The Bull Case for the Nasdaq-100: Why the AI Boom Hasn't Cracked Yet

By Yuriy Matso · The Trading Tools · August 28, 2026 · 13 min read

Research note: SEC filings for current Nasdaq-100 members and QQQ daily closes since 1999. Fundamental study frozen at August 27, 2026. How we checked it · How we use AI

Three-year aggregate revenue CAGR

11.8%

Companies growing revenue

73 of 82

Operating margin, strict cohort

15.4% → 20.3%

Free cash flow

$455B → $623B

Five-year R&D growth

2.14× · 16.8% a year

Three-year valuation contribution

0.90×

What would a real crack look like?

High spending by itself is not enough. The serious version of the AI-bubble argument is a chain: companies build capacity faster than customers need it, cash generation fails to keep up, profits stop following revenue, and share prices keep rising only because investors pay a higher multiple. That sequence is testable. I looked for the break at each link.

The spending is enormous. The cash is still carrying it.

This is the cleanest stress test. I restricted the comparison to companies with revenue, operating income, net income, cash from operations and capital spending reported over the same four quarters at both ends. That leaves 71 issuers. Their annual capital spending climbed from roughly $307B to $703B in less than three years.

Cash from operations rose at the same time, from $762B to $1.33T. After the spending, free cash flow was not merely positive; it increased by 36.8%. That is hard to square with an aggregate funding crisis.

The uncomfortable part belongs in the same chart. Capex grew much faster than free cash flow, and free-cash-flow margin barely changed. The companies can afford the buildout today. The return on the next dollar remains the open question.

Demand is broader than the AI headlines

Nvidia supplied the largest dollar increase in revenue. It did not supply the breadth. Across the 82 current members with aligned sales at both endpoints, 73 grew, 40 compounded above 10% a year, and the median company grew 9.3%. The five largest contributors still produced 59.5% of the aggregate increase. Broad does not mean evenly distributed.

R&D gives the top line a plausible engine. Among the 39 companies that disclose it consistently, spending increased 2.14× over five years and rose from 12.1% to 13.6% of revenue. The companies in the middle and highest 2023 R&D-intensity groups subsequently delivered median revenue growth near 16% a year, against 9.3% in the lowest group.

R&D intensity is measured in the filed year known by September 2023; revenue growth follows through August 2026. The rank correlation is 0.41. Sector mix and current-member survivorship remain, so this is time-ordered association rather than proof that each R&D dollar caused sales.

Sales are becoming operating earnings

Revenue is useful only if it survives the income statement. Here the cross-section is unusually clean. Sort companies by their three-year revenue growth and the operating-income result climbs with every group. The slowest third produced almost no operating-profit growth (0.8% a year on 2.5% revenue growth). The fastest third turned 16.5% median revenue growth into 28.3% operating-income growth.

This is the strongest link in the chain. Across all 59 usable companies, the rank correlation between revenue and operating-income growth was 0.65. It does not require an AI label. It requires sales growth and operating leverage.

Earnings growth is being rewarded

Prices anticipate earnings; they do not wait politely for a filing. Even with that timing mismatch, the fundamental ranking shows through. Companies in the lowest net-income-growth third produced a median 10.8% annual increase in market value. The middle third reached 15.4%. In the highest third, median net income compounded at 29.2% and median market value at 29.2%.

A 0.43 rank correlation is meaningful, not destiny. Expectations, index additions and starting valuation still matter. The point is narrower: the advance did not float free of company results.

Valuation did not power the latest double

QQQ itself rose 2.01× from the September 2023 endpoint through August 27, 2026. The matched current-member basket is different: its aggregate market value rose 2.25×. It is not the index return, because it carries today’s companies backward and does not reproduce Nasdaq’s historical weights or divisor.

That basket is still useful for the accounting bridge. Revenue rose 1.38×, the net margin contribution was 1.80×, and the valuation multiple contributed 0.90×. Multiply the three and they return the basket’s 2.25× change exactly. The re-rating was negative.

What another double would require

The margin cannot do the same work twice without approaching an implausible level. Hold both margin and the multiple flat and revenue must double. At the basket’s recent three-year revenue rate that takes about 6.2 years; at its five-year rate, about 7.0.

History says the horizon matters

There are two honest ways to define a double. QQQ finished at least twice its starting price in 46.5% of rolling five-year windows. It touched twice the starting price at some point inside 60.0% of them. Those answer different investor questions, so I would not collapse them into one probability.

The endpoint rate rises to 60.2% at seven years and 82.0% at ten. The daily windows overlap heavily; they are descriptions of QQQ’s record, not thousands of independent trials. And the dot-com years matter. A buyer at the start of 2000 waited 19.3 years for the first double; the 2001 and 2002 starts waited 13.9 and 11.7 years. “Long enough” has occasionally meant most of an investing career.

The strongest argument against this is on the same cash-flow statement

Capital spending does not stay capital spending. It becomes depreciation, on a lag set by the useful life of what was bought, and depreciation is an operating expense. So the buildout that looks affordable today is also a charge being deferred into future margins, and the size of the deferral is measurable.

Across the 48 members carrying capital spending, depreciation, revenue and operating income on the same reported year at both ends, capex rose from $203B to $413B, a factor of 2.04×. Depreciation rose only 1.36×, from $156B to $212B. Capital spending now runs 1.95× depreciation, against 1.30× three years ago.

Read it as the size of the bet rather than a prediction. 34.2% of operating income is what the spending has to earn back before it improves anything, and it is why the honest version of the bull case depends on revenue from these assets rather than on the assets themselves. That is the same margin ceiling from earlier in the article, arriving from the other direction.

Where the first cracks should appear

The bull case is intact at publication, but it now has a short list of conditions. I would change the read if any two begin moving together:

• Cash conversion: aggregate free-cash-flow margin falls below 10% while capex keeps rising. That would turn an affordable buildout into a funding squeeze.
• Demand breadth: fewer than 60% of matched members grow revenue, or aggregate revenue growth falls toward 7%—the rate that stretches a fundamentals-only double to roughly ten years.
• Operating leverage: operating income stops growing faster than revenue and the matched operating margin turns down across a majority of issuers.
• Valuation dependence: market value continues rising while earnings stall and the multiple becomes the dominant contributor again.

None had fired in the frozen study. That is the bullish conclusion. It is narrower—and more useful—than saying AI will change everything or that a high P/E must end badly.

What this method cannot see

Every tripwire above is drawn from a filing, and that is the honest limit of the whole exercise. Filings arrive weeks after a quarter ends and describe a period already over, so a break in demand shows up here one or two reporting cycles after it starts. A cross-section of income statements is the wrong instrument for anything fast.

It is also the wrong instrument for anything that never reaches an income statement first. A policy change, an export restriction, a supply shock, a customer concentrated enough that one cancelled order resets a backlog: each would reprice these companies long before it changed a cash-flow statement, and nothing in this analysis would register it until afterwards. I am not going to hand you a list of geopolitical predictions dressed as research, because I have no data on any of them and neither does this page.

So read the conclusion for what it is. The accounts do not show a crack. That is a statement about the accounts, in the window they cover, on the companies still in the index today. It is not a claim that nothing can go wrong, and the things most likely to go wrong first are the ones this method would learn about last.

How we checked it

Frequently asked