Case Studies

The Dot-com Bubble

The bubble every AI conversation reaches for. The internet was real, the visionaries were broadly right, and the Nasdaq still lost three quarters of its value.

Updated August 31, 2026

What happened

Between 1995 and March 2000 the Nasdaq rose nearly sevenfold as the internet went from curiosity to consensus future. Companies with no profits, and sometimes no revenue, reached billion-dollar valuations on user counts and "eyeballs." The Nasdaq Composite closed at 5,048.62 on 10 March 2000, its dot-com-era peak, and bottomed at 1,114.11 on 9 October 2002: a 77.9% drawdownHow far a price falls from its peak to its lowest point, in percent. The dot-com crash was a 78% drawdown for the Nasdaq.Full definition in the glossary, 78% rounded, over 31 months (FRED↗ (opens in a new window)). It did not close above 5,048.62 again until 23 April 2015, about 15 years later.

What inflated it

An IPOWhen a private company first sells shares to the public. Waves of speculative IPOs tend to cluster near market tops.Full definition in the glossary machine running at full speed, extreme valuationsWhat investors are paying for a business relative to what it earns or sells. 'Stretched valuations' means prices assume a lot of good news.Full definition in the glossary (leaders at 100+ times earningsPrice divided by yearly earnings: how many dollars you pay for one dollar of annual profit. Higher means more expensive, or more optimism about growth.Full definition in the glossary, many at 10-30 times revenue with negative cash flow), and narrowing leadershipHow many stocks are actually participating in a rally. Narrow breadth (a few giants carrying the index) has historically been a late-cycle warning.Full definition in the glossary as a shrinking set of hot names drove the index (Deutsche Bank via Investing.com↗ (opens in a new window)). Underneath the consumer dot-coms, the telecom sector was running a vendor-financingA supplier lending customers the money to buy its own products. Inflates the supplier's sales until the customers can't pay.Full definition in the glossary engine that funded much of the hardware build-out, covered in depth on the Telecom & Fiber page. Some dot-coms also padded revenue with reciprocal ad deals, each buying ads from the other and both booking the swap as sales, a small-scale round-trip arrangement.

What popped it

No single event. The Federal Reserve raised ratesWhen a central bank raises interest rates or drains money from the system to slow the economy. Most historical bubbles peaked during or within 18 months of tightening.Full definition in the glossary from mid-1999 into 2000; roughly nine months into that cycle, the index rolled over. Marquee earnings disappointments followed, funding for unprofitable companies evaporated, and the cash-burn arithmetic that had been visible all along suddenly mattered (Britannica↗ (opens in a new window)). Notably, valuation concerns were mainstream by 1998, roughly two years before the top: consensus warnings, as usual, ran early.

How far it fell, and how long recovery took

The index fell 78%; most pure dot-coms fell 90-100% and the majority disappeared entirely. Survivors with real business models (Amazon fell over 90% and lived) needed years to rebuild. Index recovery took about 15 years; for the median 1999-vintage internet stock, recovery never came.

Who captured the value

Almost everything the 1999 bulls predicted came true: e-commerce, streaming, online advertising, the death of distance. The value went overwhelmingly to companies that either survived the crash (Amazon, Google, which IPO'd after it) or were founded later on the cheap infrastructure the bubble left behind. Being right about the technology and being paid for it turned out to be almost unrelated skills.

Echoes in today's AI boom

• The honest differences matter. The dot-com bubble was powered by unprofitable startups burning other people's money. Today's AI spending is led by the most profitable companies on Earth, the core of the case against the bubble thesis. The comparison is informative in both directions.
• Valuations rhyme but don't match. Most AI leaders trade well below 1999-style multiples, though pockets (Nvidia's price-to-sales ratioCompany value divided by yearly revenue. Used when profits are small or absent. Above 10 is historically expensive; some AI names trade far above that.Full definition in the glossary, some pure plays) are in that territory. Current readings on the Warning Lights page.
• Concentration is actually worse now under any methodology fund-holdings data supports. The top ten S&P 500 companies held 37.6% to 39.1% of the index as of late August 2026 (SPY/IVV holdings), above dot-com-era concentration. The site's earlier 40.7% "record" figure is dropped; no located methodology reproduces it. See market breadth.
• "Being right about the tech" paid the wrong people. If AI follows the dot-com script, the biggest winners may be companies that buy compute cheaply after a correction, not the ones paying peak prices today. The mechanics are on Building Ahead of Demand. Where this dot-com comparison lands in the overall scorecard is on Conclusions.