IBM's Mainframe Slump Forces a $1 Million-a-Year AI Pivot

A 25% single-day stock drop is the kind of number that gets a CEO on the phone with every analyst who will take the call. For IBM, that call came after preliminary Q2 results showed mainframe revenue down 42%, prompting Arvind Krishna to spend much of last week explaining that artificial intelligence, not obsolescence, is reshaping how enterprises buy his company's hardware.

The quarter itself wasn't a disaster on paper. IBM posted $17.2 billion in revenue, $9.9 billion in gross profit, margins of 58%, and $2.2 billion in net earnings. But the mainframe collapse spooked investors enough to trigger the sell-off, and IBM followed up by trimming its full-year growth forecasts, a move that came just a week after the company had already warned investors of soft infrastructure revenue and margin pressure.

Delayed, Not Dead

Krishna's core argument is that tens of customers who were slated to buy new mainframes this quarter simply pushed the purchase, redirecting that budget toward data center and PC hardware instead, largely as enterprises scramble to build out AI capacity. CFO Jim Kavanaugh backed that framing with a structural point: IBM generates $3 in software revenue for every $1 of mainframe hardware it sells, meaning a stalled hardware refresh cycle threatens a much larger software tail. Krishna added that roughly a third of the large capex software deals delayed in Q2 had already closed in the first three weeks of the new quarter, suggesting the pause was tactical rather than a structural retreat from IBM's platform.

Complicating the picture is rising component cost. Data center and PC parts have jumped 15% to 30% amid the broader AI infrastructure boom, squeezing the same customers IBM is asking to prioritize mainframe refreshes.

Security Concerns Meet a New Product

IBM also pointed to a more unusual pressure: Anthropic's Mythos release in early April, which the company says has accelerated AI-assisted discovery of vulnerabilities buried in legacy codebases. That's a double-edged development for IBM's core enterprise base, and IBM's response was to productize it. The company unveiled Project Lightwell, an AI-driven service built to remediate and validate open source packages sitting inside legacy systems, priced at $1 million per year per subscription.

That price tag signals IBM is chasing its largest, most security-sensitive accounts rather than a broad market. Early adopters reportedly include Bank of America, Citi, Goldman Sachs, JPMorgan Chase, Mastercard, Morgan Stanley, Visa, and Wells Fargo, a roster that maps closely onto the financial institutions most exposed to legacy code risk and most able to absorb seven-figure software spend.

The bet embedded in Lightwell is straightforward: if AI tools are making it easier to find flaws in decades-old enterprise codebases, IBM wants to sell the AI tool that fixes them, turning a threat to its mainframe franchise into a new high-margin software line before competitors get there first.