
For most of 2026, the market treated SK Hynix and Samsung Electronics as a single trade: buy Korean memory, ride the AI infrastructure boom. That consensus fractured in the second week of July. On July 13, SK Hynix shares fell 15.4% in Seoul, the steepest single-session decline in the stock's history, even as the broader memory upcycle stayed intact 3. Four days earlier, Samsung had reported preliminary second-quarter operating profit of 89.4 trillion won (roughly 58.4 billion dollars), a 19-fold jump from a year earlier and the highest quarterly operating profit any technology company has ever posted 4 5. Samsung's own stock still fell as investors weighed heavy AI capital spending against the earnings beat 4. The two moves look contradictory on the surface. Together they mark the moment investors stopped pricing SK Hynix and Samsung as one undifferentiated memory trade and started pricing them on the specifics of who is actually shipping high-bandwidth memory, on what contract terms, and on what timeline.
The SK Hynix selloff was not a story about weak demand. Conventional DRAM spot prices rose roughly 30% quarter over quarter and NAND spot prices rose around 50%, confirming the AI-driven memory shortage that has defined 2026 3. The problem was mix and timing. A large share of SK Hynix's HBM volume is still sold under long-term agreements struck before this year's price spike, so its blended average selling price lagged what a fully spot-priced book would have earned 3. Layered on top of that, brokerages including Korea Investment and Securities and NH Investment and Securities pushed back their expectations for full-scale HBM4 mass production and shipment volume from the second quarter into the third, stripping out an upside catalyst analysts had already priced into estimates 3. Korea Investment projected SK Hynix's Q2 operating profit near 60.4 trillion won, about 8% below the roughly 65 trillion won consensus, largely because blended HBM pricing growth was revised down from an expected 50% quarter-over-quarter gain to under 29% 3. SK Hynix reports full second-quarter results on July 29, and that print will show whether the timing gap was a one-quarter blip or a deeper crack in its pricing power [via SK hynix Newsroom cadence, 7].
Samsung's record quarter tells the mirror-image story. After trailing both SK Hynix and Micron in HBM for most of the AI buildout, Samsung leaned on being first to mass-produce sixth-generation HBM4, starting shipments in the first half of 2026 while rivals were still finishing qualification 5. That head start, combined with improved yields and rising 4nm foundry utilization for HBM4's logic base die, helped memory-driven profit more than double sequentially from Samsung's own first-quarter operating profit of 57.2 trillion won 4. The company has told investors it expects HBM sales to more than triple in 2026 versus 2025 5. That is a sharp reversal from early 2026, when Samsung was widely described as the laggard in HBM qualification. It also validates the strategic bet Samsung made to prioritize an early, if narrower, HBM4 ramp over squeezing more legacy DRAM and HBM3E profit the way SK Hynix chose to for several quarters.
None of this means the leadership order has actually flipped. Supply-chain trackers reported in January that SK Hynix was set to supply roughly two-thirds of Nvidia's HBM4 orders for its next-generation Rubin-class platforms, with Samsung targeting early delivery of a smaller allocation to prove out its qualification 6. SK Hynix has held a majority of overall HBM revenue share since the category's inception, and its Q1 2026 results, a record 52.6 trillion won in revenue and a 72% operating margin, show how dominant that position remains in absolute terms 7. What changed in July is not who leads, but how confidently the market extrapolates that lead forward. A single quarter's contract-pricing drag was enough to erase most of SK Hynix's year-to-date valuation premium over Samsung, a sign that HBM investors are now underwriting execution quarter by quarter rather than assuming permanent separation.
The near-term setup favors patience over conviction. SK Hynix's July 29 earnings call and Samsung's July 30 divisional breakdown will show whether the HBM4 ramp genuinely lands in the third quarter as brokerages now expect, and whether Samsung's early-mover share gain is durable once SK Hynix's own HBM4 volumes clear qualification and move to market pricing. For portfolios exposed to the AI memory supercycle, the July repricing is a reminder that HBM leadership is now a moving target decided by contract structure and qualification timing, not brand loyalty. Investors chasing the next leg of the trade should watch blended ASP trends and HBM4 shipment cadence at both companies before assuming either has locked in its position for the rest of 2026.





View certificate