25% down. Worst day in 115 years. $67 billion gone. IBM’s own CEO admitted they “did not adapt and move quickly enough.” This is the story of what actually broke.
Image Source: X
IBM had the worst trading day in its 115-year history on Tuesday, July 14, 2026. The stock dropped 25.2% in a single session, closing near $217 a share. That single drop erased about $67 billion in the company’s market value, leaving IBM worth just under $205 billion.
To put that in perspective: even the 1987 Black Monday crash, when markets collapsed across the board, only took IBM down 23%. This drop was worse, and it happened on a day when the broader market was not in freefall. This was IBM’s problem alone.
Advertisement
The drop followed a letter CEO Arvind Krishna sent directly to investors that same day. In it, he got ahead of bad news that would normally wait for the official earnings report on July 22. He told investors the second quarter was “disappointing” and admitted, in his own words, that IBM “did not adapt and move quickly enough.”
Advertisement
That kind of pre-earnings warning from a CEO is itself a signal. Companies rarely volunteer bad news before they have to. When they do, it usually means the numbers are bad enough that waiting a week to disclose them would look worse.
To understand the miss, it helps to know what IBM was expecting going in. Earlier this year, IBM launched a new mainframe system called z17. It was the strongest launch quarter for a mainframe product in the company’s history. Because of that strength, IBM told investors back in April that Infrastructure revenue would only dip slightly for the rest of the year.
That did not happen. Krishna’s letter points to a shortfall specifically in “Z performance,” meaning the mainframe hardware line, and in the software that runs alongside it, particularly Transaction Processing software. That software is a meaningful revenue stream for IBM, since mainframe clients pay ongoing fees to run it.
Here is the part that connects IBM’s stumble to a much bigger trend happening across the tech industry. In the final weeks of June, IBM’s own clients changed their spending habits. Instead of buying software, they redirected their budgets toward buying servers, storage, and memory.
Why would clients suddenly want more raw hardware instead of software? Because AI data centers around the world are consuming huge amounts of exactly that kind of equipment. The buildout of AI infrastructure has made servers, storage and memory scarce and more expensive. IBM’s clients apparently decided to buy that hardware now, before prices climb even higher, rather than commit to new software contracts.
IBM says it saw some of this coming. The company had built supply chain disruption into its forecasts. What it did not predict was how large this shift would be. Krishna said this was the main reason a number of big deals failed to close when IBM expected them to.
There is a second thread to this story, and it involves Anthropic. Krishna’s letter mentions that clients were distracted this quarter by cybersecurity concerns sweeping the industry. That concern traces back to Anthropic’s release of a model called Mythos.
Anthropic said Mythos could identify cybersecurity vulnerabilities before companies even detect them on their own systems. For a company like IBM, whose mainframes run critical infrastructure for major institutions, that claim raises real questions. If a new AI tool can surface security holes that fast, clients may want to understand the implications before signing new long-term contracts. According to Krishna, that hesitation stalled several large deals during the quarter.
This is not IBM’s first run-in with an Anthropic product this year. Back in February, IBM stock fell 13% after Anthropic released a tool aimed at modernizing COBOL, an old programming language that still runs on many IBM mainframe systems. That was IBM’s worst day since the year 2000, until this week. Other software companies sold off around the same time on fears that AI tools would start automating the kind of work they charge clients for.
Put together, these two events point to a pattern. IBM’s core business has long depended on companies signing multi-year contracts for mainframe hardware, software, and support around legacy systems. That business model assumes a level of predictability that AI is now disrupting from two directions at once.
On one side, AI data center demand is soaking up the physical hardware that IBM’s clients need, pulling their spending away from software. On the other side, new AI capabilities are raising fresh questions about whether legacy systems are secure, which makes clients pause before locking into new deals. IBM is caught in the middle of both shifts.
Advertisement
