Big Tech Is Set to Spend $725 Billion on AI Next Year. Can the Demand Keep Up?

The four largest U.S. cloud companies are on course to spend around $725 billion on capital projects in 2026, most of it on artificial-intelligence data centers, up about 77% from a year earlier. Broaden the lens to the whole industry and one closely watched estimate puts total AI-related capital spending near $1.6 trillion. Numbers that large invite a plain question: is this the early build of a technology that will pay for itself, or spending that has run ahead of the revenue meant to justify it? The market cannot decide. In the last completed session, the suppliers that equip and power the buildout jumped while several of the biggest spenders fell.

Foxconn’s AI hardware just passed half its revenue

The case for a demand-backed cycle starts with the orders. Foxconn, the largest contract manufacturer of AI servers, reported second-quarter net profit up 35% from a year earlier and revenue up 41%, with cloud and networking products clearing half of sales for the first time, at 51%. The company lifted its own capital budget by 30% and is opening plants in Texas and Mexico to build more, with shipments slated to begin in the fourth quarter. Buyers are paying now.

The spenders keep raising their figures for the same reason. Meta lifted its 2026 capital-spending plan to $125–145 billion, after an earlier increase, and Alphabet raised its ceiling alongside second-quarter results, citing the rush to open data centers. There is a further distinction from the late-1990s telecom and dot-com collapse: the buyers this time are cash-generating incumbents with established revenue and large cash reserves, a profile the speculative names of that era lacked, a point even skeptics of the buildout tend to concede.

$662 billion in leases the balance sheets don’t show

The worry is not the technology. It is how the buildout is being financed, and whether the returns arrive before the obligations do. Moody’s calculated that five hyperscalers carry roughly $662 billion in data-center lease commitments that are signed but not yet started, and so sit off the balance sheet under current accounting rules. That sum is equal to 113% of the group’s adjusted debt, and the full undiscounted lease exposure reaches $969 billion, a liability the agency describes as implicit debt.

More of the cash is now borrowed. Bond issuance by the hyperscalers has surged from $16.7 billion in 2024 to $193 billion so far in 2026, and the cost of insuring that debt against default through credit-default swaps has climbed since June. Investors have started to price the execution risk directly. When Alphabet detailed its rising capital budget at its last earnings report, the stock fell 7% as the market weighed larger outlays against a payback that is still years out.

What the data shows

The split runs straight through the tape. In the latest session, the enablers led: Vertiv (VRT) rose 4.3%, chip-equipment maker ASML gained 3.8%, and two power suppliers to the data-center trade, Eaton (ETN) and Constellation Energy (CEG), added 3.2% and 2.9%. Several of the heaviest spenders moved the other way, with Alphabet (GOOGL) down 3.8%, Oracle (ORCL) off 3.7% and Dell (DELL) down 3.7%. By sector, energy and utilities led while technology and communication services lagged. The broad market barely moved. The S&P 500 traded near 7,742 midweek and the Nasdaq near 26,543, with the VIX volatility gauge under 15. Money kept flowing toward what powers and cools the sites, and eased off the companies writing the largest checks.

What would settle it

A handful of scheduled markers will test each side. Nvidia’s next quarterly report, due in late August, will show whether data-center revenue is still accelerating or leveling off. The hyperscalers’ third-quarter results this fall will reveal whether 2026 capital budgets are raised again, and whether cloud revenue is keeping pace with the spending. The Federal Reserve’s September meeting sets the cost of the borrowing that increasingly funds the buildout. And Foxconn’s new AI-server plants in Texas and Mexico are guided to begin shipping in the fourth quarter, a check on whether committed capacity turns into product on the timeline promised. Each is a scheduled event.

For now the evidence points in two directions at once. The orders and profits are real and growing; so is the share of the buildout carried by debt and by leases that come due whether or not the AI revenue lands on schedule. Both can be true. What breaks the tie is not another forecast but the next set of results.

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