The Worst Day in a 115-Year-Old Company’s History — On a Day the Market Rose
Some sessions contain two facts that refuse to sit together. On Tuesday, International Business Machines lost a quarter of its value: down 25.2%, a $73.16 slide to a $217.07 close. Roughly $67 billion in market capitalization evaporated in a single session, leaving the company worth just under $205 billion. It was the worst day in IBM’s history as a public company — worse than October 19, 1987, Black Monday itself, when the stock fell 23.7% while the entire global financial system was coming apart around it.
Which is precisely the point. This time, nothing came apart around it. The S&P 500 closed that same day up 0.38% at 7,543.59. The Nasdaq Composite gained 0.9% to 26,107.01, carried higher by semiconductors. Even the Dow Jones Industrial Average — which contains IBM, and which the stock dragged on all afternoon — squeaked out a 9.63-point gain to 52,508.27. A Dow component had the worst day of its corporate life, and the index went up anyway.
Understand that contradiction and you understand the shift now underway beneath the surface of this market. IBM’s collapse was not the failure of a single company, and it was not a sign that artificial intelligence is disappointing. It was the opposite: the first visible invoice of the AI boom — delivered not to the skeptics, but to someone inside the tent.
What IBM Actually Said
The company did something firms only do when they have no choice: it pre-announced. Preliminary results released ahead of the scheduled date are an admission that the gap between expectation and reality has grown too wide to sit on for another eight days. Full second-quarter results and updated guidance are due Wednesday, July 22, after the close.
The real explosive is CEO Arvind Krishna’s explanation, not the number beneath it. In his letter to investors, he wrote: “In the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases.” Numerous large deals failed to close on the expected timeline. The company had modeled some supply-chain drag, he added, but “we did not anticipate the magnitude of the capex reprioritization.” Then came the sentence chief executives spend careers avoiding: “These conditions require our teams to execute perfectly, and this quarter we faltered. We did not adapt and move quickly enough.” What played out, he conceded, “was worse than our expectations.”
Read that twice, because the mechanics matter more than the mea culpa. IBM is not saying customers are cutting budgets. It is not saying demand is evaporating. It is saying customers had the money — and handed it to somebody else. They spent it on metal, chips and memory, because metal, chips and memory are getting scarce and expensive, and because software can wait while lead times cannot.
The Numbers, and What They Don’t Say
Revenue came in at $17.2 billion, up 1% year over year, against the roughly $17.86–$17.9 billion analysts expected. Adjusted earnings landed at $2.93 per share versus the $3.01–$3.02 consensus. At face value these are not catastrophic misses — roughly 4% light on revenue, 3% on earnings. Stocks do not normally shed a quarter of their value over a 4% revenue miss.
The segment detail explains the violence of the reaction, and simultaneously supplies the strongest argument against the panic. Infrastructure revenue fell 7%. Consulting was flat. But software revenue rose 5%, Red Hat grew 11%, and Distributed Infrastructure jumped 37% with a $500 million backlog. IBM’s software business did not collapse. It simply grew more slowly than the market had priced in.
So it was never the number that broke the stock. It was the story implied by the explanation: if clients are reallocating budget from software to hardware mid-quarter because hardware is running short, that is not an IBM problem. That is a problem for anyone selling software into the same budget line. On Tuesday the market did not reprice IBM’s quarter. It repriced a hypothesis about an entire industry.
The Mechanism: When Memory Becomes a Rationed Good
For that hypothesis to be more than a narrative, it needs a mechanism that outlives one quarter. There is one, and it is unusually well documented.
Conventional DRAM contract prices are set to climb 58–63% this quarter versus last, according to industry trackers. NAND flash contract prices are rising 70–75%. Not year over year — quarter over quarter. A DDR5 64GB RDIMM, the workhorse module of the enterprise data center, is on track to cost twice as much by the end of 2026 as it did in early 2025. The cause is a reallocation of historic proportions: AI data centers are estimated to absorb roughly 70% of high-end DRAM in 2026, an inversion of every prior cycle, with high-bandwidth memory alone consuming around 23% of DRAM wafers. Samsung and SK Hynix have effectively sold out their 2026 capacity through long-term agreements.
For a corporate IT buyer, the logic that follows is simple and inescapable. If the server you need in twelve months will be either unavailable or materially more expensive, you buy it now. The money comes from the same pool as the license renewal, the consulting engagement, the mainframe upgrade. So the license slips. Gartner has quantified the planning consequence: enterprises need to raise PC budgets 8–12% just to buy the same number of units. And relief is distant — most forecasts see meaningful new fab capacity arriving no earlier than late 2027 or 2028.
This is the heart of it. For two years the AI buildout has been a story in which everyone wins: chipmakers sell, hyperscalers build, software vendors get a shiny new category to upsell. IBM’s warning describes something else — a zero-sum reallocation inside a budget that is not growing to accommodate it. The AI boom has stopped lifting all boats. It has begun swamping some to fill others. And the trigger is not weak AI demand. It is AI demand that is too strong.
Contagion in 24 Hours: New York, Tokyo, Frankfurt
The market took this reading seriously enough that it did not stop at IBM.
In New York on Tuesday, ServiceNow fell 5.8%, Adobe 4.3%, Salesforce 2.1%, Microsoft 1.6%, with Intuit also under pressure. SAP’s U.S.-listed shares dropped as much as 5.5%. Bloomberg characterized the update as a “devastating blow” to software and services names. Goldman Sachs warned that the IBM event “will fully validate the software bear case scenario,” and braced for a broad sector selloff.
