IBM Drops 25% and Drags Corporate Software Down: AI is Eating the Budget of Those Who Sustain It

IBM's 25.21% drop on Wednesday, July 14th, marked the company's worst trading day since 1968, wiping out $69 billion in market value in a single session. CEO Arvind Krishna attributed the miss in the second quarter, reporting $17.2 billion in revenue against a consensus of $17.86 billion and earnings per share of $2.93 versus the expected $3.02, to a "sudden reprioritization" of corporate budgets at the end of June, with customers shifting spending from traditional software and infrastructure to AI chips and memory.

The contagion was immediate and asymmetric. In the same session, Salesforce dropped 2.1%, Workday 3.5%, Adobe 4.3%, and ServiceNow 5.8%. In a longer view, Adobe, Workday, HubSpot, Datadog, and Microsoft closed the day down over 3%. On the other side of the trading floor, TSMC would announce the next day an elevated guidance of over 40% growth in 2026, with 66% of revenue coming from high-performance computing. It’s the same money crossing the street.

The Easy Thesis and Why It’s Lazy

The market's reaction was to pin it as a bubble. Bearish analysts have been saying since March that spending hundreds of billions on GPUs to displace revenue from established SaaS is value destruction, and Ed Yardeni's note about a "dual bubble" gained traction this week. The reading has appeal but ignores an operational point: a Workday license renewal and a purchase of a training cluster do not compete on the same horizon. The license is recurring revenue that the company has already paid for three years, while the cluster is a one-time capex event. What Krishna described on the call is not a customer canceling SaaS, but a customer postponing expansion.

There is evidence that undermines the very thesis promoted by some in the buy-side. Analysts from Bank of America pointed out in a note on Tuesday afternoon that a good portion of IBM's shortfall came from lower-than-expected demand for the z17 mainframe, a specific product with a heavy refresh cycle. It's not the same dynamic as a Workday competing for HR management. Confusing the two is what typically causes stock prices to drop 5% before the analyst begins the earnings call.

Where the Effect is Real

The operational impact appears in two areas where the data is robust. The first is among Indian integrators. TCS reduced its headcount by 23,460 people in fiscal year 2026, the largest cut among majors, framing the change as a pivot to an "AI-first services model," requiring less bench from each client. Cognizant, in Project Leap, is evaluating between 12,000 and 15,000 global layoffs, with a focus in India. These cuts are the symmetrical pendulum to what IBM called reprioritization: if the client buys fewer licenses and more chips, the integrator that resells licenses and implements customization loses volume before the vendor loses revenue.

The second front is Germany. SAP was not among the companies that plummeted on Wednesday, but the company's revenue structure depends on S/4HANA projects that occupy the same budget slice that is now flowing to GPUs. A survey from the consulting firm Lünendonk earlier this year indicated that 61% of large German companies planned to reallocate between 8% and 15% of their IT budget for 2026 to generative AI initiatives. If Krishna is describing a broader behavior, the results reading from SAP in October will be less placid than European analysts have been signaling.

What No One Wants to Say

There is an ongoing reorganization that does not fit the bubble label nor the transformation label. Companies are discovering, at the speed of a quarter, that discretionary IT budgets are finite, and that four years of SaaS growth were sustained by a macro condition that is not repeating: low interest rates and abundant credit for clients who were expanding headcount. Now, with CFOs demanding freed-up cash, AI spending does not come on top of software spending; it comes in place of it. This is not a bubble, it’s substitution.

The challenge for the CIO of a bank in New York, for the integrator in Pune, or for the audit partner in Frankfurt is the same, although it arrives through different fronts. It’s not about deciding between stopping SaaS purchases and migrating to AI. It’s about renegotiating contracts with the understanding that the next wave of value will come from agents running on data that is currently locked in Salesforce, Workday, and SAP silos. Those who do not include this in the addendum to the 2027 contract will pay twice: for the license they already own and for the layer that goes over it. That was the operation IBM failed to execute before Tuesday's call. It should not be the only one.