Nasdaq Falls as Lenders Tighten on AI Spending

The Nasdaq Composite closed down 694 points, or 2.6%, at 26,008.86 on October 22. Meta Platforms had just posted negative free cash flow, and Nvidia chief Jensen Huang said AI builders are short of working capital. The S&P 500 lost 1.2%, while the Dow Jones Industrial Average rose 11.89 points.

The session was a verdict from the people who write the checks for data centers, not a referendum on whether the software works.

A 694-Point Drop on a Rising Dow

The Nasdaq was down as much as 751 points, or 2.8%, before that close, its weakest day since June. The S&P 500’s 93.74-point decline, to 7,663.43, was its worst since July. Most U.S. stocks still finished higher. The Russell 2000 added 0.1% to 2,784.03, and the Dow eked out 0.02% at 52,867.98.

Chipmakers absorbed the hit. The Philadelphia Semiconductor Index fell 6.7%. Advanced Micro Devices dropped 8.1%, Nvidia 4.4%. CoreWeave, the debt-heavy AI cloud landlord, fell 10%. Meta lost 7%. Tesla slipped 3.1%.

THURSDAY’S CLOSES

Treasury prices jumped as money left tech for the safety of government paper, and yields fell hard. That is the same bid that prices corporate debt. When it shows up on a day Meta’s ad engine is still humming, the message is about funding, not about users.

THE THREE DAYS THAT REPRICED THE CHECKBOOK

1. October 21: Meta and Tesla report after the bell. Meta’s ads beat. Its cash does not. Tesla prints positive free cash flow by spending less than Wall Street modeled.
2. October 22: Huang, in Berlin, says neocloud demand is “off the charts” and working capital is not. The Nasdaq closes down 2.6%. Brent settles at $113.17.
3. October 23: U.S. index futures point lower as the same argument follows the sun into Asia.

A 7% drop in Meta on record advertising sales is a financing mark, and it reprices every supplier whose order book is tied to hyperscaler campuses.

How Meta’s Quarter Turned Into a 7% Selloff

Meta’s own release is the cleanest exhibit. Revenue was $64.60 billion, up 26%. Family daily active people averaged 3.64 billion in September, up 3%. Ad impressions rose 10% and the average price per ad rose 14%. Mark Zuckerberg, the founder and chief executive, said the company hit “major milestones this quarter toward delivering personal superintelligence for everyone,” pointing to Muse and Muse Wonder, and added, “Our investments are paying off, the core business is strong, and we’re seeing velocity across the board.”

Muse reached 10 million people in the United States and Canada in six weeks, with about 25% of those who set it up using it on a given day. The ad machine is not the problem.

Cash is. Operating cash flow was $32.79 billion. Capital spending, including principal payments on finance leases, was $34.91 billion. In the same release, Meta said free cash flow was negative $2.12 billion. Cash, cash equivalents and marketable securities still stood at $82.45 billion on September 30. Headcount was 66,775, down 15% from a year earlier.

META’S THIRD-QUARTER CASH MATH

• Revenue: $64.60 billion, up 26% from $51.24 billion a year earlier.
• The legal charge: $9.89 billion of an attorneys-general deal of up to $18.59 billion, which left GAAP operating income at $12.99 billion, down 37%.
• Earnings: $3.83 a share, or $7.40 excluding that charge.
• The spend plan: 2026 capital spending narrowed to $135-145 billion, from $130-145 billion, with fourth-quarter revenue guided at $73-76 billion.

Costs and expenses rose 68% to $51.61 billion. Full-year expenses are now seen at $175-179 billion. Strip out the legal charge and Meta still says 2026 operating income should top 2025. Investors sold the stock anyway, because the company also told them it needs a new class of shares to keep funding the race.

The board approved a non-voting Class C common stock, effective January 19, 2027, plus a one-time stock dividend that Meta expects the market to treat like a two-for-one split. The special committee’s own language is blunt. The new class exists “for capital raising, employee compensation, potential acquisitions, and other general corporate purposes,” because “the scale of investment required to compete” is large. Zuckerberg keeps control. Outside shareholders get a security the company can print without handing them votes.

Who Is Financing the $10.3 Trillion Buildout?

Columbia Business School professor Stijn Van Nieuwerburgh, in a paper presented at the Brookings Papers on Economic Activity, puts U.S. spending on data-center buildings, power, networks and specialized chips at $10.3 trillion from 2025 to 2032. That is an average of 3.63% of U.S. GDP a year, and he writes that the projected buildout “would be larger relative to the economy than the major U.S. canal, railroad, electrification, highway, and telecommunications investment booms.”

