.
- **The wealth effect makes this a Main Street problem now**: for every $100 of paper stock gains, people spend roughly $3 more in the real economy. Since stocks overtook real estate as the largest share of US household wealth this year (a first since WWII), a correction reaches ordinary households, not just the top 10% who own roughly 90% of shares.
- **Europe is one concrete diversification lever**: tech is only about 10% of the European stock index versus roughly 50% of the S&P 500. European indexes lean into banks, industrials, and healthcare paying around 3% dividends, offering a hedge specifically because they've been unfashionable and haven't absorbed AI-trade speculative pricing.