Overnight, the wave reached Asia. By Wednesday, Fujitsu had shed 5.5%, NEC 5%, Nomura Research Institute 5%, and BayCurrent Consulting nearly 7%. Japan’s IT services firms — companies that had said precisely nothing about their own quarters — were sold off because they are in the same business: selling software and consulting to enterprises now spending their money on memory.
Europe caught the wave Wednesday morning, though in muted form. SAP slipped about 1%, the DAX lost 0.73%, the CAC 40 and FTSE 100 each 0.21%, and the STOXX 600 hovered near the flatline. Three continents, 24 hours, one letter to investors.
ASML the Same Morning: The Other Side of the Same Coin
Then Wednesday morning delivered an illustration almost too neat to be true. While Europe’s software names absorbed the IBM shockwave, ASML reported — and produced the exact mirror image.
The Dutch company, without whose lithography machines not one modern AI chip exists, posted €9.3 billion in net sales at a 54% gross margin, comfortably beating consensus. It raised full-year guidance to €43–45 billion in sales at a 54–56% margin. It plans to ship roughly 65 Low-NA EUV machines in 2026, expects EUV revenue growth of about 45%, and intends to add some 30% more EUV capacity. For 2027, management said it is close to having all the orders it needs. The stock rose 3.3%. Notably, ASML also discontinued quarterly net bookings disclosure — a first in its modern history — while describing “very strong order booking” across the first half.
Same theme, same 24 hours, opposite signs. Sell the shovels to the gold rush and you raise guidance. Try to sell the miners something other than shovels and you post the worst day in your corporate history. That is not an anecdote. That is the market structure of 2026, told in two press releases.
Where the Money Actually Went
The useful exercise for investors is to stop asking whether the AI trade is intact and start asking which side of this reallocation a given holding sits on.
The receiving end is not hypothetical. Micron, Western Digital and Seagate sell precisely what IBM’s customers panic-bought in late June. Dell, Hewlett Packard Enterprise and Super Micro Computer assemble the servers those customers pulled forward. The tell is in the index math: on the very day IBM cratered, the Nasdaq closed up 0.9% on semiconductor strength. The dispersion between the iShares Expanded Tech-Software ETF (IGV) and the Philadelphia Semiconductor Index (SOX) is no longer a sector rotation story — it is a single budget being cut two ways.
The paying end includes the enterprise software complex broadly: Microsoft, Salesforce, ServiceNow, Adobe, Intuit, and the IT services layer — Accenture, Cognizant, Infosys — that sells the implementation work attached to those licenses.
One caution on the valuation math. HSBC downgraded IBM from Hold to Reduce and cut its target from $231 to $191 — not on operating weakness, but on valuation. Even after a 25% single-day loss and a 26% year-to-date decline, IBM trades at roughly 22.0x CY27 estimated non-GAAP earnings against a sector median of 16.9x. A quarter off the price does not make an expensive stock cheap. For U.S. investors, remember that any realized gain here is taxed at 0%, 15% or 20% long-term, plus the 3.8% net investment income tax above the threshold — and that a stock down 26% year to date may be more useful as a tax-loss harvesting candidate than as a falling knife to catch.
The Case Against This Thesis
A good narrative is not evidence, and the counterarguments here are strong enough to deserve real weight.
First, this may simply be timing. Krishna said deals failed to close — not that they were lost. Slipped deals can close in the third or fourth quarter, in which case revenue was deferred rather than destroyed, and Tuesday was an overreaction to a calendar problem.
Second, there is a compelling case for self-inflicted weakness. Krishna named it himself: faltered, did not adapt quickly enough. The mainframe cycle and a lumbering sales motion are IBM issues, not industry issues. Red Hat at +11% and Distributed Infrastructure at +37% show that where IBM sits close to the AI buildout, it is working.
Third — and this is the heaviest counterweight — IBM’s software revenue actually rose 5%. “AI is eating software” is a headline exaggeration. What was missed was the expectation, not the growth.
Fourth, this is one preliminary data point from a company widely regarded as a laggard in software. Extrapolating from it to Microsoft, Salesforce or SAP is an aggressive leap. The falsification test is still pending: if the pure-play software vendors report no comparable pattern in the coming weeks, this was idiosyncratic.
Fifth, memory cycles are cycles. They end. If new capacity lands from 2027, the crowding-out effect disappears as quickly as it appeared.
What Matters Now: July 22
What remains is a market that can no longer ignore an uncomfortable question. For two years the AI boom could be described as a rising tide. IBM’s darkest day is the first hard evidence that the tide also writes invoices — and that they are always paid out of the same budget.
The next test is specific. On July 22, after the close, IBM reports in full and issues new guidance. The quarter itself will not matter; it is known and digested. What will matter is what Krishna says about the third quarter. If the slipped deals close, it was timing. If guidance stays cautious because clients are still buying memory, then Tuesday’s market did not overreact — it reacted early.
Either way, the frame has changed. The question for the back half of 2026 is no longer whether AI spending is real. It plainly is. The question is who pays for it — and the answer, for the first time, is turning out to be somebody inside the technology sector itself.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
Read more in our topic hub: Topic Hub: AI Stocks Investing 2026
