Is the AI Capital-Spending Boom Peaking, or Just Getting Started?

The question hanging over the market this week was not about any single company. It was about the single largest spending program in corporate history: the hundreds of billions of dollars the biggest technology companies are pouring into AI data centers, and whether that build is running ahead of the revenue it is meant to produce. A multi-day slide in chip and AI-hardware names pointed at the debate. So did a sharp rotation into defensive sectors, and a fund-manager survey that put an “AI bubble” at the top of the worry list.

The spenders keep raising the number

Combined 2026 capital-expenditure plans for the four largest hyperscalers, Alphabet (GOOGL), Amazon (AMZN), Microsoft (MSFT) and Meta (META), now run near $725 billion — roughly 77% above 2025’s already-record total, with the majority earmarked for AI infrastructure. These were not static budgets. Meta lifted its full-year range mid-cycle, and Alphabet’s guidance of $175–185 billion is close to double what it spent the year before. Companies do not typically enlarge the biggest checks they have ever written into softening demand.

The demand signal shows up in the cloud businesses that house the workloads. In the most recent reported quarter, Microsoft’s Azure and other cloud services grew about 29% year over year, Amazon Web Services 28%, and Google Cloud more than 50%. Growth is accelerating. Executives across the group described the same bind: Google’s management said plainly that it was “compute-constrained in the near term,” and Microsoft’s said customer demand continued to outstrip available capacity. On that reading, the spending is a race to catch up with orders already in hand, and the limit is how fast power and buildings can be added.

The $600 billion question

The skeptical case starts with arithmetic that has circulated for more than a year and grown louder as the spending accelerated. Sequoia Capital’s David Cahn calls it AI’s “$600 billion question”: his estimate of the new annual revenue the AI ecosystem would need to generate simply to justify the infrastructure being built around it, a gap he argued was widening. Goldman Sachs’ Jim Covello has made a parallel argument, that the technology remains extraordinarily expensive relative to the problems it currently solves. Neither claim depends on AI failing. It only takes the returns arriving more slowly than the capital.

Accounting adds a structural wrinkle. Across the four largest hyperscalers, property-and-equipment purchases over the four quarters through the first quarter of 2026 ran near $434 billion, against reported depreciation of about $149 billion. Because that spending is expensed over multi-year server and multi-decade building schedules, today’s income statements carry only a fraction of today’s build; the rest of the charge arrives in later years regardless of what AI revenue does by then. Sentiment has shifted too. Bank of America’s July global fund-manager survey found about 45% of respondents naming an “AI bubble” the biggest tail risk to markets, up from 28% a month earlier and, for the first time, the single largest concern on the list. The same survey flagged semiconductors as the most crowded trade on record. And the month has already delivered a live test of how the market treats a stumble: Alphabet shed roughly $200 billion in market value in a single session on a report that its flagship Gemini 3.5 Pro model had slipped behind schedule.

What the data showed this week

The repricing showed up across the sector map. In the session that set the tone, the technology sector ETF (XLK) fell about 2.2%, the weakest of the eleven groups, while the most defensive corners led: consumer staples up 2.8%, health care 2.2% and real estate 2.0%. The AI-hardware complex moved almost as a single bet. Super Micro (SMCI) dropped 8.2%, Oracle (ORCL) 6.3%, Intel (INTC) 5.8%, Micron (MU) 5.7% and Arm Holdings (ARM) 5.4%. Across the full week, the Nasdaq-tracking QQQ fell roughly 4%, the technology sector shed more than 5%, and the Cboe Volatility Index closed at 18.77, up from the mid-teens a week earlier. The spread between the week’s best and worst sectors showed the selling stayed concentrated in the AI-levered names. It was a rotation, not an exit.

What would settle it

Several scheduled facts will move the argument. The next round of hyperscaler earnings will show whether cloud-revenue growth rates hold or roll over, and whether capex guidance is raised again or trimmed for the first time; that is the clearest tell on whether demand is still outrunning supply. Super Micro’s fiscal fourth-quarter report, expected in early August, will offer a direct read on AI-server demand. A firm ship date for Gemini 3.5 Pro would show whether the delays are a stumble or a pattern. Future Bank of America surveys will show whether the bubble worry deepens or fades. Until those numbers land, the question stays open.

This is a general-market summary for information only — not investment advice, and not a recommendation regarding any security.