IBM Just Told You the Emperor's Software Has No Clothes

Nobody wants to talk about IBM this week because it's inconvenient. Twenty-five percent gone in a session isn't a stock getting trimmed — that's a company getting re-rated by the market in real time, live, on camera, because it told investors that software and infrastructure demand is softer than expected. Big Blue isn't some scrappy pre-revenue AI wrapper trading on vibes. It's a hundred-plus-year-old enterprise IT bellwether with actual relationships inside actual Fortune 500 procurement departments. When IBM says demand is soft, it is reporting what CIOs are actually signing purchase orders for, not what they're saying on a keynote stage in Las Vegas.

And what CIOs are apparently signing purchase orders for, right now, is less than everyone modeled.

This is the quiet half of the AI story nobody wants on the front page, because the front page is busy with the sexier half — the $750 billion-plus in hyperscaler capex, the TSMC guidance raises, the Nvidia order book that supposedly stretches into the next decade. Fine. All true. Nobody's disputing the GPUs are being bought. The question this earnings season was always going to answer, eventually, is whether the applications built on top of all that silicon are generating enough enterprise revenue to justify the spend. IBM just handed you an early, ugly data point, and the market's response — vaporizing a quarter of the market cap in a day — tells you how little cushion was priced in for anyone who dared to disappoint on that specific question.

Compare that to what happened one day later. Chipmakers got hammered — Micron down double digits, AMD off nearly 7% — not because anyone found a flaw in the silicon, but because Meta quietly announced it has enough spare GPU capacity to start reselling it. Read those two data points together and stop pretending they're unrelated. One company that builds the compute says it has more than it needs. Another company that's supposed to be selling AI-powered software into every enterprise on earth says the demand for what runs on that compute is coming in soft. That is not two separate stories about two separate sectors. That is the same story, told from both ends of the pipe, in the same week.

Everyone's favorite rebuttal here is "but earnings are still growing 20%-plus across the index, second straight quarter." Sure. Read the composition of that growth before you cite it like it settles anything. A market where the entire multiple expansion of the last two years rests on the promise that AI infrastructure spending converts into enterprise software revenue cannot simultaneously wave off the one major enterprise software name that just told you the conversion rate is disappointing. You don't get to keep the bull case and discard the evidence against it in the same paragraph. That's not analysis, that's liturgy.

Warren Buffett picked a genuinely inconvenient week to go on television and say the quiet part out loud — that the current market rewards speculation over the kind of patient, values-based investing he's built a sixty-year career on. Cute timing, coming right as chipmakers cratered and IBM got taken to the woodshed on the same forty-eight hours. I'm not going to pretend Buffett has some prophetic edge on semiconductor cycles he's never once claimed expertise in. But when the single most successful long-horizon capital allocator alive says the tape smells like a casino, and the tape immediately proceeds to lose over a trillion dollars of chip-sector market value in the following days, maybe don't file that under "cranky old man doesn't get tech."

None of this means the AI capex cycle is a mirage. TSMC didn't raise its own spending guidance because it's worried — foundries with backlogs stretching years don't do that for theater. The compute is real, the fabs are real, the electricity bills are extremely real. What's suddenly, visibly less certain is the assumption that sat underneath every valuation multiple in this sector for two straight years: that demand for what gets built on top of all that silicon is as bottomless as the capex committing to build it. IBM just put a number on how untrue that might be, at least for one very large, very well-informed enterprise vendor, in one very bad afternoon.

The market spent two years pricing this sector for a world with no possible disappointment anywhere in the stack. This week, disappointment showed up in two different places at once — the supply side and the demand side — within about seventy-two hours of each other. That is not a coincidence you get to wave off with a earnings-season growth statistic and a shrug.