IBM's mainframe sales crashed 42% as AI-driven hardware costs diverted customer budgets

IBM Mainframe Sales Collapsed 42% as AI Costs Drained Budget

IBM's second quarter is what happens when the AI buildout gets expensive enough to rearrange enterprise IT budgets. Software revenue grew 5%. Mainframe revenue collapsed 42%. Same company, same quarter, same cause.

The Q2 report, released July 22, showed total revenue of $17.16 billion, up just 1% year over year and short of the $17.58 billion consensus compiled by LSEG 1. The segment breakdown reveals the split. Software, IBM's largest and highest-margin division, generated $7.76 billion 2. Consulting was flat at $5.33 billion 1. Infrastructure revenue fell 7% to $3.84 billion, dragged down by a 42% collapse in Z mainframe sales 1.

The damage hit the stock market before the official report. On July 14, CEO Arvind Krishna published a letter to investors warning that results would miss expectations, citing worse-than-planned performance in Z mainframe sales and transaction processing software 1. The stock dropped 25% that day, its sharpest single-day decline on record 1. When the actual earnings landed on July 22, shares rose about 4% in extended trading, a modest recovery after the numbers confirmed what investors had already absorbed 2. Through July 22's close, IBM is down roughly 30% in 2026 while the S&P 500 is up about 10% 1.

What happened to those mainframe sales? According to Krishna, customers who were due to buy new Z mainframes during the quarter chose not to. They redirected that money toward data center hardware and PCs, where costs had surged 15% to 30% due to component price inflation driven by the AI buildout 3. Krishna was firm that no one is abandoning the platform: "We see no evidence of clients moving off the mainframe" 3. The company's position is that these purchases are delayed, not permanently lost.

But CFO Jim Kavanaugh explained why a delay is more than a timing problem. IBM generates $3 in software revenue for every $1 of mainframe hardware it sells 3. When a customer defers the hardware purchase, the associated software contracts defer too. A 42% drop in mainframe revenue cascades into the Software segment through a lag.

IBM lowered its full-year revenue growth forecast to 4% to 5% in constant currency, down from the "over 5%" guidance issued as recently as April 1. Management reiterated its target of $1 billion in additional free cash flow for the year 1.

The budget pressure squeezing IBM's customers is not unique to them. Enterprise hardware makers including Dell, HP, and Apple have all warned that AI-driven component costs, particularly memory, are forcing price increases across their product lines 3. What makes IBM's numbers so stark is scale. A single mainframe system costs hundreds of thousands to millions of dollars, with maintenance contracts and software generating many millions more 3. When a customer defers a purchase that large, it shifts a quarterly P&L. A deferred PC refresh doesn't.

The same AI wave that grew IBM's Software segment 5% also delivered the company's Bob AI coding tool, now adopted by more than 80,000 employees 1. The same wave inflated the data center hardware costs that pushed IBM's customers to raid their mainframe budgets. IBM is collecting the upside of AI adoption in Software while absorbing its cost in Infrastructure.

IBM also faces a securities fraud investigation examining whether the company overstated the size of its sales pipeline before the July 14 warning 2. An investigation does not imply wrongdoing, but investors are pressing management on one question: is the missing mainframe revenue delayed, or permanently lost?

Krishna may be right that these sales return. The tech industry has predicted the death of the mainframe for decades, and it hasn't happened. But every previous threat came from a competing technology: client-server computing, then cloud migration, then distributed databases. This time, no technology is replacing the mainframe. The threat is that AI infrastructure got so expensive that mainframes lost the fight for budget. If that can happen to IBM's customers, it can happen to any enterprise running a mainframe while building out AI.