--- title: "SK hynix: Memory Rally Over; Are LTAs the Safe Harbor?" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/43019422.md" description: "SK hynix released its FY26 Q2 results (quarter ended Jun 2026) after the U.S. close, in the morning of Jul 29 (Beijing time). Key takeaways: 1) Core metrics: revenue of KRW 79 tn (+51% QoQ), below consensus (KRW 85 tn).Growth was driven by both DRAM and NAND, with memory price hikes the largest contributor. GPM was 83%, below street (84–85%)..." datetime: "2026-07-29T03:08:53.000Z" locales: - [en](https://longbridge.com/en/topics/43019422.md) - [zh-CN](https://longbridge.com/zh-CN/topics/43019422.md) - [zh-HK](https://longbridge.com/zh-HK/topics/43019422.md) author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)" --- # SK hynix: Memory Rally Over; Are LTAs the Safe Harbor? $SK Hynix(SKHY.US) (Beijing time) on the morning of Jul 29, 2026, posted Q2 FY2026 results (to Jun 2026) after the U.S. close. Key takeaways are below. **1) Headline numbers**: **Revenue came in at KRW 79 tn (+51% QoQ), missing the street at KRW 85 tn**. Growth was driven by both DRAM and NAND, with pricing uplift the largest contributor. **GPM was 83%, below the 84–85% consensus,** helped by higher DRAM/NAND prices. With GPM already above 80%, further upside looks limited. **2) Core OP**: $SK Hynix(SKHY.US) **reported core OP of KRW 60.5 tn (+61% QoQ)**, below the market at KRW 67 tn. The increase reflected strong revenue growth and margin expansion. Core Opex (R&D + SGA) was KRW 5.4 tn, up 34% QoQ. On surging revenue, the core Opex ratio fell to 7%. **3) Biz. mix: The company remains focused on memory, with essentially all revenue from DRAM and NAND**. **a) DRAM**: Revenue was KRW 57.9 tn, up 42% QoQ. Shipments rose ~8% QoQ (high single-digit), and ASPs rose ~30% QoQ, well below the traditional DRAM +55% QoQ move in Q2. This reflects prior capacity shifts to HBM, reducing exposure to legacy DRAM. **HBM update:** HBM4 started shipping in Q2 with utilization to ramp in 2H. HBM4E was sampled in 1H. **b) NAND**: $SKHY 2X Long ETF(SKHX.US) **Revenue was KRW 21.4 tn, up 85% QoQ**. Shipments rose 15% QoQ and ASPs were up 50%+, in line with market price gains. **Dolphin Research view: Earnings are rear-view; durability is the new question** Against the backdrop of sharp memory price hikes, this print was underwhelming, with both revenue and GPM below expectations. The miss is notable given the pricing tailwinds that the market had penciled in. **The shortfall largely came from DRAM** (over 70% of total). DRAM revenue rose 42% QoQ, with shipments up high single digits (~8%) and ASPs up ~30% QoQ. Versus the 50%+ QoQ move in market DRAM pricing for Q2, SK Hynix’s ASP uplift lagged visibly. **Management had shifted more capacity into HBM, leaving lower exposure to legacy DRAM, so recent price hikes provided less upside than the market expected**. **FY guide (in line)**: 1) 2026 DRAM bit shipments +~25% YoY; NAND bit shipments +~18% YoY. 2) 2026 capex at KRW 40–45 tn, roughly USD ~30 bn, vs. market at KRW 42–45 tn. **The 'post-close -10%' drop reflected the miss. The subsequent rebound** followed the first official confirmation in slides that LTAs include prepayments and related financial mechanisms. However, no amounts were disclosed (Micron has disclosed prepayments and RPO). **LTA disclosure this quarter**: Ten LTAs have been signed, and they include deposits/guarantees (first official confirmation), with core customers and typical five-year terms. Terms vary by customer and product. Beyond this print, the market is focused on the following: **1) AI capex sustainability** **Memory sits squarely in the AI capex stack, with CSPs the ultimate payors**. Street estimates for 2026 capex by the Big 5 CSPs (Google, Meta, Microsoft, Amazon, Oracle) have risen to over USD 800 bn (+82% YoY). After three years of outsized growth, investors are questioning durability. **Google last week raised its FY capex to USD 195–205 bn (vs. 180–190 bn prior)**. The ~USD 10 bn bump was in line with expectations, and is seen more as a pass-through of memory price inflation than a true step-up in AI demand. Meta’s compute leasing and the K3 model launch have amplified concerns about future token price declines and model economics. **TickerTrends shows Anthropic ARR still growing but with a flattening slope**. With leading model ARR growth decelerating, the market is sharpening pencils on the economics. That raises the risk that Big Tech will turn more cautious on AI capex. **2) Memory price peak is debated, but a slower slope is consensus** Massive memory price hikes are facing pushback from downstream customers (handsets and CSPs). **Vendors like OPPO and Vivo have begun rejecting price increases, and with GPM now above 80%, profit concentration at memory makers looks