The Nasdaq Advance-Decline Divergence Is Longer Than 2000’s. Is It Another Warning?

Research note: NYSE advances and declines since 1965, Nasdaq since 1998, our Nasdaq-listed common-stock counts since 2016 and capital spending from company filings, through October 16, 2026. Frozen October 16, 2026. How we checked it · How we use AI.

Put yourself in December 1999. The NYSE advance-decline line, the running count of rising stocks minus falling ones, had been sliding for 21 months. Over that stretch the S&P 500 had gained 31%, and on the average day decliners outnumbered advancers by 5.6% of issues. A trader watching that chart saw an index rising on the backs of fewer and fewer stocks. Three months later the S&P 500 made its top, and it fell 49% over the next two and a half years.

The same shape is on the Nasdaq today, and it has lasted much longer. The cumulative A-D line for Nasdaq-listed stocks peaked in June 2021. Since then the Nasdaq-100 has been through a 36% bear market in 2022 and still finished 118% higher at its October 6, 2026 record, while the line kept falling.

So is this another warning? The honest answer has two sides, and both have real evidence behind them. Here is the case each way, with the detail in the sections that follow.

Weighing the two, I land on the second side for now, with one reservation. The divergence is real and long, and it says the market depends on a narrow group of companies. But the breadth that matters for a 2000-style top is breadth among the stocks that carry the index, and there it has held up. That is the thing to watch.

The figures below are fixed at the October 6, 2026 record. In the eight sessions to October 16, 2026, the Nasdaq-100 fell 6.5% and two-year net breadth slipped from −2.57% to −2.88%; the latest-reading box above tracks it. The Nasdaq’s own counts, which start in November 1998, looked like the NYSE’s before 2000: decliners beat advancers by 4.6% of issues a day over the 347 sessions to the March 2000 top. That the Nasdaq line drifts lower over time is well known to chart analysts (StockCharts notes it in its guide); what is new here is measuring today’s divergence against 2000’s and locating the selling inside the Nasdaq.

Half the pace, more in total

The fairest comparison measures each divergence over its own span, from the A-D peak to the index record, so no averaging window has to be chosen.

The pace was half of 2000’s, and the Nasdaq always runs weaker than the NYSE: measured against each exchange’s average day, today’s selling is 2.2 points below normal against 6.3 for the NYSE then. But it has run for 2.7 times as many sessions, so the total is larger.

Fixed windows give different answers, which is why I don’t lean on any one of them. Ranked against each exchange’s typical record day, today’s shortfall is a share of 2000’s that moves with the window:

Which Nasdaq stocks are doing the falling?

Splitting Nasdaq-listed stocks by how much they trade shows where the decline lives.

The 100 most-traded Nasdaq stocks advanced more often than they fell and set a high in June 2026. Ranks 101 to 500 drifted slightly lower. Stocks ranked below 500 supplied 97% of the decline, at more than 3% of issues a day. That is not the picture of 1999, when the deterioration reached the established companies on the Big Board.

Has weak breadth at a record warned before?

Mild weakness has not. Since 1967 the S&P 500 set records on 987 days with two years of NYSE data behind them, and 22% were followed by a 20% fall within two years. On the 190 record days with negative two-year net breadth, the figure was 25%, and none of the 9 first such days became a bear market within two years.

Deep weakness is rarer and has a worse record, on two cases. Readings below −2% at a record came only in April 1968 and November 1999, and both were followed by falls of more than 20%. The Nasdaq has no comparable history: its weakest readings at a record came in December 2023 and January 2024, at −3.39% for Nasdaq-listed stocks, and the index kept rising.

Looking back from the tops gives the same mixed picture. The table lists the NYSE line before every S&P 500 bear market since 1968; by construction it shows only what came before tops, not how often the same readings came to nothing.

Is the AI build-out why it has lasted?

That is my leading explanation, though the numbers below support it rather than prove it. A divergence can run for as long as the companies carrying the index keep delivering, and since 2021 a handful of very large companies have poured cash into datacenters while their suppliers sold to them.

When the Nasdaq A-D line peaked, the five largest capital spenders among the large companies we track made up 26% of their capital spending. Today the five are Amazon, Alphabet, Microsoft, Meta and Oracle, and they make up 49%, spending 83% of their operating cash flow on it against 29% for everyone else.

