Revenue came in at $17.2 billion, roughly $660 million below the $17.86 billion consensus, with operating EPS of $2.93 against expectations near $3.01.
But look at the segment lines and the software-led transformation story is intact. Software grew 5%, Red Hat accelerated to 11%, and Distributed Infrastructure posted 37% growth, its best in reported history. The thing IBM has spent a decade convincing investors to value went up.
What went down was mainframes, and the reason it went down is the interesting part, because it was not weak demand.
The budget got eaten from the inside
CEO Arvind Krishna's explanation was unusually blunt for a preannouncement. He said we faltered, and that deals failed to close as clients redirected late-quarter spending toward supply-constrained servers and memory.
Read that carefully, because it inverts the usual bad-quarter narrative. Enterprise customers did not cut their IT budgets. They spent them, on something else, in a hurry, at the end of the quarter. Money earmarked for IBM software and mainframe deals got pulled forward to buy servers, storage, and memory before those became harder to get.
That is a demand story disguised as a weakness. The customers wanted to spend. They just triaged, and IBM's products lost the triage to hardware that clients were afraid they would not be able to buy later.
Why the memory part is not a one-quarter problem
The mechanism underneath is worth understanding, because it determines whether this reverses next quarter or persists.
Memory manufacturers are converting conventional DRAM production to High-Bandwidth Memory, the specialized memory that AI accelerators require. The reported displacement runs at roughly a 3:1 ratio: every unit of HBM capacity added removes about three units of conventional DRAM capacity from the market.
That is a structural squeeze, not a temporary hiccup. The AI buildout is pulling memory-fabrication capacity toward accelerators, which tightens supply of ordinary server memory, which drives up prices and lead times, which pushes enterprises to buy conventional hardware now rather than risk shortage later. The panic-buying IBM ran into is a rational response to a supply shift that is still underway.
So the honest read is that this specific dynamic, clients front-loading hardware purchases at the expense of software and mainframe deals, could recur as long as the memory squeeze lasts. It is not obvious it resolves by next quarter, and that is precisely the uncertainty the market punished.
There is a second irony worth naming. IBM is one of the companies whose AI strategy depends on the same buildout that just cannibalized its quarter. The infrastructure race that starved IBM's software deals of budget is the race IBM is trying to sell into. The same force is a headwind and a tailwind, on different time horizons.
The part that is genuinely bad, separate from the memory story
Not everything here is a sympathetic timing problem, and it would be dishonest to frame it as one.
Krishna's own words were "we faltered," which is an admission of execution failure, not just bad luck. Deals that slip because clients reprioritize are partly IBM's problem to manage, and a company with a tighter grip on its late-quarter pipeline closes more of them. Preannouncing a miss eight days early, with the CEO removing the usual ambiguity, signals the gap was large enough that IBM did not want it landing cold on the call.
More seriously, there is now an investigation into alleged securities fraud concerning the manipulation of a sales pipeline by company executives. That is an allegation, unproven, and it should be treated as such. But it changes how the "deals slipped" explanation gets received, because "deals we expected did not close" and "the pipeline was represented inaccurately" are adjacent descriptions, and an investigation into the second makes scrutiny of the first unavoidable. Every forward-looking statement on the earnings call now carries added legal weight.
What the full call has to answer
IBM's official Q2 report and call landed the evening of July 22, and the preliminary release left the load-bearing questions open.
The first is guidance. IBM entered the year projecting more than 5% constant-currency revenue growth and about $1 billion of free cash flow improvement, and it did not update that with the preliminary numbers. Whether it reaffirms or cuts is the number that actually moves the stock, because a single quarter distorted by a memory-buying scramble matters far less than what management now believes about the full year.
The second is the deal pipeline. How many of the slipped deals have since closed, how many remain, and how many evaporated. Deferred deals that convert in Q3 support the timing narrative. Deals that vanished support the execution-problem narrative. That split is the whole story.
The third is free cash flow against the dividend. IBM has paid a dividend for 56 consecutive years, and a meaningful part of its shareholder base owns it for exactly that. First-half free cash flow of $4.8 billion suggests the dividend is not in question, but confirming the full-year figure is what that base will listen for.
The distinction that matters
Strip it down and this quarter posed one question: did IBM's business deteriorate, or did its customers' purchasing sequence temporarily turn against it?
The segment data leans toward the second. Software grew, Red Hat accelerated, the recurring-revenue engine ran. The miss concentrated in mainframe hardware at the end of a product cycle, worsened by a memory shortage that pulled client budgets toward commodity hardware. That is a real problem, and it is a different kind of problem than the market's 26% reaction implies, which typically signals a broken thesis rather than a mistimed one.
The counterweight is that "we faltered" plus a pipeline investigation is not a clean timing story either, and a company does not pre-release a miss this size unless the surprise was severe. Both things are true. The memory squeeze is real and probably not over, and IBM's execution and disclosure now sit under a brighter light than a simple supply-chain excuse would leave them.