Dot-Com Bubble vs AI Bubble: What the Numbers Actually Show

Cisco 1998-2002 vs Nvidia 2023-2026 on one chart

The usual comparison pairs the two companies that sold the equipment each boom ran on. In the late 1990s that was Cisco, whose routers and switches carried internet traffic. Since 2023 it has been Nvidia, whose chips train and run AI models. To compare the two runs fairly, the chart below starts each stock at 100 and measures time in years from the start, not in calendar dates.

Two things stand out. Nvidia's gain is about twice the size of Cisco's, though it took longer to build: 3.4 years to the peak against Cisco's 2.2. And the timing has already diverged. Three years and nine months in, at the end of September 2001, Cisco's line stood near 126, with most of its gain gone, and by October 2002 it was below its 1998 starting point. Nvidia at the same distance from its start, on October 1, 2026, stood at 1,618. Starting points are a choice, of course. Begin Cisco in 1995 and its run looks even bigger. The chart compares shapes, not destinies.

Gains and drawdowns, cycle by cycle

The dot-com rows are finished stories; the AI rows are not. That asymmetry is the trap in most "AI bubble vs dot-com bubble" charts: one line has an ending and the other doesn't yet. What the table does show is that the AI trade has been anything but smooth. Each of the four AI leaders in the table lost 30% to 37% of its value at least once, yet by October 1, 2026 the Nasdaq-100 ETF and the S&P 500 were both within 2% of their record highs.

Valuations: where 2026 looks least like 2000

Cisco closed at a record $80.06 on March 27, 2000, briefly the most valuable company in the world. For the fiscal year that ended in July 2000, it earned $2.67 billion, or $0.36 a share. Divide one by the other and the stock cost about 222 times a full year of earnings, and those earnings included four months after the peak. The Financial Times put Cisco's peak P/E above 200 and its enterprise value at 31 times sales.

Nvidia's multiple has moved the other way during its rise. According to StockAnalysis data, its trailing P/E was about 51 at the end of January 2024, 48 a year later, 38 in January 2026 and 29 on October 1, 2026, with a forward P/E near 19. The price kept rising, but profits rose faster: net income went from $4.4 billion in the fiscal year to January 2023 to $120 billion in the year to January 2026. In the quarter that ended July 26, 2026, Nvidia reported $96.2 billion of revenue and $59.7 billion of net income.

There is a catch in that comparison, and it is Cisco's own history. Its earnings looked solid in 2000 too. Sales still grew 18% in fiscal 2001, to $22.3 billion, yet the company posted a $1.01 billion net loss that included a $2.06 billion charge for excess inventory. A low P/E only protects you if the E holds up. For a supplier, the E depends on how long customers keep ordering.

The broad market tells a less comforting story than Nvidia's multiple. The Shiller CAPE ratio, which compares the S&P 500 with ten years of inflation-adjusted earnings, peaked at 44.2 in December 1999. On October 1, 2026 it stood at 41.1, according to multpl.com, close to that record.

Concentration: the index is more top-heavy than in 2000

Here the AI era goes past the dot-com era. Data from RBC Wealth Management and FactSet show the ten largest S&P 500 companies peaked at about 27% of the index during 2000 and ended that year near 23%. At the end of 2015 they were about 19%. At the end of 2025 they were 40.7%, while they were expected to produce about 32% of the index's 2025 earnings.

That matters for people who think an index fund keeps them out of the argument. With roughly two-fifths of the S&P 500 in ten stocks, including Nvidia and the biggest AI spenders, a plain index fund is a sizable bet on the same theme. The S&P 500 explainer covers what the index is and what it takes to beat it.

Who pays for the build-out

The money behind the two booms came from different places. In the late 1990s, much of the internet build-out was funded with borrowed money and new stock. Young telecom carriers borrowed to lay fiber and buy gear, and equipment makers lent to customers so they could keep ordering. Lucent, for one, had to restate $679 million of revenue in late 2000, partly because sales made on credit fell through.

