Is this 2000 again? The P/E says no. The CAPE says look again.

A P/E ratio is a payback period: how many years of a company's earnings its price represents. In March 2000, buyers of the Nasdaq-100 paid about 60 years of expected earnings. Today they pay about 20 to 22. That single comparison settles the loudest version of the bubble debate — and settles nothing else, because the answer changes with every other way of counting.

Cisco's bar spans the two sourced series (Refinitiv 107× forward; FactSet 130.3× forward) — we show both rather than blending them. Nvidia's 16.9× forward assumes its earnings roughly double within a year; see the caveats in Section 03.

Section 01The indexes, basis by basis

Section 02The leaders then: what the peak actually cost

Two reputable data series disagree on the 2000 leaders' multiples, so both are printed. Anyone who shows you one clean number for Cisco is hiding the other series.

"At 10 times revenues, to give you a 10-year payback, I have to pay you 100% of revenues for 10 straight years in dividends. That assumes I have zero cost of goods sold… zero expenses… pay no taxes… with zero R&D for the next 10 years, I can maintain the current revenue run rate. Now, having done that, would any of you like to buy my stock at $64?"Scott McNealy, CEO of Sun Microsystems, on his own stock's peak valuation — BusinessWeek, March 2002

And below the leaders, there were no earnings at all. By Jay Ritter's count (University of Florida), 86% of the tech companies that went public in 1999–2000 were unprofitable; only about 14% of the 600-plus IPOs of those two years were profitable. Pets.com went public in February 2000 on $619,000 of 1999 revenue and was liquidated nine months later. eToys was briefly worth more than Toys "R" Us, which had profits and 1,500 stores; it filed for bankruptcy in March 2001.

Section 03The leaders now

One consistent provider (StockAnalysis), one date — the close of Friday, October 30, 2026. The contrast that frames the whole piece: the most expensive ordinary leader today, on trailing earnings, costs less than the cheapest of 2000's big four cost on forward earnings.

⚠ Two denominators are flattered by one-off paper gains in 2026, and honesty requires flagging them. Alphabet's 14.6× trailing P/E includes a ~$98 billion unrealised gain on private stakes; excluding it, the figure is a little over 23× (Motley Fool). Amazon's trailing earnings likewise include a ~$53 billion revaluation of its Anthropic stake, which is why some providers put its forward multiple (up to 30×) above its trailing one. Nvidia's low forward multiple leans the other way on the same lever: it assumes earnings roughly double within a year.

Section 04What is genuinely different underneath

The profits are real, and they are enormous. Microsoft's operating margin over the last twelve months was 47.1%. The S&P 500's blended net margin in Q3 2026 was 14.9% (FactSet), after a second quarter at 17.0% — the highest in more than fifteen years. The seven largest tech names earned a net margin that peaked at 25.8% (Q4 2024), nearly double the index's 13.4%. In 2000 the big four leaders were profitable too — the absence of profit lived in the IPO tail. Today there is essentially no unprofitable mega-cap cohort at all.

Earnings carry more of the value. The top ten companies are about 40–41% of the index's market value but also about 32% of its earnings — a gap, but a far smaller one than in 2000, when the top names' share of earnings was a fraction of their share of value.

The cash story, though, narrowed in 2026 — flag it. The fortress years were real: S&P 500 buybacks hit a record $1.020 trillion in the twelve months to September 2025, plus $664.9 billion of dividends. But in the first half of 2026 Alphabet and Meta bought back nothing (against $51.2 billion combined a year earlier) while spending $129.7 billion on construction and equipment. Combined quarterly free cash flow of the five largest builders fell from about $61 billion in 2024 to under $5 billion by mid-2026, and the build-out is now partly debt-funded — about 10% of global investment-grade bond issuance, against under 1% in 2024 (IMF). Stronger starting balance sheets than telecom ever had; the same direction of travel.