Profits have grown with the spending. From the end of 2022 to March 2026, total operating profit at the companies building and selling AI infrastructure rose 178% against 22% for everyone else, supplying 66% of the rise across 214 companies. Those are company-wide profits; they do not isolate a return on the AI spending itself. The index and the typical stock parted ways over the same years: the largest US companies weighted by value rose 89% from the A-D peak to August 2026, and weighted equally 47%.

What happens if the AI trade unravels?

The closest history is March 2000. XLK fell 82% to October 2002, and its price against SPY fell 66%. The S&P 500 fell 49%, and the rest of the market did not escape: the largest companies weighted equally, which had lagged on the way up, still fell 41%, and the NYSE A-D line did not regain its 1998 high until November 2004. A recession and the September 2001 attacks were part of that.

Today the five largest spenders are 20% of the market value of the 325 large companies we track, and the five largest companies by value, Nvidia, Apple, Alphabet, Microsoft and Amazon, are 33%. As a group the spenders still spend less than their operating cash flow, which leaves room to cut back. My guess is that an unwind would start as a slowdown in spending and land hardest on the suppliers whose profits depend on it.

Where I come out

A trader in 1999 was right to take the NYSE divergence seriously, but would have been early by two years. Today’s Nasdaq divergence is longer and larger in total, which deserves the same attention. What separates it from 2000 is where the selling sits: in the less-traded part of the Nasdaq, not among the companies carrying the index, whose run I think rests on the AI build-out. If that spending stalls or the weakness reaches the most-traded stocks, the comparison with 2000 gets much closer.

What would change my read

Toward 2000: the Nasdaq-100 members’ two-year net breadth turning negative (it read +3.17% on October 16, 2026), or Nasdaq-listed stocks’ reading falling below its January 2024 low of −3.39%. Both readings are in the box at the top of this page and update daily.

Toward an unwind of the AI trade: two or more of the five largest spenders reporting lower capital spending over four quarters than a year earlier, which Corporate Capex shows company by company, or XLK against SPY closing a week below its March 2000 peak of 150 on the scale above, from 169 at publication.

Toward a benign ending: the Nasdaq-100 members’ A-D line closing above its August 13, 2026 high, which the Advance-Decline Line page charts daily.

How we checked it

• The A-D line is a running total of advancing minus declining issues. Its level depends on how many issues trade, so lines are compared by when they peaked and by net breadth, not by level.
• Net breadth is advancers minus decliners as a share of issues traded that day. The main comparison averages it over each divergence, from the A-D peak to the record; the window table repeats it over fixed one-, two- and three-year windows ranked against each exchange’s record days.
• Two universes. NYSE and Nasdaq exchange totals since 1965 and 1998 include funds and preferred shares. Our own counts since November 2016 cover common stocks only, keep companies that later delisted, and track the exchange totals with correlations of 0.973 (NYSE) and 0.986 (Nasdaq). Both Nasdaq versions are shown.
• Trading-volume groups rank Nasdaq-listed common stocks each day by average dollar volume over the prior 63 sessions, using our dated listing records.
• Tops and false alarms. A bear market is an S&P 500 record close followed by a 20% fall before the next record, from daily closes since 1928. The false-alarm test uses every S&P 500 record close since 1967 and the lowest close over the next 504 sessions; a new episode starts after a year without a match. Its thresholds were chosen after looking at 2000, so treat it as description.
• The AI build-out. Capital spending, operating cash flow and operating profit come from the quarterly and annual filings of the large US companies whose financial statements we extract; the five largest spenders are ranked within each quarter. Equal and value weighting use Ken French’s CRSP size deciles 8 to 10, about the 500 largest US stocks, through August 2026.
• Limits. Daily stock counts before 1965 are not in our data, so 1929 is not tested. The Nasdaq’s March 2000 reading averages the 347 sessions available since November 1998. Nasdaq-100 members are today’s members.
• Frozen evidence. The charts and tables use the snapshot of October 16, 2026, taken at the October 6, 2026 record where noted. Only the latest-reading box updates.