Opinion: Wall Street Finally Started Pricing the AI Boom on Its Merits

Over nine days this autumn, the four companies that have bet the most on artificial intelligence opened their books to investors: Meta Platforms on October 21, Alphabet and Microsoft on October 28, and Amazon on October 29. Taken together with the government's first look at third-quarter economic growth, released the same week by the Bureau of Economic Analysis, the results amounted to the clearest evidence yet that Wall Street has stopped rewarding AI spending simply because it is AI spending. It has started asking what that spending buys, and it answered two of the four companies with a sell-off.

Alphabet reported revenue of $127.6 billion, up 25 percent, with Google Cloud revenue more than doubling to $30.7 billion, a 103 percent jump. Yet the stock still fell roughly 8 percent in after-hours trading, because capital expenditures hit $51.4 billion for the quarter, more than double the prior year according to Alphabet's own quarterly filing, and an $8.7 billion loss on equity holdings dragged diluted earnings per share down to $2.12 from $2.87. Meta's numbers were worse in the places that mattered most to investors: revenue rose 26 percent to $64.6 billion, but capital spending of $34.9 billion outran the $32.8 billion in cash its operations generated, producing negative free cash flow, and earnings per share of $3.83 badly missed the $6.67 Wall Street expected. Meta's shares tumbled about 11 percent after hours. Amazon and Microsoft, by contrast, were rewarded. Amazon's cloud division, AWS, grew 39 percent, its fastest pace in nearly five years, on its way to a record 13.9 percent company-wide operating margin, and investors shrugged off the fact that trailing twelve-month free cash flow swung to a $42.4 billion outflow from a $14.8 billion inflow a year earlier. Microsoft's Azure revenue, disclosed in dollar terms for the first time, grew roughly 48 percent to around $33 billion, and its shares rose even as full-year capital spending guidance climbed toward $175 billion.

A Trillion Dollars and Counting

Add up what the four companies have told investors to expect for 2026 on their own terms: roughly $220 billion at Amazon, which said in its third-quarter filing that it "continues to expect 2026 cash capital expenditures to be approximately $220 billion"; $195 billion to $205 billion at Alphabet; $135 billion to $145 billion at Meta; and on the order of $170 billion in Microsoft's current fiscal year. The total lands somewhere near $750 billion, in the same range outside estimates have placed it. The Bank for International Settlements, in its latest annual economic report, put the scale in starker terms: the five largest hyperscalers are on pace to spend more than $1 trillion on AI-related capital projects across 2025 and 2026 combined, an amount the BIS said is increasingly financed not from operating cash but through debt and private credit markets that are harder for regulators, and investors, to see into.

This spending is no longer a Silicon Valley curiosity; it shows up in the national accounts. The Bureau of Economic Analysis's advance estimate for the third quarter found that real GDP grew at a 3.4 percent annual rate, and that the rise in investment "primarily reflected increases in private inventory investment and in equipment," with the increase in equipment led by information-processing gear, mainly computers and peripheral equipment, the components that fill AI data centers. Independent economists at firms including ING and Bridgewater have argued that AI-related investment now accounts for somewhere between a quarter and a third of the economy's overall growth this year. Even taken conservatively, the AI buildout has become one of the load-bearing walls of the current expansion, which means its condition is no longer a question for four companies' shareholders alone.

What the Market Is Starting to Price

The useful thing about the past two weeks is not the dollar figures, which anyone following the sector already expected, but the pattern in how the market reacted to them. Alphabet's cloud business had its best growth in years and the stock still fell, because the earnings miss and the cash burn said something about discipline that the revenue number did not. Sundar Pichai tried to frame the quarter around demand rather than cost:

"Google Cloud revenues grew 103 percent, powered by huge demand for our AI infrastructure and AI solutions."

Investors heard him, and bid the stock up on cloud growth in isolation plenty of times before. This time they also did the arithmetic on free cash flow and the one-time equity loss, and sold anyway. Amazon's Andy Jassy made a similar case for patience, telling investors that AWS revenue growth should exceed 40 percent in both 2027 and 2028 and that the company plans to "spend meaningfully more" next year; this time the market credited the growth story because the current quarter's margin and backlog numbers were already working. Meta offered growth without the offsetting proof and was punished hardest of the four.

