The continued expansion of AI inference demand is propelling the global memory market into a new structural upcycle, with SK Hynix emerging as a primary beneficiary. Analysts note that AI inference demand is driving a surge in KV cache storage needs, leading to a significant jump in the average selling prices (ASPs) of both DRAM and NAND in the first quarter. Concurrently, memory suppliers are shifting from quarterly contracts to long-term agreements spanning 3 to 5 years. This transition is expected to significantly enhance profit visibility, reduce cyclical volatility, and lead to a structural improvement in the quality of SK Hynix's earnings. The target price for SK Hynix has been raised, with a "Buy" rating maintained. It is projected that the company's operating profit for the first quarter of 2026 will reach 39.1 trillion won, representing a 104% increase from the previous quarter and a 426% rise year-over-year, substantially exceeding the market consensus of 35.1 trillion won. Based on the current share price, this implies a potential upside of approximately 64.6%.
First-quarter earnings are anticipated to surpass expectations, driven by a substantial leap in pricing power. For the first quarter of 2026, SK Hynix's DRAM ASP is forecast to increase by approximately 63% compared to the previous quarter, while the NAND ASP is expected to rise by about 70%. The core driver is the explosive growth in KV cache storage demand within AI inference scenarios. Regarding shipment volumes, DRAM shipments are estimated to have decreased by around 6% quarter-over-quarter, and NAND shipments by approximately 2%. However, the substantial increase in ASPs is more than sufficient to offset this seasonal decline, driving total revenue to an estimated 52 trillion won, a quarterly increase of about 58%. From a profitability perspective, the operating profit margin for the DRAM business is projected at 80% for the quarter, and 61% for the NAND business. The overall operating profit margin is expected to be around 75%, a significant improvement from the 58% recorded in the fourth quarter of 2025.
Concerns regarding macroeconomic uncertainties and potential demand suppression from AI algorithm optimizations are viewed as overstated. Underlying demand for 64GB server DDR5 DRAM and HBM remains on a steady growth trajectory, unaffected by recent market fluctuations. Advancements in AI algorithms are expected to further increase demand for server DDR5 and HBM, rather than diminish it. On the pricing front, prices for 64GB server DDR5 DRAM are anticipated to remain strong at least until the second half of 2026, with HBM demand also set for continued expansion. As the widespread adoption of AI agents leads to growth in token generation, CPU memory demand for both AI training and inference scenarios is expected to receive a concurrent boost.
The duration of the global memory supply shortage may extend beyond current market expectations. The proliferation of multi-year binding long-term agreements is likely to intensify this trend. Memory suppliers are upgrading traditional quarterly contracts to long-term agreements lasting 3 to 5 years, although specific details are still to be confirmed. This shift indicates the memory market's evolution from a traditional commodity model towards a highly customized, client-specific model—a trend comparable to the development path of the wafer foundry business. HBM business valuation is already being benchmarked against relevant multiples. Based on this rationale, the 1.7 million won target price for SK Hynix is affirmed.
Comments