The AI build-out has mostly been paid for by some of the most profitable companies ever listed. But spending is now outgrowing even their cash. As of late April 2026, Microsoft, Alphabet, Meta and Amazon were guiding to about $725 billion of capital spending for 2026, up from roughly $410 billion in 2025. Alphabet generated $174 billion of operating cash in the twelve months to March 2026, then raised its 2026 capex range in July to $195-205 billion. The funding mix is shifting as a result:

• Alphabet announced $80 billion of new equity on June 1, 2026: $15 billion of mandatory convertible preferred stock, $15 billion of common stock, a $40 billion at-the-market program and a $10 billion placement with Berkshire Hathaway. It said its total debt had passed $100 billion.
• Meta spent $31.1 billion on capex in the second quarter of 2026 and kept $784 million of free cash flow. Its long-term debt was $83.7 billion on June 30, 2026, up from $28.8 billion at the end of 2024.
• Microsoft spent $72.4 billion on capex in the six months to December 2025. Its $625 billion commercial backlog included a $250 billion commitment from OpenAI.
• Nvidia signed a letter of intent in September 2025 to invest up to $100 billion in OpenAI as OpenAI deploys 10 gigawatts of Nvidia systems. A supplier putting money into a customer that buys its hardware echoes the vendor financing of the 1990s.

The difference from 1999 is the size of the cushion: these companies have large, profitable core businesses and the AI spending sits on top of them. The similarity is the direction of travel. More of the build-out now depends on outside capital and on a small number of customers, OpenAI above all, being able to pay.

What the dot-com crash did to the survivors

Being right about the technology did not save investors who paid peak prices. In our dataset, with dividends reinvested, the Nasdaq-100 ETF needed until February 2015 to get back to its March 2000 high. On the same basis and measured from their own dot-com highs, Microsoft took until July 2014, Intel until March 2018 and Cisco until August 2021. Cisco's share price, without dividends, first closed above $80.06 on December 10, 2025, more than 25 years later. On May 14, 2026 it jumped 13% in a day after the company raised its forecast for AI orders, and by October 1, 2026 the stock was up 151% since January 2023, dividends included.

Amazon is the survivor everyone cites. Its adjusted stock fell 94.4% from December 1999 to September 2001 and did not regain the 1999 peak until October 2009. The full dot-com story covers the companies that did not survive, and the S&P 500 spent most of the following decade below its 2000 high, as described in the lost decade article.

Amazon or Pets.com: the problem in 2000

With hindsight Amazon looks like the obvious keeper. In 2000 it did not. That year Amazon lost $1.41 billion on $2.76 billion of sales, and it did not post an annual profit until 2003, when it earned $35 million. Pets.com went public in February 2000 and announced its shutdown that November. On the information available in early 2000, both were loss-making online retailers asking investors to wait for scale to pay off.

Remembering only the Amazons is survivorship bias. Today's version of the question is which AI spenders will earn a return on hundreds of billions of capex, and which suppliers' order books are funded by customers who may run short of money. Nobody in 2026 has the answer key, just as nobody had one in 2000.

So, is AI a bubble?

The evidence points both ways, and an honest comparison stops short of a verdict:

• Like 2000: a supplier stock with a bigger run than Cisco's, index concentration above its 2000 peak, a market CAPE near its 1999 high, capex outgrowing cash flow at some of the biggest spenders, and deals that tie suppliers to customers.
• Unlike 2000: the leaders are highly profitable, Nvidia's P/E is a fraction of Cisco's, the biggest spenders still fund most of the build-out from operating cash, and the AI leaders have already absorbed drops of 30-37% without a collapse.
• Unknown: whether AI revenue grows fast enough to pay for the spending, and whether today's profits are a plateau or a peak. Those were also the unknowns in 2000, and they were only answered later.

Trying it without knowing the ending

Every chart in this article was drawn knowing how the dot-com story ended. In Blindfolio you get real prices, real headlines and companies under codenames, and you decide as the clock moves forward week by week, with a pause whenever you need to think. Pick a random start and the years are masked too, so you do not even know which era you are in. A round that starts in the 2010s may or may not reach the AI years. That uncertainty is the point, because it is what investors in 1999 or 2024 actually faced.