3 wars. 3 crashes. #4 is coming

When SK Hynix reports Q3 results on October 28, the numbers will almost certainly be extraordinary. Market consensus projects consolidated revenue of ₩98.23 trillion and operating profit soaring 576% year over year. That translates into a 78.4% operating margin, with Q4 profit expected to climb further still. And yet SKHY shares, even after jumping recently in Seoul, sit roughly 57% below their late-June peak.

That gap is the debate. Either the market is pricing in something analysts are not, or analysts are right and one of the widest valuation discounts in the semiconductor sector is about to correct violently. Professional investors will be arguing both sides before the Seoul webcast archive is even uploaded.

Why the Bulls Think the Market Is Wrong

Analysts carry a Strong Buy consensus on SKHY, with an average price target of $254.70, implying about 60% upside from current levels. Barclays analyst Simon Coles expects SK Hynix to return about 15% of its market value to shareholders, and sees the Q3 report as the event that forces the market to price that commitment in.

On estimated 2026 earnings, SK Hynix trades at roughly 5.3 times, compared to a sector median of about 28.7. The bull case is simple: that multiple makes sense only if profits collapse. The structural argument against collapse is the long-term contract book. Speaking on the Q2 earnings call, management said it had completed long-term supply agreements with around 10 customers, with durations typically around five years. A key additional driver this quarter is HBM4, but investors should be careful about assuming meaningful HBM4 revenue is already embedded in Q3: Nvidia’s Vera Rubin platform uses HBM4, and Nvidia and SK hynix have publicly described a multiyear technology partnership around that platform, yet the timing and scale of commercial HBM4 shipments that would move quarterly financials are not something the company has clearly pinned to Q3 in its disclosures.

The long-term contract dynamic is not unique to SK Hynix. Micron’s record quarter surfaced the same structural question about whether multi-year AI memory commitments have permanently changed the cyclical math for DRAM suppliers.

The special dividend is the other card. Bloomberg reported SK Hynix plans to return roughly 80% of gains from selling down its Kioxia stake through a special dividend of about ₩16 trillion, or $12 billion, citing a local newspaper report. The company has also said it considers its share price undervalued relative to intrinsic value and has committed to acquiring and cancelling approximately ₩40 trillion in treasury shares.

Why Bears Are Not Moving

Investors are focused on slowing sequential growth: the question is no longer whether earnings beat expectations, but whether depressed valuations have already priced in the worst, with Q3 revenue growth decelerating to roughly 24% quarter over quarter after 60% and 51% in the preceding two periods.

The China threat reinforces the caution. CXMT has said it plans next-generation DDR5 server memory based on a denser 4F² DRAM architecture, raising concerns it could eventually take global market share from Samsung, SK Hynix, and Micron. Citi analysts have argued the manufacturing remains several generations behind industry leaders, with mass production of the new architecture unlikely before 2028. Bears use that two-year window not as reassurance but as a countdown.

Profits and margins are still rising, but as sequential growth normalizes, an earnings beat may carry less weight than management’s 2027-2028 pricing and capacity outlook. Tonight’s guidance language will matter as much as the headline number.

Guidance language is carrying more weight across the entire AI hardware supply chain right now. Nvidia’s own forward volume forecast has become the anchor number investment committees use to stress-test 2027 memory demand assumptions.

What Investors Are Missing

The framing of “peak memory” misses a structural shift in how memory is procured. As memory becomes a critical bottleneck for hyperscalers building AI infrastructure, long-term supply agreements are becoming more common, and SK hynix has been explicit that it is pushing multi-year agreements with around 10 key customers, typically around five years. That changes the cyclical math. Prior memory downturns were amplified by customers who could simply stop ordering. Customers now locked into multi-year commitments, often with mechanisms like deposits, cannot.

The bears pricing in 2028 oversupply are extrapolating a market structure that no longer exists.

Stocks to Watch

• SKHY: The direct read. Tonight’s margin outcome and management’s guidance tone on 2027 pricing are the two numbers that matter more than the dividend size.
• MU (Micron): Memory chip stocks including Micron have already been rising in premarket trading as Bank of America rethinks what comes next for the sector. A strong SK Hynix report confirms the AI memory cycle has not peaked.
• NVDA (Nvidia): Vera Rubin’s HBM4 supply chain begins here. SK Hynix’s guidance on HBM4 ramp and pricing is a direct read on Nvidia’s next platform economics.
• SNDK / WDC (SanDisk / Western Digital): The NAND side of tonight’s report matters for both. SK Hynix has said it decided to abandon a potential Solidigm IPO, and that choice raises fresh questions about how it will finance future NAND capacity expansion in the U.S., with potential joint venture structures that could reshape competitive dynamics.