To The Moon

SK Hynix reported quarterly revenue of 104.6 trillion won and operating profit of 83.3 trillion won, with an operating margin approaching 80%. Yet net income plunged 58%. Despite the sharp decline in net income, the company’s U.S.-listed shares surged 13.3% on Oct. 28, while its South Korean shares rose 10.89%. The rally extended across the semiconductor sector, lifting Samsung Electronics 9.01%, Micron Technology 5.86% and SanDisk 7.66%. The market’s reaction reflected confidence in SK Hynix’s underlying operating performance rather than its headline net income. Strong shipment guidance, rising memory prices and progress in high-bandwidth memory (HBM) reinforced the company’s position in the AI supply chain. But the results also raise broader questions. How sustainable are its extraordinary margins? Can long-term supply agreements reduce the risks of the memory cycle? And as AI workloads evolve, could NAND flash become a more important part of the industry’s growth story? Strong Demand Drives the Rally Before the earnings release, investors had grown increasingly concerned about the sustainability of AI infrastructure spending. Questions about the returns on massive investments in data centers and computing capacity had begun to weigh on the semiconductor sector. SK Hynix’s latest outlook offered a more encouraging picture. The company expects fourth-quarter DRAM shipments to rise roughly 20% to 25% sequentially. That would indicate that the market’s strength is being driven not only by higher prices but also by growing shipment volumes—a healthier combination than price increases alone. Production execution will be equally important. The ramp-up at the M15X facility and the transition to advanced 1c DRAM manufacturing appear to be progressing smoothly. Meanwhile, HBM4 has entered volume shipments, with yields reportedly ahead of expectations. These developments matter because the memory industry’s profitability depends on more than demand. Manufacturers must also deliver sufficient output, improve yields and transition to new technologies without allowing production costs to erode margins. SK Hynix’s position in HBM gives it exposure to the growing need for memory in AI accelerators, while its conventional DRAM and NAND businesses provide additional sources of revenue. Strength across these product categories could make the company less dependent on any single segment of the AI hardware market. The company also announced a shareholder-return package worth roughly 100 trillion won, including more than 30 trillion won in dividends and over 60 trillion won in share repurchases and cancellations. It also ended its evaluation of a potential Solidigm IPO. Together, these moves reinforce confidence in management’s expectations for future cash generation. The key question is whether strong demand and pricing can translate into sustained profitability as the industry invests to meet customers’ needs. 2.The 32.6 Trillion-Won Earnings Gap Despite operating profit of 83.3 trillion won, SK Hynix reported net income of just 39.8 trillion won. The gap was largely attributed to two categories of non-operating losses. The first was approximately 11.6 trillion won in foreign-exchange losses, reportedly associated with the Korean won’s appreciation. Currency movements can affect reported earnings when a company holds assets, liabilities or cash balances denominated in foreign currencies. The second was around 22.9 trillion won in investment valuation losses as semiconductor-related equities declined. Together, the two items represented a substantial drag on the company’s bottom line. The distinction between operating profit and net income is important. Operating profit provides a clearer view of the performance of the company’s core business, while net income also reflects financing costs, taxes and other items outside day-to-day operations. Some investment valuation losses may be non-cash when recognized and could reverse if the underlying assets recover. However, that outcome is not guaranteed, and foreign-exchange movements can have real economic consequences depending on the company’s exposures and hedging arrangements. The losses therefore should not simply be dismissed. Instead, they need to be assessed separately from the performance of the memory business. If DRAM and NAND prices remain firm, shipment volumes increase and production yields improve, SK Hynix could continue generating strong operating profits even when non-operating items create volatility in net income. For investors, the central issue is the durability of those profits. An operating margin approaching 80% is extraordinary, but the memory industry is cyclical. Whether such profitability can persist will depend on pricing power, manufacturing efficiency and the pace at which competitors add capacity. 3.Long-Term Contracts: Less Risk, Potentially Less Upside SK Hynix is also seeking greater visibility into future demand through long-term supply agreements. The company has reportedly signed five-year agreements with 14 core customers, including major cloud-service providers, securing commitments for more than 50% of its total capacity in 2027 and 2028. The agreements reportedly include deposits. The strategic logic is straightforward. AI infrastructure requires substantial investment, but the pace of adoption and the commercial returns from AI applications remain uncertain. If customers slow their spending, memory manufacturers without firm commitments could face weaker sales and excess inventory. Long-term agreements can reduce that uncertainty by giving manufacturers a clearer view of future demand. They may also help customers secure access to critical memory