Big Tech Just Promised $725 Billion of AI Capex This Year

The four U.S. hyperscalers reporting earnings this week handed investors a number that is hard to put in context: roughly seven hundred and twenty-five billion U.S. dollars in combined 2026 capital expenditure, the bulk of it earmarked for AI infrastructure.

Alphabet led the surge. The parent of Google reported Q1 results late on Wednesday and lifted its full-year capex guidance to $190 billion, up from a previous range of $175 to $185 billion. Google Cloud revenue grew 63 percent year over year to $20 billion, a faster clip than either AWS or Microsoft Azure posted in the same period. Shares jumped roughly six percent in extended trading, with CNBC noting the company's enterprise AI business has now overtaken its YouTube ads segment in absolute revenue.

Microsoft matched Alphabet's $190 billion capex figure and warned investors that data-center capacity is likely to constrain growth through the rest of the year. Chief financial officer Amy Hood told analysts the company is "still capacity-bound on the AI side, and will be for several quarters." Fortune reports that Azure's AI services revenue alone has more than doubled in twelve months.

Meta was the outlier. The Facebook parent raised its 2026 capex range to $125 to $145 billion but disappointed on user growth and ad pricing, sending shares down about seven percent after hours. Investors appear less convinced that Meta's open-source Llama strategy and its consumer Meta AI assistant will pay back the spend at the same rate.

What the number actually means

Seven hundred and twenty-five billion dollars is more than the combined annual capex of the U.S. railroad, electric utility and oil-refining sectors. Bloomberg points out that on current trajectories, AI capex from the top five U.S. tech firms will exceed total federal R&D spending this year for the first time in history.

Most of the money flows to Nvidia GPUs, to custom in-house silicon (TPUs at Google, Trainium at Amazon, MAIA at Microsoft), and to the buildings and power needed to run them. Yahoo Finance's earnings recap notes that data-center construction labour is now the second-biggest constraint after GPU supply. Power utilities in Northern Virginia and Texas are reportedly turning down new AI campus connections because grid upgrades cannot keep pace.

Apple reports on Thursday and is expected to come in much lower on capex, in keeping with its more cautious AI rollout. Whether the wider hyperscaler spending pays off is the question every analyst on yesterday's calls eventually circled back to. Cloud revenue growth has held up so far. Whether end-customer enterprise AI demand justifies the capex through 2027 is, as TNW puts it, "the trillion-dollar bet of the decade."