Wall Street has spent months debating whether the money pouring into artificial intelligence is the start of a productivity revolution or the inflating of a bubble. JPMorgan has now weighed in firmly on the optimistic side — with conditions attached.
What JPMorgan argues
In a research note, the bank estimated that global spending on AI infrastructure is on course for around $5.5 trillion by 2030, as reported by Fortune — a figure other accounts of the same research put in a range of roughly $5 trillion to as much as $7 trillion, according to Data Center Dynamics. Its core argument: the spending is, so far, supported by real cash flows rather than pure speculation.
A quick translation. Capex — capital expenditure — is money spent on long-lived physical assets like buildings, servers and power systems, as opposed to day-to-day running costs. Hyperscalers are the handful of companies that run cloud computing at vast scale — chiefly Amazon, Microsoft, Google and Meta — and they are doing most of the spending.
The bull case
JPMorgan's optimism rests on the spenders' financial strength. The hyperscalers are generating more than $700 billion a year in operating cash flow and reinvesting roughly $500 billion of it into capital spending, the bank's analysis shows. In other words, the companies building AI data centers are funding much of it from their own profits, not betting the firm on borrowed money. Credit markets, too, remain willing: loans against AI data-center projects are being made at high loan-to-value ratios, a sign lenders are comfortable with the underlying assets.
The bank also argues the payoff will come less from consumer subscriptions than from productivity gains spread across corporate customers — automation and analysis in finance, healthcare and manufacturing — a demand base it views as durable.
The "for now" caveat
JPMorgan is careful not to call this a forecast of lasting success. Its own math sets a daunting bar: earning a 10% return on the AI investments modeled through 2030 would require about $650 billion of new annual revenue, sustained indefinitely, the bank found — an enormous figure to hit and then hold. And as annual funding needs climb past $1.4 trillion by 2030, the spending will increasingly have to be financed by debt and outside capital rather than cash flow alone.
The bank has also pointed to history, warning against repeating the late-1990s telecom and fiber build-out, when a flood of capital produced overcapacity, collapsing returns and bankruptcies. It expects this cycle to produce "spectacular losers" alongside winners.
The bubble debate
JPMorgan's note is a direct intervention in a live argument. Skeptics — including the investor Michael Burry, who shorted the mid-2000s housing bubble — have warned that AI valuations echo the final phase of the dot-com era, and that some hyperscalers may be flattering profits by depreciating AI chips too slowly. Others have called the financing structure overstretched and too concentrated in a few interlinked players.
JPMorgan's counter is narrow but pointed: as of mid-2026, the spending is anchored in measurable, growing enterprise revenue, not just hope. The key word in its own framing is "for now." Whether AI adoption scales fast enough to justify trillions in infrastructure between here and 2030 is, by the bank's own account, the open question — and the concentration of the spending in a few companies means any pullback would ripple far across suppliers, lenders and power developers.
