IBM CEO Arvind Krishna says delayed software deals are coming back; a third have already closed

IBM CEO Arvind Krishna says the delayed software deals are coming back—a third have already closed.

IBM lost more than a quarter of its market value in a single trading session on July 14, the sharpest one-day fall in the 115-year-old company's history. The trigger was an unusual pre-announcement—a letter from CEO Arvind Krishna warning investors that the quarter had gone worse than planned, with mainframe sales collapsing 42 percent and the software attached to those machines going down with them. Wall Street read it as the AI boom finally coming for Big Blue's oldest business. Krishna reads it differently. On the July 22 earnings call and in an interview afterwards, he made the same argument repeatedly: customers didn't walk away from IBM software, they just spent their money somewhere else first. Servers, storage and memory prices are climbing 15 to 30 percent as the AI infrastructure race eats the silicon supply chain, and enterprise buyers rushed to lock in hardware before it got worse. "The majority of what didn't happen in the second quarter was large capex deals at large clients," Krishna said. The mainframe order that slipped took three to four dollars of software and services with it—that ratio is why a hardware miss hurt IBM this badly.

IBM Q2 2026 results: a third of the delayed mainframe deals have already closed

The number Krishna kept returning to is the one he thinks settles the argument. Of the large deals that failed to close in the low tens by quarter-end, roughly a third closed within the first three weeks of the new quarter. He expects 80 percent of them to close inside six months, and says that pace is faster than deferred deals normally move. He is careful not to oversell it. "The fact that a third have already closed in the first three weeks gives us an indication, not yet full evidence, but a good indication that this was deferral and not destruction," he said. Evercore's Amit Daryanani put the question plainly on the call—deferred or destroyed—and that framing dominated the analyst Q&A. IBM still cut its full-year constant-currency growth guidance to 4-5 percent, down from the 5 percent-plus it forecast in April.

IBM Lightwell: a $5 billion bet on selling AI that cleans up what AI uncovers

The second half of the call was about where IBM thinks the money goes next. Krishna argued that Anthropic's Mythos release in early April sharply accelerated the discovery of security vulnerabilities in old open-source code sitting inside enterprise systems. IBM's answer is Lightwell—a $5 billion commitment, more than 20,000 engineers across IBM and Red Hat, and a subscription priced at $1 million a year to remediate and validate open-source packages whose maintainers moved on years ago. The early customer list is almost entirely finance: Bank of America, Citi, Goldman Sachs, JPMorganChase, Mastercard, Morgan Stanley, Visa, Wells Fargo, BNY, RBC and State Street. Krishna called it a multibillion-dollar addressable market. It is a neat commercial loop—AI is getting good at finding holes in legacy code, so IBM sells AI that patches them.

Red Hat, Power servers and quantum computing: the parts of IBM that didn't miss

Underneath the mainframe number, several lines held. Red Hat growth accelerated to 11 percent. Distributed Infrastructure had its best reported quarter ever at 37 percent, driven by Power servers and storage, exiting with roughly $500 million of backlog. HashiCorp and Confluent both performed well. Software revenue still rose 5 percent to $7.76 billion, and IBM generated $4.8 billion in free cash flow across the first half—while Alphabet turned cash-flow negative for the first time since going public. Krishna's longer bet sits further out. IBM plans to spend over $10 billion on quantum in five years, is building the Anderon wafer foundry with $1 billion in CHIPS money, and is buying HRL Laboratories. He now thinks quantum adds a trillion dollars of value by the end of the 2030s. Whether investors wait that long is a separate question.