What happens when the AI bubble pops?

Each bubble's area is a company's market value. Dashed rings show today's size.

Selected companies, value change

–

S&P 500 proxy stress

–

US output shortfall, year 1

–

Unemployment pressure, year 1

–

Set the scenario

A couple dozen companies carry the AI build-out, from the chips to the power plants. Pick a scenario or set your own assumptions to see which of them deflate, by how much, and what that does to the US economy.

Where the shock lands

Scale the demand drop for each layer of the stack. 1.0× = the headline number.

Prices and disclosures are dated observations; the scenario inputs are assumptions you can change. Each bubble holds a company's whole business, not only its AI revenue.

Data dates & coverage

Follow the money

Figures from company reports, each with its date, to give the scenario some context: how AI makes money, how much cash goes into it, and where funding is tight. Companies use different time periods and definitions. These are chosen examples, not a live ranking of companies.

Stock by stock

Two methods use the same scenario inputs. Fundamentals cuts a company's earnings or total value. Factor stress uses how much each stock has moved with the market and with chip stocks in the past. If the two agree, that does not prove either one is right. The bars show a rough range, not a measured probability. Open a company to see its assumptions and regression details. Click a column to sort.

Company	Market value	Valuation basis	Fundamentals	Factor stress	Scenario estimate	Historical comparison

Company assumptions & sensitivity ranges

These numbers are selected scenario inputs, not estimated coefficients or measured pre-boom P/Es. Issuer links support the business context. The company-assumptions slider moves between the stated sensitivity endpoints.

The way down

An assumed path shape, alongside monthly closes rebased to March 2000 month-end. Historical comparisons do not validate the scenario.

Where it hurts

Fall in market value by layer of the stack.

The economic knock-on

A rough estimate of how much less the economy produces than it would with no crash. Two things drive it: less investment at home, and households slowly spending less as their wealth falls. Year 1 is the average of the first twelve months of the scenario. This is not a GDP growth forecast.

Macro assumptions

How the model works

A stress test based on research. Real data sets the starting point. The assumptions decide what happens in the "what if" world.

Fundamentals

A share price is earnings times a multiple, so a crash is some mix of lower earnings and a lower multiple:

P₁/P₀ = (1 − ε·d) × (1 − m·resettable share) × (1 − c)

d is a miss against expectations. ε is an assumed exposed-revenue share times an assumed earnings amplification. The resettable share is the forward P/E above an assumed normal, not an identified AI premium or observed historical norm. m resets that share and c assumes a general-market decline. P/E also reflects growth and accounting choices; this identity is not the Campbell–Shiller present-value decomposition.

CoreWeave and Nebius instead use enterprise value: stress revenue and the valuation multiple, then subtract dated net debt. Cash, debt and share counts are held fixed; refinancing, cash burn and dilution remain outside the calculation.

Factor stress

For every stock we regress two years of weekly returns on two factors:

rᵢ = α + β_mkt·r_SPY + β_semi·SEMI + ε

The second factor is SMH beyond its market exposure, which includes semiconductor influences unrelated to AI. Fundamental scenario returns are weighted using dated IVV holdings (an S&P 500 proxy) and actual SMH holdings, then pushed through the measured betas. Unmodeled holdings receive only the assumed general-market decline. Both calculations share assumptions; neither is an independent check on the other.

Blend, range and history

The estimates are averaged in log space with the weight you choose. The illustrative band combines residual volatility, model disagreement and company-parameter sensitivity. Basket noise preserves observed cross-stock residual dependence. It has no tested probability coverage and omits factor-shock and coefficient uncertainty. Sensitivity endpoints are choices, not confidence limits.

Dot-com counterparts are selected by business role. Traded analogs use 1999–2003 monthly-close drawdowns; four delisted analogs use approximate fixed losses. The severe preset was originally tuned to the 2000–02 S&P decline. This is calibration, not out-of-sample validation.

The economy

Output gap = (domestic investment loss + phased wealth effect) × (1 − cushion)

Dated global cash investment is annualized as an explicit scenario baseline, then bridged to assumed AI spending, US installation and domestic value added. These are distinct steps. FRED household equity assets anchor the wealth base; an editable portfolio exposure maps it to the scenario. The research estimate of 3.2¢ of annual spending per lost dollar at two years phases in over an assumed 24 months, applied to each monthly wealth loss. Output effects are divided by current-dollar GDP as a first-order volume approximation with prices held fixed. Okun's coefficient gives illustrative unemployment pressure, not predicted layoffs or a dated unemployment peak.

What this is not

A forecast. It answers "if this happens, then roughly what?", not "will it happen?". Nothing here is investment advice.
Precise about each company. Assumptions and their ranges are explicit; forward P/E estimate horizons and accounting definitions can differ across providers.
Crash-calibrated. Historical betas, volatility and correlations can change under stress; the direction of the resulting error is unknown.
A full macro model. It leaves out second-round effects such as credit losses on data-center debt, layoffs feeding back into spending, and dollar and rate moves.

Sources

Prices, market values and P/E ratios: Yahoo Finance (unofficial endpoints, delayed). Index history 1999–2003: Yahoo Finance.
US GDP, observed growth, unemployment, labor force and tech investment: FRED (BEA, BLS). Household equity wealth: Federal Reserve Financial Accounts, including nonprofits, direct and indirect holdings, and foreign equities.
Portfolio weights: iShares IVV and VanEck SMH, normalized over equity holdings. IVV is a tradable proxy, not official index constituent data.
Wealth effect: Chodorow-Reich, Nenov & Simsek, "Stock Market Wealth and the Real Economy", AER 2021 / NBER w25959.
Wealth sensitivity varies with ownership and income: Federal Reserve research (2025). This supports examining a range, not automatically replacing the stock-specific MPC estimate.
Investment and imports: Federal Reserve research (2026); Minneapolis Fed AI Trade Tracker. Neither identifies a universal domestic share.
Unemployment relationship: Ball, Leigh & Loungani, Okun's Law: Fit at 50?. The coefficient is an approximation, not a causal employment forecast.