Mirae Asset Securities logo

Growth outlook remains intact

Maintain Buy and TP of W2,800,000

We maintain our Buy rating on SK Hynix with a target price of W2,800,000. Despite solid industry conditions, the stock has pulled back sharply relative to its fundamentals amid heightened macro uncertainty. At the current share price, 12-month forward P/B and P/E stand at only 2.1x and 4.1x, respectively. We see few signs of deterioration in either industry conditions or the company’s fundamentals.

Against this backdrop, the key question is whether growth can be sustained. On Aug. 19, the company announced a W40tr share buyback and cancellation program, in line with our estimate, while also revising its shareholder return framework to at least 50% of 2025-27 cumulative free cash flow (from up to 50% previously). We believe this underscores management’s confidence in the visibility of future cash flow generation.

We maintain our operating profit forecasts of W267tr for 2026 and W389tr for 2027. We expect memory supply/demand conditions to remain tight through 2028, with DRAM ASP likely to grow 188% in 2026 and a further 23% in 2027. With more than half of revenue now covered by long-term agreements, earnings volatility is also declining. Despite moderating growth, the company’s elevated ROE supports a higher multiple.

We visited seven semiconductor companies in China from Aug. 17 to 22. HBM was by far the most prominent topic among Chinese accelerator makers. Even companies that currently use GDDR plan to adopt HBM in their next-generation products, while one company said it intended to use proceeds from a planned Hong Kong SAR listing partly to strengthen its access to HBM supply. All in all, the growing adoption of China-made accelerators points to strong HBM demand growth in the country.

However, China’s ability to meet its HBM needs domestically appears lower than expected. Based on the assessments of local firms, China remains roughly two years away from achieving self-sufficiency in HBM3-class products. Meanwhile, the more capacity CXMT reallocates to HBM to address the shortfall, the less conventional DRAM it will produce (with HBM consuming nearly four times as much capacity). We therefore expect limited China-driven disruption to DRAM supply/demand dynamics. Instead, this dynamic should put upward pressure on DRAM prices.

At Hot Chips 2026, SK Hynix said that its 16-high HBM is undergoing customer qualification and that it is developing hybrid-bonding technology for 20-high and above. By eliminating the gaps created by micro-bumps, hybrid bonding allows the core dies in a 20-high stack to be up to 24% thicker while reducing thermal resistance by approximately 35%. A narrower bonding pitch can also support greater bandwidth, with the benefits increasing as stack heights rise. However, SK Hynix is considering adoption of hybrid bonding only for post-HBM4E generations. Maintaining the integrity of the vast number of bonding interfaces―and thus achieving acceptable yields―remains an industry-wide challenge. As HBM stack height rises, more capacity is consumed, making supply expansion more difficult. Against this backdrop, we expect SK Hynix’s lead in high-stack technology to eventually be fully reflected in its valuation.