IBM Says Delayed Capex, Not Demand Loss, Cut Its Outlook

IBM cut its full-year sales outlook after softer mainframe demand and weakness in capex-sensitive parts of its portfolio. On-screen Bloomberg coverage framed the development bluntly: “IBM cuts outlook on softer mainframe demand.” ? arvind-krishna characterized the shortfall as a late-quarter reprioritization of capital expenditure among a subset of Fortune 100 customers, not as evidence that IBM’s pipeline had been overstated.

Krishna said IBM reviewed the deals that did not close and checked with clients that the opportunities were real. About one-third of the slipped business had already returned, he said. That is the basis for management’s view that customers deferred projects rather than abandoned them; Krishna said the next several months would determine whether that interpretation holds.

The immediate source of the disruption, Krishna said, was higher infrastructure-component pricing. He cited memory prices rising three to four times over the prior 18 months, networking infrastructure costs up 60% to 80%, and rising fiber and connector costs. Customers had shifted capital spending toward servers, storage, and memory late in June.

IBM had not seen that spending shift continue in July, Krishna said—a point also carried in Bloomberg’s on-screen headline: “IBM CEO: Didn’t see customer spending shift continue in July.” But he allowed that capex caution could reappear around the end of September, when much enterprise capital spending is committed. If it does, IBM expects to reach the low end of its 4% to 5% growth range; if delayed deals return, it expects the high end.

Krishna pointed to demand for IBM’s current Z17 mainframe as a counterweight to the weaker quarter. Since its May 2025 introduction, the Z17 has delivered aggregate capacity growth of 130% relative to its predecessor, he said, while 85% of clients have increased capacity. He called that one of the strongest patterns IBM has seen in a long time, though he also said software revenue trails hardware-capacity growth.

Romaine Bostick pressed Krishna on the risk that IBM was mistaking a structural computing transition for a temporary slowdown. Bostick invoked the early 1990s, when pressure on IBM’s mainframe business was initially discussed as temporary even as workloads were moving toward Unix-based systems.

Krishna’s answer was that customer behavior and unit economics, rather than IBM’s view of its own technology, should decide the question. In the earlier transition, he said, IBM could not plausibly claim that the mainframe was cheaper for workloads moving to midrange Unix systems. If a cheaper alternative emerges for a workload, he said, migration follows over time.

For a narrower class of workloads, Krishna argued that the economics now favor the mainframe. He said systems requiring protection, resilience, and burst capacity—credit-card authorization was his example—can be five to 15 times cheaper to run on a mainframe than elsewhere. For those workloads, he called the mainframe the architecturally superior platform.

That argument is deliberately bounded. Krishna did not say every workload belongs on a mainframe; he distinguished workloads with those requirements from those that do not. His defense of the installed base rests on the claim that there is not presently a cheaper alternative for this particular class of work, unlike the workloads that left the platform decades earlier.

The central commercial adjustment is to direct more resources toward software that does not depend on customer capital budgets. ? arvind-krishna said 80% of IBM’s software business is already annuity-, consumption-, or operating-expense-based and is growing at about 8%. The remaining 20% is capex-oriented and more exposed if customers continue to reprioritize infrastructure spending.

IBM plans to add forward-deployed engineers and technical help for customers implementing software, with the aim of accelerating growth in the recurring portion of the portfolio. Krishna named Red Hat, Confluent, and Hashi as examples of products that fit that model. On the capex side, he said IBM should continue serving the market without depending on outsized growth from it.

Supply-chain execution is part of that response. Krishna said IBM ended the second quarter with $500 million in backlog across distributed infrastructure and needs to ensure it can meet demand for storage and Unix systems.

He also pushed back on the premise that IBM’s cash-flow improvement has primarily come through cuts. Over the past four years, he said, adjusted EBITDA has been the main driver of cash-flow growth. IBM’s model has been to grow revenue roughly 4% to 5% while growing cash flow 7% to 9%, because incremental revenue is more productive and profitable than the first dollars of revenue. If revenue growth falls by one point, he said, productivity can make up the difference.

That does not necessarily mean headcount cuts. Krishna said IBM’s headcount has been “more or less flat” for years and that it sees substantial room to become more efficient in third-party spending.

The Starbucks example illustrates both a vulnerability and an opportunity in this strategy. Bloomberg reported that Starbucks was replacing some IBM software with internally developed tools. Krishna said Starbucks is a little over a $2 million-a-year IBM client and that the product being replaced is Tririga, a real-estate lease-management tool whose installed version at Starbucks is nearly 10 years old.

He said he was unsurprised by the replacement. Software that is principally interaction-based and depends on ease of use, rather than specialized capability, can be replaced by AI and AI agents, he argued. But he suggested IBM could still expand its relationship with Starbucks through Hashi or security offerings, and said he would not be surprised if Starbucks became a larger IBM client in the following year.

Ed Ludlow asked what supports IBM’s confidence in its revised guidance while infrastructure costs remain elevated. Krishna began with a macro assumption: global GDP growth of roughly 2% to 3% for the year, including regions facing energy disruption and instability.

He expects technology spending to grow two to four percentage points faster than GDP, putting the broader technology market in a 5% to 7% range. IBM’s exposure is uneven. Krishna said the recurring software business and distributed infrastructure can participate directly in that demand, while consulting is likely to grow only 1% to 3%, not at a double-digit rate. IBM nevertheless sees demand for transformational work in its signings and client discussions, he said.

Krishna estimated that technology once represented about 3% of the average enterprise budget and is now closer to 5% or 6%. By 2035, he said, he would not be surprised if it reached 10%. That is a long-run management forecast, not a solution to the present quarter’s capex pressure. In the nearer term, IBM’s case depends on whether its recurring software and distributed-infrastructure businesses can capture the categories of spending that customers continue to treat as necessary.

? arvind-krishna said IBM hired three times as many college graduates this year as it did the prior year. As other companies appear to reduce college hiring, he sees an opportunity to bring in talent and develop it internally. IBM has roughly 20 million resumes in its applicant database, he said, so candidate supply is not his principal concern.

The harder issue is whether employees can do the work employers will demand. Krishna said he does not expect an overall lack of employment, but expects AI and productivity tools to become a condition of competing effectively with peers. The implication for IBM is that hiring alone is not enough: the company needs employees able to deploy and operate technology with clients as it shifts resources toward more technical implementation support.

Quantum computing sits outside the near-term guidance case. Krishna said quantum is now an engineering challenge rather than a purely scientific one over the next five years. He expects IBM to deliver a large-scale, fault-tolerant quantum computer within two years and a machine capable of 100 million computations by 2029.

The commercial projection is much further away. Krishna said IBM and third parties estimate quantum could create about $1 trillion in total market value by the end of the 2030s. He also said IBM had announced that morning that it bought HRL from GM and Boeing, an acquisition intended to add expertise in materials, spintronics, quantum sensing, and related sensor technologies. The near-term claim is about building a fault-tolerant machine; the trillion-dollar estimate is a longer-dated view of the value quantum could create if that technology becomes commercially useful at scale.