supplies in a market where advanced products can be difficult to obtain quickly. The trade-off is pricing flexibility. If memory prices rise sharply in 2027, SK Hynix’s ability to capture the full upside will depend on how contract prices are determined. Agreements that allow prices to adjust with market conditions may preserve some upside, while tighter pricing limits could leave the company earning less than it might in the spot market. In effect, SK Hynix is trading some potential upside for greater demand visibility. That may be a sensible strategy in a capital-intensive industry where overestimating demand can have costly consequences. More predictable orders could also make production planning and capacity investment easier. Rather than responding entirely to short-term price movements, the company can align part of its output with contracted customer requirements. Still, long-term agreements cannot eliminate the memory cycle. Demand can change, technology can shift and new capacity can eventually put pressure on prices. The value of these contracts will depend not only on their volume but also on their pricing terms and flexibility. 4.Beyond HBM: NAND’s AI Opportunity The AI memory opportunity is expanding beyond high-bandwidth memory. DRAM shipments reportedly rose around 10% sequentially, while average selling prices increased more than 20%. NAND shipments edged lower, but average selling prices also rose by more than 20%. The combination suggests that pricing strength is spreading across the broader memory market. For SK Hynix, this matters because conventional DRAM and NAND can contribute meaningfully to earnings alongside its HBM business. A further opportunity may emerge from changes in AI system architecture. AI agents and long-context models require increasingly large KV caches, which store information that models need to access during inference. As these workloads grow, keeping all relevant data in expensive, capacity-constrained HBM and DRAM becomes increasingly costly. NAND-based storage could help address this problem. SSDs cannot replace HBM or DRAM, which offer different performance characteristics, but they could complement faster memory in a tiered architecture. Data requiring immediate access would remain in high-speed memory, while other information could be stored on lower-cost media. Such an approach could help AI infrastructure balance speed, capacity and cost. If the amount of context processed by AI systems continues to expand, storage capacity may become an increasingly important part of the overall infrastructure equation. SK Hynix is reportedly developing SALT-KV software and CMM Hybrid technology, while exploring High Bandwidth Flash (HBF) as a potential bridge between HBM and SSDs. These technologies remain part of an evolving market, and their commercial impact will depend on performance, cost and customer adoption. Nevertheless, they illustrate why the long-term AI memory opportunity may extend beyond the fastest memory products. The gains in SanDisk and Micron alongside SK Hynix reflect this broader market interest. Investors are increasingly looking at the entire memory supply chain rather than treating HBM as the only potential beneficiary of AI infrastructure spending. 5.The Risks Ahead Despite the strong results, three questions will shape SK Hynix’s outlook. First, can DRAM shipments meet guidance? Fourth-quarter growth of 20% to 25% would support the view that demand remains robust and that the M15X ramp-up and advanced 1c manufacturing are progressing as expected. A shortfall could reflect production constraints, customer timing or weaker demand. Second, why is the profitability gap between HBM and conventional DRAM narrowing? If conventional DRAM margins are rising as prices strengthen, SK Hynix could benefit from a broader-based memory upcycle. If the narrowing instead reflects pressure on HBM profitability, investors would need to examine whether competition or customer pricing is affecting returns. Third, will higher investment eventually trigger another supply glut? Management argues that capital spending does not immediately translate into additional output: factories can be built ahead of demand, while equipment installation and production ramps are staged over time. Even so, the industry has repeatedly demonstrated how quickly favorable economics can encourage capacity expansion. If new supply arrives faster than demand grows, pricing power could weaken. Chinese manufacturers such as ChangXin Memory Technologies are another factor to monitor as their manufacturing capabilities develop. If capacity expands just as AI spending slows, the market could face renewed pricing pressure by 2028. For now, supply-demand conditions remain favorable, but current margins should not be assumed to last indefinitely. SK Hynix’s results highlight the strength of the current memory upcycle. Rising prices, growing AI demand and progress in HBM are supporting its operating performance, while the expansion of AI workloads could create opportunities across conventional DRAM and NAND. Long-term supply agreements may provide greater visibility into future demand, although they also introduce trade-offs if market prices rise beyond contracted levels. Meanwhile, the risk of overcapacity remains a central concern for an industry accustomed to sharp swings in profitability. The challenge is to capture the AI boom without investing too aggressively ahead of the next downturn. SK Hynix’s ability to maintain its technological edge, manage capacity and secure sustainable demand will determine whether the current cycle develops into a more durable phase of growth—or another peak in the memory market. $SK海力士(SKHY)$