The biggest technology companies are committing enormous sums to AI infrastructure, but the business case remains unsettled. Amazon, Microsoft, Alphabet, and Meta plan at least $720 billion in capital expenditures this year, with most directed toward AI, alongside at least $1 trillion in commitments for coming years.
Spending plans keep growing
The latest earnings calls offered little sign that the companies are ready to slow their data center expansion. Alphabet expects capital expenditures of between $195 billion and $205 billion this year and plans to increase spending “significantly” in 2027. Amazon expects to spend $220 billion this year, then “meaningfully more” in 2027.
Amazon CEO Andy Jassy described the opportunity as “once in a lifetime,” arguing that the pace of change makes it impractical to spread investment over a longer period. Meta expects to spend between $135 billion and $145 billion this year. The company did not specify next year’s planned spending, though major financial institutions expect it to rise.
Microsoft reported $45.2 billion in capital expenditures for the quarter, below the expected $50 billion. The company attributed the lower figure to “third party capacity delays,” rather than a decision to cut back. Together, these plans suggest that the buildout depends on companies’ expectations of future demand, even as the costs arrive now.
Revenue is rising, but so are questions
There are signs of business growth: Amazon reported its fastest growth in five years for the quarter, Meta’s ad revenue is surging, and Google Cloud’s revenue more than doubled in a year. Still, investors are asking whether those gains can justify the scale of spending and when the investments might produce returns.
That question is especially visible at Alphabet. Philipp Schindler, the company’s senior vice president and chief business officer, said growth in Google Cloud had come at the expense of its core Search business. He also said that more queries in AI Mode affected revenue because those queries had lower monetization.
Google Search ads revenue grew 12% in the third quarter, compared with 17% growth in the same quarter last year. Alphabet says its ad experiments for AI Mode have shown promising results, but Schindler did not give specific examples. The company is therefore managing a tension: AI features may attract use while bringing in less advertising revenue than the search business they affect.
Alphabet’s stock fell 13% in one day, its worst one-day drop in history, as investors reacted to the company’s plans and the uncertainty around returns. The broader spending announcements also triggered a sharp selloff in tech stocks, which began to taper off as investors shifted focus to Tuesday’s midterm elections.
Infrastructure costs reach beyond the big platforms
AI developers face their own challenge in balancing growth against the cost of computing. Anthropic’s revenue grew to $17 billion in the quarter, more than 1,000% above the prior year, while its compute costs reached $17.5 billion. The figures show how fast rising sales can coexist with costs that outpace them.
Anthropic has also committed nearly $900 billion in future compute and infrastructure costs and plans to build its own data centers. Some of that spending will go to Google, which sells its TPU chips to AI labs. Analysts asked why Google would sell capacity to other companies if it is itself capacity-constrained.
Google is paying to build the hardware before it is ready for use. Contracts with AI labs are backed by financial obligations that come due when the hardware is available, but they do not put cash in Google’s hands immediately. Selling infrastructure can create future business, while leaving the company to fund construction upfront.
Who might change the trajectory?
For now, the companies’ own earnings calls have not pointed to a pullback. Executives continue to describe demand and opportunity as reasons to keep investing, even though the timing and scale of returns are uncertain. Meta CEO Mark Zuckerberg said the Muse agentic AI program would become profitable only if Meta could reduce compute costs. He expects compute efficiency to increase “tenfold” next year, an effort he says will require “significant investment.”
Outside pressure could affect the plans. Candidates elected in the midterm elections favor moratoriums and other restrictions on data centers, creating uncertainty about regulation. That uncertainty may matter to investors, who are already weighing spending commitments against the prospect of future revenue.
The buildout is not simply a question of whether AI use is growing. Companies must also make infrastructure affordable, find ways to earn from AI services, and sustain investment while waiting for returns. Until those pieces align—or outside constraints alter the plans—the earnings calls suggest that the largest players expect to keep building.