August 14, 2026, before US open
Position
Long SKHY. Reference $166.19 pre-market, prior close $165.67. Fifty-two week range $124.80 to $194.80. Intended holding period of six to eighteen months.
Thesis
The case rests on one claim: high-bandwidth memory has stopped behaving like commodity memory, and SK Hynix is positioned to capture the difference.
Conventional DRAM is interchangeable and priced on spot, which is why the industry has spent four decades competing on cost per bit. HBM works differently. It is sold under multi-year agreements at negotiated prices, qualified customer by customer, and constrained by advanced packaging capacity that sits outside the memory maker’s own fabs. A supplier who wins qualification holds that socket for a product generation.
The second quarter print is where this shows up. Operating profit of ₩60.54 trillion on revenue of ₩79.32 trillion works out to a 76% operating margin, higher than TSMC’s. A commodity producer does not earn a foundry’s margin, even at a cycle peak.
The packaging constraint is also easing in a way that favors the incumbent. TSMC disclosed on August 11 that its 5.5 times reticle CoWoS solution entered mass production, accommodating up to twelve HBM4 stacks at yields above 98%. HBM4 volume in 2027 required that capacity to exist at production yields, and it now does. SK Hynix is the primary qualified supplier into it.
There is a further possibility that tightness widens rather than resolves. On August 14 SK Group chairman Chey Tae-won said the company is weighing a joint venture structure for new fabs, specifically to share capital expenditure and overcapacity risk, and that supply could tighten further into 2027. He is talking his own book and I discount it accordingly. It is still the clearest indication so far that the supplier does not intend to solve the shortage by building alone.
Valuation is a supporting observation rather than a reason. The stock trades near 19.5 times trailing earnings against Micron at 20.7. Trailing multiples flatter any company whose earnings are expanding this fast, so I do not lean on this.
Catalyst
The 2027 HBM4 contract round, negotiating in the second half of this year.
Every HBM price currently in force reflects an agreement struck when supply was the binding constraint. The 2027 round is the first real price discovery since the sold-out period began. It is where I find out whether pricing power survives contact with CXMT capacity and with customers who have now absorbed two years of increases. Flat or higher pricing confirms that HBM has escaped commodity economics. Concessions would be the earliest read available on saturation, arriving two to three quarters before it reaches any income statement.
Risks
The largest risk is that the instrument does not track the thesis. On August 6 the ADR fell 10.37% on SanDisk NAND guidance. On August 12 it rose 8% alongside SanDisk, Western Digital, and Micron. It rose again on August 13 and 14 on cooler US inflation data and on SanDisk’s capital return announcement. That is four observations unrelated to Korean fundamentals. If SKHY prices as US memory-complex beta, then my Korean sourcing does not reach the price, and my reason for preferring it over Micron is weak.
CXMT is the competitive risk. Its July 27 Shanghai listing raised ¥57.92 billion, earmarked for domestic HBM3 mass production by the end of 2026. Current output is trailing-node commodity and absent from leading-edge HBM. That may well hold. It now has capital attached to the question.
Downstream, demand destruction is already visible. Samsung’s mobile division posted the first operating loss in its history in the same quarter its chip division set a record. TrendForce has third-quarter conventional DRAM contract prices decelerating to plus 13 to 18% from 58 to 63%, attributed to buyers who cannot absorb further increases. Alphabet has raised phone prices citing a severe memory crunch. HP reports on August 26 having already guided to a margin trough.
Sell-side consensus is also one-directional. Every analyst initiating coverage since the July listing has been bullish, which tells me nothing about whether they are right and does tell me the marginal buyer is scarce.
Won exposure is unhedged and cannot be hedged in this account. The ADR is five weeks old, so its premium or discount to the Seoul line has no history I can reason from.
Entry
The stock has risen 27.7% from its August 6 close of $130.17. The KOSPI closed today at a record 6,977.94, a fifth consecutive gain, up 11.5% on the week after seven losing weeks, with roughly $2 billion of foreign net buying. A soft US inflation print and a NAND company’s capital return announcement did most of that work. Though the consecutive days of momentum aren’t favorable for entry, this doesn’t change the underlying fundamentals and value I see in SK Hynix.
Limits
As seen by the investment-mandate, the position is capped at 30% of the account with thesis invalidation at 25% price stop, if 2027 HBM4 contracts settle below 2026 levels, if SK Hynix posts two consecutive quarters of declining HBM revenue, or if conventional DRAM contract prices turn negative quarter over quarter.