There is a particular sound a market makes when a blue chip breaks. It isn’t a crash, exactly. It’s the sound of a dinner party when someone drops the good china — a sharp intake of breath, then everyone pretending to continue their conversations while doing arithmetic in their heads.
Thursday morning, before the open, IBM (IBM) warned that its second-quarter results would disappoint. Consulting demand is softening. The AI roadmap — the thing that was supposed to make a 112-year-old company young again — is foggier than management had let on. By the close, the stock had fallen 25% — the worst single trading day in the company’s history, and this is a company whose history includes 1929, 1974, 1987, the dot-com unwind, and 2008. It took more than a century of practice to have a day this bad.
CEO Arvind Krishna blamed macro headwinds and software pressures, which is the corporate equivalent of blaming the weather for the state of your roof. The actual message, once you strip the conference-call upholstery off it, was simpler: the backlog isn’t there. Enterprises that were supposed to be signing nine-figure AI transformation engagements are instead doing what the American consumer is doing this summer — looking at the price tag, nodding thoughtfully, and putting it back on the shelf.
Silicon and Slideware Are Not the Same Trade
For two years, “AI” has been sold to investors as a single trade — one big rising tide, sloshing indiscriminately over chipmakers, cloud vendors, consultants, and any company whose CEO could pronounce “large language model” on an earnings call. Thursday was the day the tide went out selectively, and it turns out the trade was actually two trades wearing one trenchcoat.
The first trade is silicon: the chips, the memory, the foundries, the physical stuff that has to exist for any of this to run. That trade is underwritten by signed capacity commitments and order books you can read. The second trade is slideware: the consulting engagements, the transformation roadmaps, the promises to install the future at $600 an hour. That trade is underwritten by enterprise budgets — and enterprise budgets can be postponed with a phone call.
IBM is not a bad company. It is an important company having an honest moment in public, which is rarer and more painful. But here is the cold arithmetic of Thursday: if the earnings picture could deteriorate that much, that fast, without the market having any idea, then the market’s information about the business was stale — and everyone holding the stock was, whether they knew it or not, trusting a narrative rather than a number.
Is It Cheap Now? Sure. So Is a Flooded Car.
The tempting question, the one that arrives by text message from a brother-in-law around 2 p.m. on days like Thursday: isn’t it a buy down here? And the honest answer is that a 25% markdown tells you the price changed; it tells you nothing about whether the problem is fixed. A flooded car is also 25% off. Everything works except the part you can’t see, and the part you can’t see is the part that matters.
What would change our mind is boring and specific: two consecutive quarters of stabilized consulting billings. Not a press release. Not a partnership announcement. Not a new roadmap with better fonts. Billings — the number that measures whether enterprises are actually signing, or just taking meetings. Until then, catching this particular falling mainframe is a job for someone else’s money.
The most instructive part of Thursday wasn’t IBM alone — it was the split screen. The same morning IBM fell down the stairs, UnitedHealth (UNH) beat expectations and raised its guidance for the year. One blue chip repriced 25% down; another calmly told the market things were better than promised.
That’s not noise. That’s the sorting we’ve been writing about all summer: the must-pay economy (health premiums, staples, the electric bill) keeps performing, while the can-wait economy (consulting engagements, new sneakers, the furniture) keeps getting postponed. Thursday just rendered it in unusually high contrast.
The general principle, and it’s the one we’d want every client to take from the week: single-name risk is the risk you feel after it happens. Nobody holding IBM on Wednesday night thought they were one phone call from a 25% drawdown — that’s precisely what single-name risk means. The defense isn’t clairvoyance. It’s sizing: building the book so that no one company’s worst day in 112 years gets to be the worst day of your retirement. If you’re not sure how much of your plan rides on any single name behaving, that is a fifteen-minute conversation, and it’s free.
IBM stays off the books — avoided, not bottom-fished. The thesis: a pre-market warning that erases a quarter of a blue chip’s value means the consulting backlog deteriorated faster than disclosure kept up, and we don’t underwrite management credibility mid-repair. Re-entry test is two consecutive quarters of stabilized consulting billings, nothing softer. The silicon half of the AI trade is unchanged and unbothered: Broadcom (AVGO) on Apple’s signed $30 billion commitment, Taiwan Semiconductor (TSM) as the foundry that has never missed, Micron (MU) on $250 billion of committed U.S. capacity — all held. UnitedHealth (UNH), which beat and raised into the same tape, stays held as the must-pay economy’s standard-bearer. The broader lesson lands in sizing discipline: every model book is built so a −25% day in any single name is survivable by construction. Thursday was the argument.