The Memory Trade Just Became The Hardest Call On The Book

SK Hynix posted numbers that would make a hedge fund manager weep, and the market shrugged.

Let me lay it out. In Q2 2026, this South Korean memory maker reported record revenue of KRW 79.32 trillion, an operating profit of KRW 60.54 trillion, and a staggering operating margin of roughly 76%. Net margin? About 118%. Profits up 257% year-over-year on the top line, 557% on the bottom. For context, a year earlier that operating margin sat near 41%. This wasn't a good quarter for SK Hynix; it was a generational one.

And then there's the stock. As of October 24, it was trading at $137.02, down 26% on the month and 19% year-to-date, after the KOSPI-listed shares had been up roughly 159% year-to-date at one point in October. It had been sitting around 31.4% below its 52-week high of $199.87. The market, in other words, has already decided the story is priced. Loudly.

Here is where it gets interesting, and why I'm not going to pretend this is a coin flip.

You're Not Buying A Memory Stock. You're Buying A Monopoly At A Fire Sale.

Let's talk about what's actually happening under the hood, because the headline numbers are still understating the moat.

SK Hynix holds roughly half the HBM market and is the leading HBM supplier to NVIDIA. It kicked off volume shipments of HBM4 in Q2, signed multi-year supply agreements with roughly 10 customers, and already shipped 12-layer HBM4 samples pushing up to 16 Gbps per pin. HBM4 burns about three times the wafer capacity of conventional DRAM to build. That's not a feature; that's a supply-side choke point, and SK Hynix sits on the good side of it.

The inventory picture is almost surreal. Samsung and SK Hynix are holding fewer than 10 days of finished DRAM inventory. DRAM average selling prices jumped roughly 30% quarter-over-quarter in Q2, with another 21% QoQ gain penciled in for Q3. NAND ASPs were moving even faster, into the mid-50s percent range. And 2027 HBM capacity? Reportedly sold out. Hyperscalers are extending supply horizons out to 2031, some pushing toward 2035.

This is not a company begging for orders. This is a company with an assembly line running hot and a backlog that writes its own checks.

So Why The Blood?

Because the tape has a cruel sense of humor, and "record earnings" is not the same as "record reaction."

The bears aren't dumb. They've got real ammunition. SK Hynix's DRAM revenue share slipped to 24.9% in Q2 from 28.8% in Q1 as Samsung accelerated its own HBM4 push and clawed toward 39.4% share. There are reports HBM4 shipments may have missed expectations for a second straight quarter, tangled in NVIDIA's Rubin rollout delays. Analysts flag that the 76% margin embeds "extraordinary pricing," and that memory margins normalize eventually — they always do. A certain analyst model even implies a demanding ~22% ten-year EBITDA CAGR just to justify the current price.

I'll give the bears their points. They earned them.

But here's the thing that makes me uncomfortable sitting on the sell side of this argument: none of that is new information, and the stock already fell about 31.4% off its high on it. The tape digested the Samsung overtake, the HBM4 teetering, the margin-sustainability nagging — and kept falling. That's not a market pricing in bad news. That's a market that ran out of buyers and is now repricing a monopoly like it's a commodity.

The Setup That Matters

Look at the valuation, because it's the kind of divergence you don't see twice. The current P/E sits near 8.9, forward P/E at a near-absurd 3.93, price-to-sales at 7.74 against $129 billion in sales. Meanwhile the consensus target is $258.45 — nearly double where it trades. Consensus targets are rarely right, but a target that's well above the ask usually means someone in that room thinks you're wrong about the severity of the story.

And there's a floor being laid right under the stock that most people ignore. In August, the board approved a KRW 40 trillion repurchase-and-cancel program — the largest treasury-share cancellation in South Korean market history — covering about 3.3% of shares. They pushed their shareholder-return commitment above 50% of free cash flow. They're also dropping KRW 54 trillion into new DRAM (Yongin) and NAND (Cheongju) fabs. When a company with a balance sheet this fortifies (debt-to-equity of 0.08, cash ratio that would make a treasury secretary proud) starts buying back its own shares at fire-sale prices while profits are printing at 76% margins, management is telling you something.

My read? The buyback is the smart money's put option against your fear.

The One Thing I'm Watching

October 28, when SK Hynix reports Q3. Consensus is pointing at record territory again — revenue around KRW 98.2 trillion, operating profit near KRW 77 trillion, an operating margin creeping toward 78%. The real questions are the ones the numbers can't answer cleanly: Is HBM4 actually ramping or still stalling? What happened to ASPs in Q3? And did they lose more share to Samsung?

If the call confirms the HBM4 execution is fixed and margins hold, the current discount looks like a mistake the market will correct fast. If it confirms the delays and the share bleed, then maybe the bears were right and the normalization story is just getting started.

Either way, I'd rather be on the right side of the tape than fence-sitting. The fundamentals haven't deteriorated in lockstep with the price. They've accelerated. The only question is whether enough people are willing to buy a monopoly at 8x forward earnings before someone taps them on the shoulder.

I'll be looking. And if the Q3 report lands clean, I suspect a lot of people will wish they had been.