unhealthy and underscores cyclicality**. **On the cycle:** **a) Some expect prices to roll over in 2H27**: Memory majors are aggressively lifting capex. **Dolphin Research estimates 2026 memory capex by the Big 3 will exceed USD 95 bn (+30% YoY)**. Faster investment could pull forward capacity and bring an earlier downturn. **b) The more constructive view is that LTAs can soften and delay the downcycle**: **The market expects SK Hynix to lock in 60–70% of volume/price over five years**. Five-year terms often follow a 2+3/3+2 structure, with the first 2–3 years fixed and the back half partially floating. LTAs improve earnings visibility but do not fully lock capacity. If management can provide more incremental LTA detail (re-signed enhanced LTAs, broader capacity coverage, financial guarantees), it would bolster confidence and offer better downside protection. **More than the peak debate, a slowing second-derivative is consensus**. TrendForce shows ASPs up 90%+ QoQ in Q1, 50%+ in Q2, but only ~15% expected in Q3. That 90% → 50% → 15% deceleration intensifies concerns about an approaching downcycle. **At a current mkt cap of USD 948.7 bn, the stock trades at ~4.3x 2027E core net income** (assumes two-year revenue CAGR +129%, GPM 86%, tax rate 25%, USD/KRW = 1,400). While a 2027 PE near the low end of history looks optically cheap, peak-cycle multiples are fragile given cyclicality. **Most houses expect memory prices to start falling in 2H27**. The key question is whether earnings can hold up through the downcycle, which takes us back to LTA coverage and terms. **Among memory majors, Micron has a U.S. home-court edge and is most assertive on LTAs, supporting a higher multiple (6x PE) vs. SK Hynix (4x PE)**. **i) Pricing bands**: Micron has set price floors and ceilings to protect margins. The market believes SK Hynix uses fixed prices in early years, then flex in later years. **ii) Financial guarantees**: Micron disclosed USD 22 bn in financial commitments and customer RPOs. **SK Hynix said LTAs include guarantees/deposits, which lifted confidence post-close (-10% to green), but disclosed no amounts on the call, and shares faded again**. **Bottom line**: In a fragile AI setup, an underwhelming print dents confidence. With a likely downcycle from 2H27, the current ~4x PE bakes in that view. Even if ~4x is near trough history, it is not a free pass, as earlier price declines would cut earnings and mechanically lift PE. Better LTA disclosure would help de-risk the path. **The second derivative on memory pricing has rolled over**. Investors are prioritizing LTA depth to buffer cyclicality, and while SK Hynix confirmed guarantees in LTAs this quarter, it withheld specifics on amounts and terms. **Compared with Micron’s explicit prepayment and RPO disclosures, SK Hynix’s LTA detail still looks light. That makes it hard to close the valuation gap**. Note: As full financials are not yet available, this note is based on the presentation and call. We will update after the full release. Charts for SK Hynix’s results follow: **Dolphin Research history on SK Hynix:** Jul 10, 2026 First Take: '[SK Hynix heads to the U.S., seeking a fair valuation?](https://longportapp.cn/zh-CN/topics/42574773)' Risk disclosure and disclaimer: [Dolphin Research Disclaimer and General Disclosure](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [SKHY.US](https://longbridge.com/en/quote/SKHY.US.md) - [SKHX.US](https://longbridge.com/en/quote/SKHX.US.md) ## Comments (10) - **北畠显家 · 2026-07-30T07:18:27.000Z**: The investment income included, didn't mention selling Kioxia common stock at the high point - **汗舞大地 · 2026-07-30T02:39:46.000Z**: The profits from the past three years were turned into a negative by memory stocks this month, which is terrifying. - **左储右光 · 2026-07-29T08:25:09.000Z · 👍 1**: Their earnings hit a record high, yet Wall Street insists it missed expectations. After all, whenever they short, they claim it missed the mark. - **Dolphin Research** (2026-07-29T08:43:50.000Z): Mainly, the stock price has already priced in the expectations. - **real88 · 2026-07-29T03:27:29.000Z · 👍 5**: Whose expectation is this exactly? It's malicious. - **未来战士** (2026-07-29T07:21:35.000Z): Of course it's Wall Street. This sharp drop is clearly the US shorting South Korea. - **Dolphin Research** (2026-07-29T08:43:22.000Z): Consensus among major brokerages 😂 - **serenity! · 2026-07-29T03:23:14.000Z · 👍 1**: Time will tell. - **W时空 · 2026-07-29T03:21:47.000Z · 👍 8**: When others quote high prices, you say it's a cycle. When long-term agreements lower prices slightly, you complain about earning less. Hahahaha 😁 Truly, with just a movement of the lips, you manage to profit from both rising and falling markets. - **君悦 · 2026-07-29T03:21:13.000Z**: The momentum from the early session was smashed down again.