How long did the dot-com bubble last, and what does it say about the AI bubble?

How long did the dot-com bubble last from start to finish?

Wikipedia dates the rise of the Nasdaq from 1995, the peak to Friday 10 March 2000, and the low to October 2002 (Wikipedia, Dot-com bubble). Counted from 1995 to October 2002, the full cycle lasted about 7 years and 7 months. The rise took about 5 years, and the fall took 2 years and 7 months.

Two features stand out. The rise lasted about twice as long as the fall, and the cycle was far longer than a single year of enthusiasm. A reader who expected a short episode in 1996 or 1997 would have waited years for the burst, which is the experience that makes timing so difficult. For the AI bubble the first lesson is humility: a count of years describes the past cycle well and describes the present cycle poorly, because the end date is unknown while the cycle runs.

How long did the dot-com bubble rise before the peak?

From 1995 to March 2000 the Nasdaq climbed to 5,048.62 (Wikipedia, Dot-com bubble). The climb ran for about 5 years, with pauses and sharp setbacks inside it. Each setback ended with a recovery, and each recovery taught buyers that dips were opportunities. That habit is what extends a bubble.

Long rises also change who holds the asset. Early buyers sell to later buyers, and later buyers have shorter memories of risk. By the final year, new participants dominate the market. The count of years therefore matters less than the change in who is buying and why. Ask of any market today how many participants joined in the last 12 months and what they know of the previous downturn.

How long did the fall last after the dot-com bubble burst?

From the peak on 10 March 2000 to October 2002, the Nasdaq Composite lost 78% of its value over 2 years and 7 months. Most internet stocks had already lost 75% by November 2000, so the steepest drop came in the first 8 months. The slower years that followed wore down the companies that survived the first wave. Counting by quarters, the fall covered about 10 quarters, which is long enough for a company to burn through a full funding round.

A fall in two stages gives a practical rule for planning. Expect a fast first leg driven by funding and a slower second leg driven by results. A business that reaches the second leg with cash and customers has a path through it, and a business that enters it without them faces the harder years alone.

Why do bubbles like the dot-com bubble last longer than forecasts?

A forecast needs a date, while a bubble needs a belief to end. Warnings can circulate for years before prices fall, because each warning that fails makes the next one less credible. The Barron’s article of 20 March 2000 was one such warning, and it landed at the peak. Warnings of that kind help most when they arrive with evidence on cash and customers.

Funding also stretches the duration. As long as new money arrives, weak companies stay alive and prices stay high. The bubble ends when the supply of new money slows, not when someone proves it is a bubble. Track funding conditions and you track the true clock of the cycle.

How does the AI bubble run-up compare with the dot-com years?

The EQOS Asia AI study records a ×4 valuation rise over 18 months (EQOS Asia AI study, April 2026). Eighteen months is much shorter than the 5 years of the dot-com rise. A short run-up can mean a faster cycle, or it can mean an earlier stage of a longer one. The data cannot yet settle which. Readers who need the underlying numbers can open the study and recompute the score.

The study also scores the enterprise AI market at φ(S) = 0.337, in the fragile zone, and it records about $200B a year invested. Those readings describe the state of the system and its tolerance for a shock. They do not give a date, and the study states a bifurcation verdict, which leaves the soft landing and the break both open.

Which measures beat duration for judging an AI bubble?

Replace the question of how long with 3 measures. First, the pace of new funding. Second, the share of customers who pay for results. Third, the gap between valuation and earnings. Each measure moves when conditions change, and each one gives a warning before the price does. Duration gives no warning because it is known only afterwards.

Record each measure every quarter and note the direction. Three measures moving the wrong way in the same quarter justify a review of your commitments. The method asks for a short page of notes, and it replaces a debate about dates with a record of evidence.

What should you do when nobody can time the AI bubble?

Prepare for several durations at once: a short cycle, a long cycle, and a soft landing. Size commitments so that your business works in all three, and stage large purchases. The EQOS Asia method page explains how scenarios feed a recomputable score. This page is general information, not investment advice.

The Λ simulator gives a φ(S) estimate in 10 minutes, and a 30-minute pre-diagnostic applies the scoring to a decision in Singapore, ASEAN, or Hong Kong. You bring the decision, and the session returns a score and a ranked set of options.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile