Why the "cheap" memory/semiconductor stocks may not be cheap — a P/E analysis

**TL;DR:** Memory chips are in an AI-driven boom, and some of these stocks look "cheap" on *next year's* earnings (\~10× forward while AI peers sit at 40×). For a commodity cyclical, that forward-cheap is often a late-cycle *warning*, not a bargain...

Here's a more detailed POV:

A lot of us screen for low P/E as a value habit. For most businesses that's reasonable. For **commodity cyclicals — memory (DRAM/NAND/HBM), steel, shipping, solar — it runs backwards**, and memory is the cleanest example happening right now.

**The setup.** After a sideways 2022–23, memory makers cut production; then AI demand for HBM exploded. DRAM contract prices jumped \~90%+ in a single quarter recently, NAND 70%+, and the industry expects a price peak around Q3–Q4 2026. When a commodity is this hot, producers print money — and their stocks start to look "cheap" on P/E.

**Why the P/E inverts.** P/E = price ÷ earnings. At the cycle *top*, earnings are huge → P/E looks *tiny* ("cheap") right when risk is highest. At the *bottom*, earnings vanish → P/E looks *huge* ("expensive") right when it's safest. So a low P/E on a cyclical is often closer to a sell signal than a buy. (Peter Lynch also says this; memory is its purest case.)

**The 2026 wrinkle:** the *trailing* P/E on these names actually looks fairly normal (low-20s) because the price ran up with the earnings. It's the *forward* P/E — price ÷ next year's peak-extrapolated profits — that screens cheap (\~10×). And that's the trap: it's only cheap if peak earnings hold.

**Why peak earnings aren't real earnings.** Across a memory cycle, gross margins swing from **50%+ at the top to the low-20s or negative at the bottom**, revenue 25–40% peak-to-trough. The profit you'd be capitalising at a "cheap" multiple near the peak can halve or vanish in a year or two.

**What to use instead of trailing P/E:**

* Where you are in the cycle (spot vs contract prices, inventory days, fab utilisation)
* Normalised *mid-cycle* earnings, not the peak quarter
* Price-to-book (book value is far steadier than peak earnings)
* Cost-curve position (who survives the bust)

**And the flip side — where the money's actually made:** because the P/E runs backwards, a cyclical looks *best* (a low, reassuring multiple) near the *top*, and *worst* (P/E sky-high or negative, grim headlines) near the *bottom*, where the next upcycle is closest. So the discipline inverts: get interested when it looks ugly, wary when the screen flashes "bargain." The opportunity hides in the ugliness, not the cheapness.

**The honest other side:** this is the best "this time is different" case memory has ever had. DRAM consolidated to **3 players** (discipline); HBM is \~**4× more wafer-intensive** so every AI bit structurally tightens commodity supply; a lot of it is on **multi-year fixed-price contracts** (less spot whiplash); and capex is restrained — analysts don't model the next downcycle until **late 2027**.

So the real question isn't "is the P/E low," it's: **are these record margins the new floor, or the old peak?** Nobody knows. The tell that decides it: do the three makers stay disciplined, or chase capex into the strength? "Cautious" ≠ "short."

Curious what others here think — **new normal, or just a louder version of the same script?**