Section 05What rhymes — the honest counterpoint

• The slow gauges are near their extremes. CAPE at ~41 against a peak of 44.2. Trailing P/E ~26× against ~31×. Price-to-sales 3.78× against a long-run 1.81×. A dividend yield near 1.06% in October 2026 was described as the lowest on record — below even 2000's 1.1%.
• Concentration is not an echo of 2000; it exceeds it. Top ten ≈40–41% of the index versus ~26–27% then. An index fund today is a materially bigger bet on about ten AI-linked names than it was on tech in 2000.
• The building boom is bigger than telecom's. The big four builders' 2026 spending plans total roughly $720–745 billion — mechanically about 2.3% of US GDP, against ~1.2% for telecom at its 2000 peak (other definitions run from $600 billion to J.P. Morgan's broader $950 billion). Telecom left most of its fibre dark for a decade on forecasts that proved directionally right and a decade early. One disanalogy cuts each way: fibre barely ages, while AI chips obsolesce in three to four years — an overbuild cannot hibernate the way dark fibre did. But 2000's builders were debt-funded shells; 2026's were, until this year, paying out of operating cash flow.
• Forward-earnings optimism is load-bearing. The friendly 18.7× exists only if consensus is right: earnings growth of +32.9% for calendar 2026 and ~+16% for 2027 (FactSet). And about $160 billion of 2026's gains are private-stake revaluation gains — paper, not operations; strip them out and the forward multiple is well above 19. Meanwhile the average earnings "beat" in Q3 2026 was only +1.9–2.6%, against a five-year norm near 7.5%: beats are thinning even as growth stays huge.
• Circular financing is back. The IMF's October 2026 stability report counts roughly $300 billion of hyperscaler equity invested in weaker AI firms, backed by ~$650 billion of debt, locking in ~$1.4 trillion of future revenue commitments — structurally the vendor financing of 1999, at larger scale.

"There does not appear to be a valuation bubble, but there may be an earnings bubble."Goldman Sachs strategists Peter Oppenheimer & Christian Mueller-Glissmann, autumn 2026

The two strongest readings, stated fairly. "This time is different": forward multiples are a half to a third of 2000 at index and leader level; the leaders convert revenue to 30–60% margins; index earnings growth is actually printing at +25–30% for three straight quarters rather than being projected from eyeballs; and there is no Pets.com cohort at scale. "It rhymes": every trailing and cyclical gauge sits at or near its 1999–2000 extreme; concentration exceeds 2000 by some fourteen points; the spending boom is twice telecom's relative to GDP and increasingly debt- and circular-financed while the builders' free cash flow collapses toward zero; and the one gauge that looks normal — forward P/E — looks normal only by capitalising +33% earnings growth and record margins indefinitely, the exact assumption that failed in 2000–01.

The IMF's market-stability chief made the same distinction in October 2026: valuations were not stretched because P/Es fell as earnings beat — "the key difference from the dot-com bubble" — while warning that a reassessment of AI profitability "could trigger a market correction" amplified by concentration and leverage. J.P. Morgan's strategists, from the other side, note the largest names' forward P/E relative to the market is at a ten-year low: the de-rating, they argue, is largely complete.

Section 06What happened last time — and what today requires

After March 2000, the Nasdaq Composite fell from 5,048.62 to 1,114.11 — a 77.9% decline over 31 months; the S&P 500 fell 49.1%. The Nasdaq did not close above its peak again until April 23, 2015 — fifteen years later. Cisco fell 88–89% and did not regain its $80.06 high until December 2025, twenty-five years and eight months on, despite its revenue roughly quintupling in the interim. The technology was right. The prices were early.

Today's prices make a specific demand: consensus needs +32.9% earnings growth in 2026 and ~+16% in 2027 merely to validate the current forward multiple at flat prices. At the infrastructure level, Goldman Sachs Research estimates roughly $300 billion of incremental AI revenue is needed for the spending cycle to break even — the live version of the "$600 billion question" Sequoia's David Cahn posed about the annual revenue required to justify the AI build-out. Payback arithmetic (the McNealy test, updated): Nvidia at 16.3× sales must convert sales to cash at margins Cisco never sustained, for a decade; the index at a CAPE near 41 needs roughly a decade of above-trend real earnings growth to grow into its multiple at historical returns — the same arithmetic that, in 2000, took fifteen years and a 78% drawdown to resolve.