Morgan Stanley collected 2.3 billion dollars in debt and equity capital markets fees in the first half of 2026, up from 1.4 billion dollars in the same period a year earlier, and the bank’s own numbers say the driver was AI infrastructure financing rather than the usual mix of corporate refinancing and opportunistic issuance. A 64 percent jump in fee income at a single bank would be a notable data point on its own. What makes it a signal rather than a one-off is that Morgan Stanley’s CEO has been explicit about where he thinks the cycle stands: by his own account, the AI infrastructure buildout is only 10 to 15 percent of the way through its investment cycle. If that estimate is even directionally right, the fee growth banks are reporting now is closer to the beginning of a multi-year financing wave than the peak of one.
The forecasts keep getting revised up, not down
The more telling detail than any single dollar figure is the direction of Morgan Stanley’s own forecast revisions. The bank’s data center capex estimate for 2026, set in November of last year at roughly 575 billion dollars, is now tracking closer to 850 billion dollars. The 2027 estimate, previously around 700 billion dollars, now stands at 1.3 trillion. The 2028 figure is projected at 1.5 trillion, with Morgan Stanley Research putting cumulative AI-related infrastructure investment near 3 trillion dollars by that year and more than 80 percent of it still to be deployed. Forecasts that get revised upward every quarter, rather than converging toward a stable number, are themselves evidence of a cycle still in its early, accelerating phase rather than one approaching maturity.
This is a banking-system pattern, not a Morgan Stanley story
The same week Morgan Stanley reported its numbers, JPMorgan and Goldman Sachs both posted record quarters of their own, and all three of Wall Street’s top trading desks beat equities estimates by hundreds of millions of dollars. Global AI-related debt issuance is on pace to reach nearly 570 billion dollars in 2026, more than double the prior year, with roughly 236 billion dollars already priced through the end of May, about four times the volume over the same stretch in 2025. The specific deals behind those totals read like a buildout still ramping rather than one leveling off: Amazon’s record Canadian dollar maple bond and the largest euro-denominated corporate bond ever sold, a roughly 13 billion dollar financing package Morgan Stanley and JPMorgan are arranging for a Meta data center in Texas, and a 4.25 billion dollar bond from Hut 8 to fund a single Texas data center project. None of that activity resembles a financing market clearing out a backlog. It resembles one still building capacity to meet demand that keeps arriving faster than the capital markets can price it.
Why the debt is the real tell
Hyperscaler capital spending in 2026 is on pace to consume close to 100 percent of operating cash flow, against a ten-year average closer to 40 percent. That is a genuinely elevated leverage posture, and it deserves to be tracked as a credit risk in its own right. But it also cuts in favor of the early-cycle framing rather than against it. Mature, self-funding businesses do not need to tap the maple bond market or arrange 13 billion dollar structured financings for a single data center. The fact that the world’s largest, most cash-generative technology companies are increasingly financing this buildout with other people’s capital, at the pace and scale the numbers above describe, is the clearest evidence available that the spending itself has not yet caught up to where the demand is being priced. That gap, not this quarter’s fee total, is what Wall Street is actually underwriting.
