Hook
70% operating margin. 93.9 trillion won net profit. A month later, stock down 40%. The numbers don't add up—not because the revenue is fake, but because the market is pricing a cliff that hasn't arrived yet, s heart.
Revenue hit 79.3 trillion won. Operating profit 60.54 trillion. These are historical highs for a memory manufacturer. Yet when SK Hynix reported on July 25, 2026, the stock initially fell 3% before recovering 0.19%. Then came the real move: a 40% drawdown over the following weeks. The divergence between reported earnings and price action is not irrational. It is a signal that the market sees a structural ceiling from which the company's margins will inevitably revert.
Context
SK Hynix is the world's second-largest DRAM manufacturer and the dominant supplier of High Bandwidth Memory (HBM) for AI accelerators. Its HBM3E—stacked DRAM dies connected via TSV and MR-MUF packaging—is the critical memory component driving NVIDIA's H100 and B200 GPUs. The company's competitive advantage stems not from lower wafer costs, but from a 6-to-12-month head start in HBM packaging yield and capacity. That lead is now being challenged by Samsung Electronics, which struggled with HBM3E yields in 2025, and by Micron's delayed but credible roadmap.
Core: Systematic Teardown of the Earnings Mirage
Revenue composition reveals over-concentration. AI-related products (HBM, enterprise SSDs) now account for an estimated 55-60% of revenue. The remainder comes from DDR5 and LPDDR5T for PC and mobile—both undergoing a temporary price recovery driven by supply discipline, not demand expansion. This concentration means any deceleration in AI CapEx directly translates into revenue risk.
Operating margin of 76% is a statistical outlier. For context, TSMC's best-ever operating margin is ~45%. NVIDIA's is ~55%. SK Hynix's margin is the highest in the semiconductor industry's history for a company of its scale. This is not the result of structural efficiency; it is the result of a temporary supply bottleneck in HBM packaging. The MR-MUF process, while superior to TC-NCF, remains a yield-limited art form. As Samsung and Micron close the packaging gap, HBM pricing will compress. A 50% margin compression would still leave SK Hynix with 38% operating margins—above historical averages, but far below current levels.
Capital expenditure is accelerating into a potential peak. The company announced expansion of its Cheongju M15X line and the long-term Yongin cluster. Based on historical depreciation schedules, the new capacity will begin hitting the balance sheet in 2027, just as the competitive landscape intensifies. The risk is not that demand vanishes, but that capital becomes trapped in capacity built at the cycle's zenith. SK Hynix holds 69.4 trillion won in net cash, offering a buffer, but the accounting standard of straight-line depreciation means that any revenue shortfall after 2028 will expose these fixed costs.
Customer concentration masks true bargaining power. NVIDIA alone consumes an estimated 35-40% of SK Hynix's HBM output. The remaining share goes to AMD, Intel, and cloud hyperscalers. NVIDIA is known for aggressively dual-sourcing: it has already pre-announced Samsung as a qualified HBM3E supplier for the B200 successor. The moment Samsung achieves parity, SK Hynix's pricing power will collapse from a premium to parity. The company's "multi-year contracts" are back-to-back agreements with volume commitments, but price renegotiation clauses are standard. Downward price risk is real.
Supply chain vulnerability is underestimated. HBM production depends on EUV lithography from ASML (100% dependency for 1β nm and below), high-purity chemicals from Japan (Shin-Etsu, JSR), and advanced packaging equipment from TEL and Disco. Any disruption—geopolitical or natural disaster—would halt HBM output. The US CHIPS Act includes provisions for onshoring HBM packaging, but building fabs in the US costs 30-40% more than in Korea, and the timeline extends to 2028. The company's cash hoard is partly a hedge against this risk, but it's also a reflection of the inability to convert cash into guaranteed capacity.
The market is right to sell the news. The 40% drop is not a panic. It is a repricing of future earnings from 76% margin to a normalized 35-40% margin over 18-24 months. The market is applying a terminal growth rate consistent with a cyclical memory supplier, not a hyper-growth AI proxy. The divergence between reported Q2 2026 profit and forward guidance is the largest in the company's history. s heart.
Contrarian: What the Bulls Got Right
First, the demand floor is real. Enterprise AI infrastructure investment is driven by cloud service providers whose budget cycles are multi-year. Microsoft, Google, and Meta have publicly stated that AI CapEx will grow 50%+ in 2027. This is not a speculative bubble—it is a long-term capital deployment that requires memory. Data center DRAM content per server is 4x higher for AI workloads than for traditional compute. Even if margin compression occurs, the total addressable market for HBM will grow from $22B in 2025 to $60B+ by 2030, according to industry estimates.
Second, SK Hynix's technology moat in packaging is real and defensible. The MR-MUF process is not easily replicated. Samsung's yield struggles in 2025 cost it an estimated 5-7% market share. SK Hynix's experience with TSV and micro-bump stacking gives it a multicycle learning advantage. The shift to HBM4 (expected 2027) will introduce hybrid bonding—a technique that increases stack height to 16+ layers. Early R&D investment means SK Hynix may retain a 12-month lead into the next generation.
Third, the net cash position (69.4 trillion won) provides operational flexibility. In a downturn, cash-rich memory players historically outspend rivals on R&D and capacity, then emerge stronger. If AI demand slows in 2028, SK Hynix can sustain losses longer than Samsung (which has a larger, lower-margin foundry business). The cash acts as a call option on the next cycle.
Takeaway: The Accountability Call
The SK Hynix story is not a fraud or a bubble—it is a classic case of mean reversion being mispriced. The market has priced in a 40% decline in share price not because the company is failing, but because the 76% margin is an artifact of a temporary supply-demand imbalance. The question every investor should ask: Do you believe HBM margins can stay above 50% past 2027? If not, the current valuation is a value trap masked by record cash flows. The metrics are real. The risk is structural. s heart.