After their latest earnings calls, Amazon, Microsoft, Alphabet and Meta have told investors they plan to spend at least $720 billion on capital projects this year, most of it on AI infrastructure, on top of at least $1 trillion in commitments already arranged for the years ahead. The money is real; the payoff, as an analysis in The Verge points out, is still mostly a promise.
The individual numbers are striking. Amazon plans about $220 billion this year, the most in its history, and says it will spend meaningfully more in 2027; chief executive Andy Jassy called it a once-in-a-lifetime opportunity. Alphabet expects between $195 billion and $205 billion and plans a significant increase next year. Meta is expected to spend between $135 billion and $145 billion. The scale spooked investors: Alphabet suffered its worst one-day fall on record, 13%, before recovering somewhat.
Revenue is growing fast, which is the bull case. Amazon reported its quickest growth in five years, Meta's ad revenue is surging and Google Cloud's revenue more than doubled in a year. But Google also showed the strain. Its chief business officer said heavy use of AI Mode in search had weighed on results because those answers earn less, and search ad revenue grew 12% in the quarter against 17% a year earlier. Meta's Mark Zuckerberg said its Muse agent will only become profitable if compute costs come down, and that he expects compute efficiency to rise tenfold next year.
The labs are spending too. Anthropic's updated prospectus showed quarterly revenue of $17 billion, more than ten times a year earlier, against compute costs of $17.5 billion, with nearly $900 billion committed to future compute and infrastructure, some of it bought from Google as TPU capacity. For anyone using the tools we review, this spending is what keeps new models arriving and prices falling; it is also why subscriptions and usage limits keep shifting as companies look for margin.
What could slow it is outside the companies' control. The Verge's argument is that Big Tech will not cut back on its own, and the most likely brake is political: the midterm elections put into office candidates who favour moratoriums and other limits on data centres, a shift we cover separately. The forecasts here are company guidance, not commitments, and the gap between spending and AI profits is an open question rather than a settled verdict.