That is, in miniature, exactly the kind of discipline critics of the AI buildout have been asking markets to apply for two years: stop paying for capital expenditure as a story and start paying for capital expenditure as a result. It is a healthy sign. It is also an unreliable one to depend on by itself, for a reason Microsoft's own numbers illustrate. Secondary analysts could not even agree on Microsoft's quarterly capital spending figure, with some reporting $39.4 billion and others $45.2 billion, a gap traceable to a change in how the company now amortizes data-center leases, stretching their useful life from fifteen years to twenty-five. That single accounting choice lowers near-term reported depreciation and smooths the capex figure that investors use to judge discipline, through no misstatement of fact but through a presentation choice that is Microsoft's alone to make. If professional analysts cannot reliably compare one hyperscaler's reported spending to another's, the market's capacity to discriminate between good and bad capital allocation is weaker than last week's stock moves suggest.

The Case Against Complacency

There is a reasonable objection to all of this: markets just demonstrated they can discipline themselves, so what exactly is there for anyone else to do? Boards and executives, the argument runs, are better placed than any regulator to judge whether a given data center will pay for itself, and the sell-offs in Meta and Alphabet stock are proof the system is working as intended.

That argument would be more persuasive if the spending were still funded the way it was two years ago, largely out of the hyperscalers' own enormous operating cash flow. It increasingly is not. The BIS's report is explicit that AI investment is shifting toward debt and non-bank credit vehicles whose exposures are opaque even to the lenders involved, a concern echoed in surveys of institutional fund managers who have repeatedly named hyperscaler capital spending as their leading candidate for the next systemic credit event. A bet financed with a company's own cash that goes wrong is a write-down. A bet financed through layered private-credit structures that goes wrong can propagate through balance sheets that have nothing to do with artificial intelligence, in the way mortgage losses once did. The scale involved, a trillion dollars from five companies in two years, is now large enough that an abrupt unwind would not stay contained to technology stocks; it would hit the index funds that make up the bulk of ordinary retirement accounts and a GDP figure that, by the BEA's own breakdown, is already leaning on this one category of investment.

There is also a more uncomfortable asymmetry sitting underneath the earnings reports: the same companies pouring unprecedented sums into AI infrastructure have been cutting their own workforces. Amazon alone has eliminated roughly 30,000 corporate positions since October 2025, including a round of 16,000 cuts announced in January, even as its quarterly capital spending climbed to $77.3 billion. None of that is necessarily contradictory; companies can simultaneously automate work and expand infrastructure for entirely separate reasons. But shareholders and the public deserve a clearer account of the connection than "AI shifts workforce priorities," particularly when the same balance sheet shows free cash flow swinging $57 billion in the wrong direction over a single year.

What Should Happen Next

None of this is an argument for throttling AI investment, which may well prove as transformative as its proponents claim; the BIS itself credits AI optimism with helping global growth absorb this year's trade shocks. It is an argument for making sure the market's newly visible appetite for discipline has the information it needs to do its job consistently, not quarter by quarter and company by company.

• Securities regulators should press for standardized, comparable disclosure of AI infrastructure spending, including how much is funded from operating cash versus debt or off-balance-sheet vehicles, so that one company's lease-accounting choice cannot flatter its numbers against a peer's.
• Banking and financial-stability regulators should take up the BIS's recommendation to map exposure to AI-related debt now, before a shock forces them to do it after the fact, given how much of that debt runs through private credit funds outside routine bank supervision.
• Boards tying executive pay to AI capital spending or to cloud growth rates should tie it as well to realized returns, backlog conversion, and free cash flow, not simply to the size of the investment.
• Companies cutting headcount while expanding AI budgets owe shareholders and workers a specific explanation of how the two decisions relate, rather than parallel press releases that leave the public to guess.

The market did something useful last week: it stopped treating every AI capital expenditure announcement as automatically bullish. That is the floor for sound capital allocation, not the ceiling. Regulators, boards and the companies themselves should build on it before the next earnings season arrives and, perhaps, before the debt behind this buildout comes due.