Four Dates, One Thesis

Four times this year I wrote about Apple in public, and each time I said something specific enough to be wrong. Not a mood, not a vibe about the culture at Cupertino, but a dated, falsifiable claim, the kind that either happens or does not. Most commentary about a company this large is built to survive contact with the future by never quite touching it. Say Apple is "behind," and you can be right forever, because "behind" has no expiration date. I tried to write the other way: name the mechanism, name the date it would show up, name what would prove me wrong. Three of those four pieces carried a claim built that way. This is the accounting on all three, done honestly, including the one number I got wrong the first time.

The week Ross Gerber went on television and said Tim Cook needed to go, Apple quietly closed a two billion dollar acquisition of an Israeli sensor startup called Q.ai, its second largest deal ever. The piece I published that Friday argued the market was reading the wrong signal. Nobody spends two billion dollars fixing a chatbot; you spend that kind of money on a platform shift, and platform shifts get assigned to whoever will own the next era, not the current one. The acquisition sat under hardware, not software, which meant it sat under John Ternus. I named him directly: head of hardware engineering, Cook’s presumptive heir, a man who would need a signature platform to define his own tenure the way the iPhone ecosystem had defined Cook’s.

That was a structural bet on a person, made from capital allocation rather than from any leaked memo or insider tip. It is the kind of prediction easiest to get right by accident and hardest to get right on purpose, because the tell was invisible unless you were already looking at org charts instead of press releases.

Three weeks later, Bloomberg ran a piece framing a global memory chip shortage as the crisis that would finally catch Cook flat-footed. My response argued the opposite: that Apple’s vertical integration and pricing power functioned as what I called a volatility exoskeleton, letting the company absorb a DRAM and HBM supercycle that was squeezing everyone else. I did not leave that claim unhedged. I set the falsifier in the piece itself, in writing, before the numbers existed: if Apple’s gross margin fell materially below 47 percent in the second fiscal quarter of 2026, the exoskeleton argument would weaken and I would update the view publicly.

That is the whole discipline in one sentence. A prediction without a number attached to its own failure is not a prediction, it is a mood wearing a prediction’s clothes.

By late May the coverage had settled into a single, comfortable sentence: Apple missed AI, Cook was too cautious, the operations man had finally run out of road. I argued that sentence was answering the wrong question. The right question was never where Apple’s own model was. It was where Apple wanted intelligence to live, and the answer had been visible since at least 2014: not in a chatbot Apple owned outright, but in the operating system layer underneath whichever chatbot the user picked. That same week, Reuters reported Apple was preparing a feature for iOS 27, iPadOS 27, and macOS 27 called Extensions, letting users route Apple Intelligence, Siri, and Writing Tools to third-party models while Apple kept the operating system, the privacy boundary, and the interface. I called that cloaking mode confirmed in product form, and set its own falsifier: it would only count if Extensions turned out to be real platform-level routing rather than a cosmetic menu of choices Apple still fully controlled.

Here is what each of those three claims looked like against what actually happened, dated against public reporting rather than against my own memory of having been right.

A calibration ledger that only ever reports being right is not a calibration ledger, it is a highlight reel, and the two are easy to mistake for each other if nobody checks the sourcing. So here is the correction, stated plainly rather than buried. An earlier draft of this piece, assembled from a secondhand summary of the Q2 results, cited Apple’s second fiscal quarter gross margin as 48.7 percent. That figure is wrong. Apple’s own newsroom release and the corresponding SEC filing report Q2 FY26 gross margin at 49.3 percent, up from 47.1 percent a year earlier. The floor still held, comfortably, both before and after the correction. But the discipline this piece claims to practice does not get to skip its own gate just because the correction happens to help the thesis. The number is fixed above; it was wrong in an earlier draft, and it is not wrong here.

There is a second nuance worth stating with the same plainness. Q3 FY26 gross margin of 50.1 percent is real, and it is also not pure structural margin. It included an approximate two percentage point favorable impact from tariff refunds, a one-off, not a repeatable feature of the business. Reporting 50.1 percent without that caveat would be technically accurate and quietly misleading, the exact failure mode this whole piece exists to argue against in other people’s coverage of Apple. So: the floor held at 49.3 percent on its own structural merits in Q2, and it held again at 50.1 percent in Q3 with roughly two points of that figure attributable to a refund, not to the exoskeleton argument itself.

None of this proves I understand Apple better than the analysts I was arguing against in real time. It proves something smaller and, I think, more useful: that writing a falsifier into a claim before the outcome exists, and then returning to check the claim against reporting rather than against memory, is a habit that survives contact with being wrong. Most of what gets published about a company like this is built to be unfalsifiable by design, vague enough to always have been directionally correct. The alternative is slower and occasionally embarrassing, as the 48.7 correction above demonstrates. It is also the only version of this kind of writing that means anything eight months later, when the receipts, not the headlines, are what is left.