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

Dot-Com Peak vs AI Boom: The Numbers

The table sets the dot-com era beside the AI boom on the measures that defined the 2000 bubble: valuations, the bellwether stock, infrastructure spending, how much of the new capacity was used, how it was funded, and how many new listings had profits. Every figure is dated and sourced, and ratios marked as Ecomerate calculations are our arithmetic on the cited figures.

Dollar figures from the 1990s are nominal, not adjusted for inflation, and the two eras use different definitions in places. Read the rows as orders of magnitude rather than exact like-for-like comparisons.

What Happened in the Dot-Com Bubble

The Nasdaq Composite closed at a record 5,048.62 on March 10, 2000. Over the next 31 months it fell about 78%, to 1,114.11 in October 2002, and it did not close above the 2000 record again until April 23, 2015.

Cisco was the emblem of the boom. On March 27, 2000 it became the world's most valuable company, worth about $569 billion at its high, according to Cisco. For the fiscal year to July 29, 2000, it reported net income of $2.67 billion on sales of $18.93 billion, so the market valued it at more than 200 times a full year of earnings, or about 146 times the $3.91 billion pro forma figure Cisco also reported, which excluded acquisition charges and investment gains.

The telecom build-out left a glut. Telecom companies laid about $90 billion of fiber-optic cable between 1997 and 2001, according to the Federal Reserve Bank of Dallas, and usage of the long-distance backbone fell below 3% by April 2001. From March 2000 to September 2001, the S&P communications-equipment index fell 86%, erasing $793 billion of shareholder equity.

New listings showed the same pattern. In Jay Ritter's IPO data, 76% of US companies that went public in 1999 had negative earnings per share, and their shares rose an average of 71.2% on the first day of trading.

AI Capex by Company (2026 Guidance)

The five companies below gave their latest capital spending guidance between July and September 2026. Microsoft, Alphabet, Amazon, and Meta guide on a calendar-year basis, and together that comes to roughly $720 billion to $745 billion for 2026 by our sum of their ranges.

The figures are not perfectly comparable. Microsoft counts finance leases but not operating leases, and said in July 2026 that moving more data-center leases to operating leases lowered its calendar 2026 figure to about $175 billion while its underlying plans were unchanged. Meta includes principal payments on finance leases, and Oracle reports gross spending, part of which customers prepay.

Dot-Com Bubble vs AI Bubble: Where They Match

• • Spending ahead of proven returns: in both eras, companies committed capital on forecasts of demand. Telecom carriers kept building until backbone usage fell below 3% in April 2001, and Microsoft said in July 2026 that its capex for the fiscal year to June 2027 will grow again.
• • One supplier as the bellwether: Cisco then, NVIDIA now. Data Center made up $89.0 billion of NVIDIA's $96.2 billion of revenue in the quarter to July 26, 2026, so its results track the build-out directly.
• • More outside money: the telecom boom ran on debt. AI spending is still mostly funded from operations, but Alphabet raised $49.6 billion of equity and $20.3 billion of notes in the second quarter of 2026, and its long-term debt more than doubled from the end of 2025 to $98.2 billion.
• • Loss-making leaders funded by investors: like many 1999 IPOs, the leading AI labs lose money and rely on outside capital. Anthropic is preparing an IPO, according to the draft prospectus Reuters reviewed.
• • Valuation stretched on some measures: the S&P 500's trailing P/E was 26.0 on October 16, 2026, above its 10-year average of 23.6, even though its forward P/E was in line with its average.

Dot-Com Bubble vs AI Bubble: Where They Differ

• • Lower market-wide valuation: the S&P 500 traded at 19.0 times forward earnings on October 16, 2026, against 24.4 at the March 2000 peak, according to FactSet.
• • Profitable bellwether: NVIDIA's net income in one quarter, $59.7 billion, was more than 20 times Cisco's for the whole of fiscal 2000, $2.67 billion.
• • Incumbent spenders: most AI capex comes from companies with large, profitable core businesses. Meta earned $15.85 billion in the second quarter of 2026 at a 31% operating margin. The telecom boom, by contrast, brought a deluge of new carriers and unprecedented debt issuance, according to the Dallas Fed.
• • Revenue arriving: Google Cloud revenue rose 82% to $24.8 billion in the second quarter of 2026, and its operating income more than tripled to $8.8 billion. In April 2001, usage of the long-distance backbone was below 3%.
• • Calmer IPO market: 53% of 2025 IPOs had negative earnings and the average first-day gain was 29.3%, against 76% and 71.2% in 1999.
• • Earnings growth: FactSet's blended estimate of S&P 500 earnings growth for the third quarter of 2026 was 29.0% as of October 16, 2026.

Is AI Losing Money?

Some AI companies are losing money, heavily. OpenAI burned about $3.7 billion of cash in the first quarter of 2026 on $5.7 billion of revenue, The Information reported, citing documents shared with shareholders. Anthropic lost about $42 billion in 2025 on revenue of nearly $4.6 billion, according to a draft IPO prospectus reviewed by Reuters; about $34 billion of that was a non-cash accounting charge, and the operating loss was more than $8 billion.

The companies selling to them are profitable. NVIDIA earned $59.7 billion in the quarter to July 26, 2026, and Google Cloud's operating income rose to $8.8 billion in the second quarter of 2026 from $2.8 billion a year earlier.

The biggest spenders sit in between. Their core businesses earn a great deal, but capex now absorbs most or all of their operating cash flow: Alphabet's free cash flow was negative $5.9 billion in the second quarter of 2026, Amazon's was negative $7.6 billion over the 12 months to June 30, 2026, and Oracle's was negative $5.4 billion in the quarter to August 31, 2026.

Reported profits also include paper gains on stakes in AI labs. Amazon's second-quarter 2026 net income of $62.6 billion included $53.4 billion of pre-tax non-operating income, primarily from its investments in Anthropic, and Alphabet's included $98.0 billion of other income, primarily net unrealized gains on equity securities.

When Did the AI Bubble Start?

There is no official start date. Most accounts date the AI boom to November 30, 2022, when OpenAI released ChatGPT. Bubbles, if that is what this turns out to be, are dated after the fact from the peak.

NVIDIA's results show the scale since then. Its Data Center revenue was $3.83 billion in the quarter to October 30, 2022, the last full quarter before ChatGPT's release, and $89.0 billion in the quarter to July 26, 2026, about 23 times as much.

Is AI a Bubble? What to Watch

An AI investment bubble would show up in the returns on all this spending, and those will take years to measure. These are the measures that would point one way or the other as new filings arrive.

• • Cloud and AI revenue against capex: whether revenue keeps growing fast enough to cover the depreciation that today's spending will create.
• • Free cash flow and outside funding: whether negative free cash flow at the largest spenders proves temporary, and how much new debt and equity they raise.
• • AI lab economics: whether OpenAI and Anthropic narrow their losses as revenue grows, and what audited figures show if they list.
• • Backlog conversion: how quickly contracted backlog, such as Oracle's $664 billion of remaining performance obligations, turns into revenue.
• • Valuations: whether forward P/E ratios stay near long-run averages or move toward 2000 levels.
• • Capacity use: whether new data centers run close to full, the opposite of the idle fiber of 2001.