Damodaran: AI crash could hit harder than dot-com bust

Aswath Damodaran, a finance professor at New York University, warns a potential crash in the AI sector could be more painful than the dot-com bubble around 2000. On the podcast "Intangible Economy," he argued that AI’s reliance on large, debt-financed physical infrastructure and per-use compute costs make the industry vulnerable to a deep correction.

How does Damodaran say AI’s business model differs from traditional software?

Damodaran says AI is not a traditional, capital-light software business: it requires massive physical infrastructure, much of it financed with debt, and every additional use burns compute rather than moving costs toward zero. He compares AI’s per-use compute to Spotify paying for each stream, and contrasts that with Netflix, where high content costs are spread across growing subscribers. This makes economies of scale "far weaker" for many AI businesses and leaves growth paired with thin margins that "could actually destroy value," he said on the podcast.

Damodaran also flagged a specific competitive pressure: price erosion from Chinese competitors, naming Deepseek as an example. He noted margins are already low, which compresses the room to absorb rising infrastructure costs or debt servicing.

What are the risks to companies, workers and markets?

Damodaran warns a correction would not only hit shareholders but could ripple across society because of the debt backing AI infrastructure. He contrasted the current wave to the dot-com era, saying the fallout could be "more painful than the bursting of the dot-com bubble around 2000." On the upside-turn scenario he calls the "AI fever dream," he warned that if AI’s biggest promises come true, the model becomes about replacing entire jobs, not selling tools, and could lead to "half of white-collar workers" losing their jobs.

He said big tech firms are entering unfamiliar territory by shifting from capital-light operations to building "massive factories and infrastructure" that will be depreciated over ten years but could be obsolete after five. That combination of heavy capex, debt financing and potential rapid obsolescence creates a pathway to steep write-downs and broader financial strain. Damodaran added a personal portfolio note: he owns five of the seven so-called "Magnificent Seven" stocks and has held Amazon on and off since 1997, which frames his concern about how these companies are changing.

He was blunt about the scale of the bet large companies are making: "I'm not sure they really know what they're getting themselves into," he said, arguing investors must now analyze capital expenditures and depreciation in ways they did not when these companies were capital-light.

Why it matters

The shift from software that scales to near-zero marginal cost toward compute-intensive, asset-heavy AI changes how valuations should be built. Investors who assumed exponential margin expansion may find those assumptions invalid when firms face rising depreciation schedules, debt service and the risk of rapid obsolescence. The social angle is material: if the bull case means substantial job replacement, the economic cost will extend well beyond balance sheets to labor markets and public finances.

Damodaran’s comments force a re-evaluation of common valuation anchors: growth alone can destroy value when tied to per-use costs and thin margins, and systemic risk spreads when large players fund infrastructure with leverage.

What to watch

Look for large-scale capex write-downs, rising depreciation charges, or debt distress at companies investing heavily in AI infrastructure, and for aggressive price moves from competitors such as Deepseek that could further compress margins. Also track concrete labor-market signals tied to automation, since Damodaran flags the possibility that up to "half of white-collar workers" could be affected if the replacement scenario plays out.