His central case adds about 183 gigawatts of computing capacity by 2032. To earn a 10% unlevered return on that stock of plants, the same paper’s math implies roughly $3.7 trillion of annual AI-related revenue by 2032, which would require growth near 80% a year from a combined OpenAI and Anthropic sales base around $100 billion. Carl Weinberg, chief economist at High Frequency Economics, has called the wave the largest tide of investment in U.S. history.

S&P Global has a narrower cut of the same boom. It expects the top six U.S. hyperscalers, Amazon, Microsoft, Alphabet, Oracle, Meta and SpaceX, to spend more than $7 trillion on data centers and related outlays between 2025 and 2030, and to run negative free cash flow until at least next year. That $7 trillion is a company list over six years. The $10.3 trillion is the whole U.S. build, including power and chips, through 2032.

The cash machine that used to pay for this is no longer enough. Capital Group’s credit desk counted the five big U.S. hyperscalers, Alphabet, Amazon, Meta, Microsoft and Oracle, as having issued $240.7 billion in debt year to date through August 31, $66 billion of it in non-dollar currencies. Add dollar bonds from SpaceX and Nvidia and the pile across currencies is $290.7 billion, against $321 billion of 10-year U.S. Treasuries issued over the same stretch.

TWO DIFFERENT BILLS, TWO DIFFERENT CLOCKS

U.S. investment-grade spreads, on the Bloomberg U.S. Corporate Index, were still 78 basis points at the end of August, unchanged on the year. Spreads on 10-year-plus hyperscaler bonds had already widened. Capital Group expects more than $25 billion of extra paper from those five names in the last four months of 2026, then $100 billion to $200 billion in the dollar investment-grade market in 2027 and 2028.

The rest of the money is harder to see, which is Van Nieuwerburgh’s warning. Risk is moving off the big corporate balance sheets and into joint ventures, private credit, securitizations, special-purpose vehicles, leases and guarantees. Those structures, he writes, then depend on “uncertain AI demand, rapid technological change, timely access to power and hardware, and the continued credit quality of a small number of [data center] tenants.” He says it would be “premature to conclude that AI infrastructure already poses systemic risk comparable to earlier credit booms,” and that off-balance-sheet structures matter “because they may make correlated exposures hard to observe before a downturn.”

WHERE THE MONEY NOW COMES FROM

• Senior unsecured bonds: Still the main tap for Amazon, Alphabet, Microsoft and Meta, and the bid Capital Group says has so far absorbed the supply.
• Project vehicles: Beignet Investor, a Blue Owl Capital holding company, sold the largest single-tranche bond in the U.S. investment-grade market to fund a 2.1-gigawatt Meta campus in Louisiana. The notes are liquid and not index-eligible.
• Leases and guarantees: Long-dated campus leases and landlord backstops that never hit the headline debt line, which is why S&P’s tech analyst, Habib, told a London roundtable it is “safe to assume that there is more (debt) out there that we are not seeing.”
• New equity that does not vote: Meta’s Class C stock, built to raise cash without loosening Zuckerberg’s hold.
• Backup bank lines: Tesla, even as it cut spending, secured capacity to borrow up to $30 billion.

S&P cut Oracle to BBB-, the last rung of investment grade, in July on AI spending and what it called an uncertain path to profit. CoreWeave’s five-year bonds had already fallen through 80 cents on the dollar earlier in October, the distressed line. Credit had started to mark this trade before the Nasdaq tape did.

Jensen Huang Named Capital as the Bottleneck

Huang said the quiet part on the same Thursday the Nasdaq fell, speaking in Berlin on the sidelines of Nvidia’s GTC event. Demand at the so-called neoclouds, the specialist AI landlords that rent clusters to labs and hyperscalers, is not the issue. Money is.

If you look at the fundamentals, the demand for the neo clouds is off the charts. What is missing in the middle is the working capital to fuel that growth, because there’s just so much demand across so many sectors of AI, there’s a scarcity of capital. The demand of the capital is much greater than the availability of the capital. That’s going to get solved.

Jensen Huang, chief executive, Nvidia, speaking in Berlin

He told the same Berlin audience that a 1-gigawatt data center costs about $50 billion. Nvidia is trying to make its hardware an asset lenders will finance, and it has already wrapped large customers in guarantees and capacity-purchase promises. Hock Tan, Broadcom’s chief executive, has made the same point from the other side of the chip counter: capital, not silicon, is the binding constraint.

Adam Crisafulli, head of Vital Knowledge, put the equity version of that bind in a note the same day. AI growth depends on capital spending, he wrote, “and since capex is dramatically outpacing op. cash flow, companies are reliant on the largess of equity and debt investors to underwrite all the spending, but lately those markets are becoming increasingly less accommodating.” Slowing debt issuance as credit tightens, he added, could hinder the buildout itself.

