IBM's 42% drop in Z mainframe revenue wasn't proof the mainframe is dead, but it did show how quickly AI hardware can steal the budget conversation.
IBM gave investors the kind of quarter you don't shrug off. On July 14, the company released selected preliminary second-quarter results, and the stock fell more than 25%, its worst one-day move on record by several accounts. The market noticed. IBM said revenue was $17.2 billion, up just 1% from a year earlier, while infrastructure revenue fell 7% and IBM Z revenue later came in down 42% in the full earnings release on July 22.
Arvind Krishna has a story he wants you to believe. In his July 14 letter to investors, the IBM chief executive said clients shifted late-June capital spending toward servers, storage, and memory to secure supply-constrained infrastructure before expected price increases. He also said IBM did not adapt quickly enough, and that numerous large deals failed to close when the company expected them to close. That is real pressure. But it is also convenient.
The mainframe didn't lose a technical argument in Q2. It lost a budget fight.
The mainframe defense
IBM's strongest defense is the z17 cycle. The company said on July 14 that z17 remained at nearly 130% program-to-program, ahead of z16, and that clients representing 85% of installed MIPs were maintaining or growing capacity. As The Register reported earlier this year, IBM had been talking up z17 as a strong mainframe launch, with Krishna and CFO James Kavanaugh pointing to IBM Z revenue strength after the new system arrived. That evidence matters. A platform customers are abandoning doesn't usually keep running ahead of the prior cycle.
SiliconANGLE reported after IBM's July 22 earnings that Krishna said about one-third of the delayed business had already closed in the first few weeks of Q3. Kavanaugh also told Reuters, according to the same report, that the mainframe stack of hardware and transaction processing software hit IBM's growth by more than five points in the quarter, when the company had expected only about one or two points of impact. Those are not small misses. They are the kind of misses that force investors to ask whether the sales cycle slipped or the spending priority changed.
Here's the thing. Both can be true. IBM can still have loyal mainframe customers, and those customers can still move money away from mainframe upgrades when AI infrastructure becomes the louder emergency. Mainframes process financial transactions, airline reservations, and other workloads that companies don't casually move. No one should pretend otherwise. But customers don't need to rip out a mainframe for IBM to feel pain. They only need to defer a Z upgrade, delay the software stack attached to it, and spend the quarter's money on memory and servers instead.
The budget fight is bigger than IBM
For anyone selling into enterprise technology, IBM's warning is a useful data point because it came from the old center of enterprise IT, not a venture-backed software vendor trying to explain a missed pipeline number. Axios reported on July 14 that Krishna said IBM did not anticipate the magnitude of the capex reprioritization, and that clients had moved spending toward servers, storage, and memory purchases in the last few weeks of June. Look at the mechanics. A deal can be strategic in April and still lose to hardware availability in June.
That is the uncomfortable part. Enterprise IT budgets are not expanding neatly to match every AI ambition. They are being reshuffled. If you're selling software, services, security tooling, or infrastructure that doesn't sit directly inside the AI buildout, you should expect more late-stage budget friction. The buyer may still like your product. The purchase order may still move. But the AI server bill is now sitting on the same desk, and it is not waiting politely for the old refresh calendar.
IBM's official July 22 release fixed one important point the earlier draft got wrong: the company did not cut full-year revenue to $11.5 billion to $12.5 billion. IBM now expects 2026 constant-currency revenue growth of 4% to 5%, down from more than 5%, and it still expects free cash flow to rise by about $1 billion year over year. That sounds dry. It is not. Guidance is where management tells you how temporary it thinks the problem is, and IBM's guidance says the company expects damage beyond one ugly week in June.
That matters. Krishna is right about one important thing: AI isn't killing the mainframe in any direct technical sense. A GPU cluster is not about to take over the world's COBOL transaction processing because an executive wants to sound modern on an earnings call. But AI is changing the order in which enterprise buyers spend money, and IBM just showed you what happens when even a sticky platform gets pushed down the list.
Q3 now carries the test. If the delayed Z deals keep closing and z17 stays ahead of the z16 cycle, July's stock reaction will look too harsh. If the pause stretches, IBM's problem becomes harder to dismiss as timing. The data will settle it. It always does.
Also read: ServiceNow bets $40 million on an Indian AI banking startup to lock down enterprise agent governance • STMicroelectronics posts 25% revenue growth as its AWS deal turns data centers into its fastest-growing business • Nokia beats Q2 2026 earnings estimates as AI data center demand rewrites the company's story
