The Semi-Story Is Weaker Than It Looks
Part 2 of 3 · The chart says late-stage. Now we turn it face-down — and our own screening reveals that «semiconductors» is not one trade, but three, stacked by quality.
Part 1 (Climax Run), Part 2 (Story is weaker than you think), Part 3 (Semi-Sell Rules)
«It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.»
Mark Twain, American writer (1835–1910)
In Part 1, we read the tape: a textbook three-stage parabola, semiconductors at 22% of the S&P 500, record inflows, a five-sigma momentum stretch. The chart says late-stage.
But a stretched chart, on its own, is not a sell. Great trends get stretched and stay stretched for years. The chart only becomes dangerous when something else is true at the same time — when the story underneath the price is quietly weaker than the price assumes.
So this piece does what we always do at arvy: we turn the chart face-down and interrogate the «Good Story». We are quality investors first. Before we trust a parabola or fear it, we want to know two things — whether the earnings are real, and, more importantly, whose earnings are real. Because the most important discovery we have made about this sector is that «semiconductors» is not one trade. It is three different businesses wearing one ticker-cloud, and our own screening pulls them apart with uncomfortable clarity.
But first, the foundation. Are these earnings real at all?
The Check Writers and the Check Receivers
Here is the chart that reframed this entire sector for us.
Chart 1: The «check writers» versus the «check receivers» — hyperscaler spenders against AI-chip recipients
Check writers vs check receivers
It splits the AI complex into two groups. The check receivers — the chip and AI-hardware names that book the revenue from the build-out — have soared. The check writers — the hyperscalers whose capital expenditure is that revenue — have started to wobble, their stocks lagging as the bills come due.
Why does that gap matter? Because it exposes the mechanism underneath the boom, and the mechanism is circular.
A vast share of the AI-chip industry’s revenue is hyperscaler capital spending. One company’s capex is another company’s sales — and that spending cascades through the supply chain, booked again and again as revenue across chips, networking, and infrastructure. The reported profitability of the whole AI complex is inflated by this recycling of the same dollars.
And follow the money all the way back to its source. Where does the hyperscaler capex ultimately come from? Overwhelmingly, from advertising revenue — one of the most cyclical income streams that exists. So beneath the most celebrated secular growth story of the decade sits a deeply cyclical foundation: ad budgets funding capex, capex becoming chip revenue, chip revenue capitalised into trillion-dollar valuations. When the check writers slow down — and the chart shows them beginning to — the check receivers’ revenue is what slows next.
A secular story, resting on a cyclical base. That is the first crack.
Cyclicality Wearing a Secular Mask
Here is the trap that catches even good quality investors — we have watched it happen, and it cost a lot of careful managers dearly in 2022.
Nvidia was an unquestionably great company in 2018. It was an unquestionably great company in 2021. And the stock fell over 50% in the first case and roughly 66% in the second — not because the business was bad, but because it is cyclical, and the market eventually remembers that. (Tellingly, it then recovered to new highs within about eighteen months each time — which is precisely the double-edged lesson we return to in Part 3: the cyclical crashes were also the buying opportunities.)
This is the oldest and most expensive mistake in our profession: mistaking a cyclical business for a secular compounder. Semiconductors have always been cyclical — booms of over-ordering followed by brutal inventory corrections. The AI narrative insists this time is different, that demand is structural and permanent. Perhaps. But every cycle in the sector’s history has been accompanied by a story explaining why that particular boom would not end. It is the commodity cycle wearing a new costume.
When you pay a secular multiple for a cyclical business at the peak of its cycle, you are not buying growth. You are buying the most optimistic possible version of an inherently mean-reverting story.
The Tell: When the Price-Maker Becomes the Price-Taker
If you want to know how late it is in a cycle, do not watch the cyclical company. Watch its most powerful customer. And in the summer of 2026, the most powerful customer in consumer technology blinked.
Apple is the most disciplined pricer in the history of business — a company that has trained the world to pay more and feel privileged doing it. And in June 2026 it did something it almost never does: it raised prices across much of its lineup — Macs, iPads, and more — with no new chips, no new designs, no «one more thing». The same hardware as the week before, simply more expensive. A naked, pure cost pass-through, from the one company that never has to do that. The stock had its worst session in over a year, shedding hundreds of billions in value in hours.
Why would the great price-maker suddenly behave like a price-taker? Because the input underneath it had caught fire — one boring, grey input: memory. The same DRAM and high-bandwidth memory we have been circling. With AI demand pulling every available wafer toward high-margin memory for data centres, the supply of ordinary consumer memory was starved, and contract prices roughly doubled over a few quarters. By Apple’s own account, memory now costs it several times what it did a year earlier.
Here is the part that should make a quality investor sit up. Look at the two gross-margin lines over the last year. Apple sits where it always sits — a remarkable, steady mid-to-high-40s. The memory makers? Their gross margins exploded — Micron’s vaulting from the high-30s a year ago to the mid-80s last quarter, briefly richer than Nvidia’s, on revenue that more than quadrupled. That is not a healthy supplier earning a fair return. That is the sound of a supplier transferring its customer’s profit into its own pocket — and the customer in question is the most powerful brand on Earth.
Chart 2: Gross margins — Apple (steady ~mid-to-high-40s) versus Micron (~39% to ~85% in a year).
And this is precisely the evidence for our thesis, not against it. A cyclical business throwing off margins this far above its own history — margins so extreme they are visibly destroying the economics of its largest, most powerful customers — is not a business that has discovered a durable secular moat. It is a business at the violent top of its cycle. The squeeze is the symptom. When a sixty-year-old commodity industry suddenly has the most powerful company in consumer tech over a barrel, the right question is not «how much higher can the memory makers go» — it is «how much longer can a margin this abnormal possibly last». Peak power over your customers is what the top of the cycle feels like from the inside.
The Depreciation Time Bomb
Now the part that is most technical, and most damaging to the «earnings are real» case.