That is the hidden stakeholder. Bond funds, project lenders, lease counterparties and the buyers of Meta’s future Class C shares can slow the $10.3 trillion plan without a single model missing a benchmark. Huang says the shortage will get solved. The October 22 tape was the market asking him to prove it.

Tesla Cut Its Capex Plan and Still Slipped

Tesla’s quarter should have been the control experiment. Revenue was $28.05 billion, flat with a year earlier. GAAP net income was $787 million. Operating income was $718 million. Non-GAAP earnings were $0.48 a share. Operating cash flow was $6.26 billion. Capital spending was $4.73 billion, up 110% from $2.25 billion a year earlier and down from $5.79 billion in the second quarter. That left free cash flow of $1.53 billion, after a $1.09 billion outflow in the second quarter. Cash and short-term investments ended at $45.45 billion.

CFO Vaibhav Taneja told analysts the firm is “revising our CapEx guidance to be approximately $20 billion for the current year as a result of managing our capital deployment efficiently,” and that it plans to fund that “primarily from our cash flows from operations.” The prior guide was more than $25 billion. Tesla still lined up up to $30 billion of borrow capacity “to support the expected growth in CapEx.” The stock fell 3.1% anyway.

Spend less, print cash, get marked down next to Meta, which spent more and burned cash. The common factor is the cost of money around AI, not the sign on one quarter’s free-cash-flow line. Ryan Lee, a senior vice president at Direxion, said the cash beat was pleasant and that, paired with the lower spending, it “raises questions on the next step,” and “suggests less of an ‘all-in’ approach to the AI arms race than one would expect from Elon.”

Oil Near $113 Added a Second Shock

Brent crude, the international grade, settled at $113.17 a barrel, up 4.00%, the highest since May 4. West Texas Intermediate settled at $99.47, up 2.39%. President Donald Trump, asked about Iran late on October 21, said the United States is not negotiating and that “something one way or the other” will “happen fairly soon.”

Dearer crude hits this trade three ways at once. It keeps the Federal Reserve cautious on inflation. It raises the power bill inside every campus Huang prices at about $50 billion a gigawatt. And it lifts the discount rate on debt that already has to fund spending in excess of cash flow. Software that uses AI inside workplaces was not the name being sold. The names that borrow to pour concrete and stack chips were.

The Reports Still Ahead for Amazon and Microsoft

Alphabet, Microsoft and Amazon still have to report. Those three, with Meta, are the cash engines Capital Group and S&P are modeling as they swing from funding the boom with operating cash to funding it with bonds, leases and vehicles. A second down tape after those prints would tell lenders the October 22 bid for Treasuries was not a one-day scare. A shrug would tell them the high-grade window is still open, just more expensive.

WHERE EXPERTS DISAGREE

• The window is still open: Dubravko Lakos-Bujas, JPMorgan’s global market strategy head, wrote on October 22 that the investment-grade market has substantial capacity to take hyperscaler paper, that operating cash flow for the group is seen exceeding $900 billion through 2027, and that heavier outside funding is not expected to weaken operating fundamentals.
• The window is the risk: Crisafulli says equity and debt investors are “increasingly less accommodating,” and Huang says working capital is the missing middle. S&P’s Habib says the rated totals understate the true AI-linked load.
• The middle path: Capital Group’s credit team is modestly overweight selected hyperscaler bonds and data-center builders, underweight chip names that are stretching their own balance sheets, and selective on 144A special-purpose vehicles, while warning that AI outlays still “vastly exceed AI revenues.”

JPMorgan can be right about Amazon and Alphabet and still leave CoreWeave, Oracle and the Louisiana project paper in a tighter box. High-grade borrowers with ad and cloud cash can keep issuing. Everyone else needs Huang’s scarcity to get solved on his timetable, not the bond market’s.

On October 22 the people who price that paper, not the people who train the models, set the Nasdaq’s close. Van Nieuwerburgh’s policy ask is dull on purpose: better measurement while the capital structure is still being written. Until that ledger is in daylight, every 694-point day will look like a software scare and behave like a credit one.

Disclaimer: This article is news reporting and analysis of market moves, company filings and research notes. It is informational only and is not investment advice, a recommendation to buy or sell any stock, bond, fund or commodity, or a forecast of future returns. Readers should consult a licensed financial adviser or broker who understands their own holdings, time horizon and risk tolerance before making any investment decision. Index levels, oil prices, debt totals and company figures reflect the filings and market sources cited as of October 22, 2026, and those values can change in the next session.