Research/Research note

Demand isn't AI's problem. Paying for it is.

StoryVector · October 29, 2026

Caterpillar's power and energy sales rose 18% last quarter, led by data centers, and its order book hit a record $78.1bn. Its shares fell about 9% before Thursday's open.

That is the AI trade after two weeks of earnings. Demand is running at full speed, from chips to power equipment. The question investors are asking has changed: who pays for it?

On Monday Meta began selling $35bn of stock to help fund its buildout, five months after Alphabet sold $84.75bn. Borrowing costs sit near their highest in years. And this year chipmaker ETFs are up 64.1%, while the ETFs built around Meta and Alphabet are down 9.8%.

+48%growth in Azure, Microsoft's cloud, last quarter
$120bnof stock sales launched by Alphabet in June and Meta this week to fund AI, with up to $60bn more planned
−9.8%Meta and Alphabet's ETFs this year, against +64.1% for chipmaker ETFs

The builders are booked

Nearly every supplier that reported in the past two weeks showed more demand than forecast. Lam Research, which makes chipmaking equipment, guided above forecasts. Seagate and Western Digital beat and guided higher. Vertiv, GE Vernova and Bloom Energy, which keep data centers powered and cool, all raised their outlooks.

What the AI buildout reported

Results reported from October 19 to the morning of October 29, 2026. The arrow is our read for the AI trade: ↑ supports it, ↔ mixed. Shaded: the three strongest numbers.

ReadCompanyIts part in AIWhat it reportedKey number
↑MicrosoftBuys and sells AI computingBeat forecasts; Azure, its cloud, grew 48%+48% Azure
↑CaterpillarGenerators and turbinesBeat; power sales rose mainly on data centers; backlog a record but growing more slowly$78.1bn backlog
↑Vertiv, GE Vernova, Bloom EnergyPower and coolingAll beat and raised their outlooks; Bloom for a third time3 of 3 raised
↑Super MicroAI serversRaised its revenue range for the September quarter$18.5bn to $19.5bn
↑Seagate, Western DigitalStorageBoth beat and guided the next quarter above forecasts$5.1bn revenue guide
↑Lam ResearchChipmaking equipmentBeat and guided above forecasts; shares slipped lateGuided above
↔MetaBuys AI computingBeat on revenue and raised the floor of its 2026 spending range; then launched a stock sale$35bn stock sale
↔ServiceNowBusiness software, selling AIRevenue beat; AI contract target raised; slower bookings forecast; shares down about 19% before the open$1.7bn AI target
↔AlphabetBuys and sells AI computingCloud revenue doubled; free cash flow negative as capital spending reached $51.4bn; profit cut by an $8.7bn investment loss−$9.1bn cash flow

Then comes the bill

The builders are being paid. The buyers, Meta, Microsoft, Alphabet and the AI labs, have to find the money: from profits, by borrowing or by selling stock.

The buyers have been borrowing all year. By one count the AI industry has sold about $300bn of high-grade bonds in 2026, roughly triple last year, including $25bn from Alphabet in August and Microsoft's first US bond sale in more than nine years. Bonds were not enough. In June Alphabet sold $84.75bn of stock, a record, $10bn of it to Berkshire Hathaway. Now Meta is selling $35bn, with up to $60bn more planned from early 2027, against 2026 spending it expects at $135bn to $145bn.

Alphabet showed why last night. Its cloud revenue doubled to $30.7bn, yet it spent $51.4bn on capital projects and its free cash flow was negative $9.1bn. Its shares fell about 8% after hours.

Lenders saw it first. Earlier this month buyers of CoreWeave's bonds wanted 14.8%, about twice the rate it was charged in March, as our CoreWeave note showed. Stock investors flinched too: on October 23 chip stocks led the Nasdaq to its worst day since June on worries about AI financing.

The revenue is coming

The spending is starting to earn. Azure, Microsoft's cloud, grew 48%, and Google Cloud doubled. Anthropic's listing filing showed $16.3bn of revenue in six months. ServiceNow raised its year-end target for the yearly value of its AI contracts to $1.7bn or more.

Beat, raise, sell

Good results are no longer enough. ServiceNow beat on revenue and was down about 19% before Thursday's open, after forecasting slower bookings. Caterpillar beat, but its backlog grew more slowly than earlier in the year. Lam slipped after guiding above forecasts. Investors have priced in a faster buildout, and any hint of a slower one gets sold.

Where it shows in ETFs

The gap shows inside our narratives. Memory-chip ETFs are up 94.9% this year and chipmaker ETFs 64.1%. The communication services ETFs are down 9.8%; Meta and Alphabet make up 35% to 43% of the three largest, XLC, VOX and FCOM. In October the builders gave some back: memory fell 10.4% and chips 7.5%. Application software rose 5.4% and cloud infrastructure 5.7%, before ServiceNow reported.

Builders, buyers and earners

Our categories inside each narrative: total return, ETFs weighted by assets, to the October 28 close, before ServiceNow and Caterpillar reported. Under each, its three largest ETFs.

Show
Memory chipsDRAM, DISK, KMEM · 7 ETFs+94.9%ChipmakersSMH, SOXX, SOXQ · 17 ETFs+64.1%Data centersDTCR, SRVR, SNSR · 9 ETFs+23.8%Grid and power equipmentGRID, VOLT, AIPO · 10 ETFs+12.5%Application softwareCIBR, IGV, HACK · 15 ETFs+36.1%Cloud infrastructureSKYY, CLOU, WCLD · 5 ETFs+33.5%Meta and Alphabet's ETFsXLC, VOX, FCOM · 8 ETFs−9.8%S&P 500SPY+14.1%
buildersearnersbuyersthe market

By our measure, the AI compute core, the chips, memory and big buyers, is up 38.6% this year but down 4.1% in October. The productivity handoff, the software companies that have to earn the money back, rose 3.7% in October. Some of the AI trade sits in funds that never mention it: Caterpillar and GE Vernova are two of the ten largest holdings in XLI, State Street's $30bn industrials ETF.

The AI trade by narrative, this year

Total return, ETFs in each narrative weighted by assets, to the October 28 close.

-20%-10%0%+10%+20%+30%+40%+50%JanFebMarAprMayJunJulAugSepOctThe AI compute core +38.6%The productivity handoff +30.6%SPY +14.1%The power constraint +1.6%

The bottom line

The builders are booked and the revenue is coming. What changed this month is who pays, and at what price. That, more than demand, is what the market is now pricing. If Microsoft or Amazon follows Alphabet and Meta in selling stock, it will say more than any earnings beat.

Notes and method. ETF returns are total returns at the close on October 28, 2026, before ServiceNow and Caterpillar reported. Narrative returns weight each ETF by its assets, with leveraged and inverse ETFs left out. Categories are ours, inside each narrative; a category's return weights its ETFs by assets. Holdings are each ETF's ten largest, at our latest fetch on October 1, 2026. Share moves after hours and before the open are as reported and may change once trading opens. Sources: company results October 19 to 29, via the StoryVector wire; Caterpillar, third-quarter results, and Reuters, October 29; ServiceNow earnings call, October 28; Meta offering, October 26; Alphabet third-quarter results, Fortune, October 28; Alphabet stock sale, June 3; AI bond total, The Daily Upside, October 26; ServiceNow shares, 24/7 Wall St., October 29; Alphabet bond sale, Bloomberg Law, August; Microsoft bond sale, Bloomberg via Reuters, July; Microsoft results, October 28; Anthropic, Form S-1, October 26. All dates 2026.