The KOSPI opened up over 2% today. Samsung and SK Hynix led the charge. The headlines write themselves: AI narrative intact, semiconductor cycle bottom in, HBM demand is the new oil. Let me dissect this narrative with the cold eye of someone who has lived through five market cycles, audited enough smart contracts to spot a flaw in governance, and watched more leveraged portfolios liquidate from overconfidence than from bear markets.
Hook: The Price Action Anomaly
Here is the data point that matters: the KOSPI rally was not a broad-based tech bounce. It was concentrated in two names — Samsung and SK Hynix. The rest of the index lagged. This is not a sign of a healthy market rotation. It is a symptom of a market that has run out of ideas and is doubling down on a single narrative: AI hardware scarcity. When the rally is so narrow, it becomes structurally fragile. The market is pricing in perfect execution from two companies that face asymmetric risks.
Context: The Machinery of Memory
Samsung and SK Hynix are not just any chip makers. They are the gatekeepers of global memory supply — DRAM and NAND for everything from your smartphone to the data center. But the game changed with HBM (High Bandwidth Memory). HBM is not a commodity like DDR5. It is a bespoke, vertically integrated product that requires advanced packaging (TSV, micro-bumps, MR-MUF or TC-NCF) and tight coupling with logic foundries like TSMC. The barrier to entry is high. The capital required is staggering.
In 2024, SK Hynix owns roughly 50% of the HBM market. Samsung is close behind at 40%. Together, they form a duopoly that commands pricing power with clients like NVIDIA and AMD. The problem? Both companies are pouring "tens of trillions of won" into capacity expansion. This is not optional. It is a strategic bet that AI demand compounds faster than the capex cycle. The market is paying for that bet today, but the bill comes due in 2025–2026 when depreciation starts to eat margins.
Core: Order Flow Analysis — The Mechanical Reality
Let me state my bias upfront: I trade the structure, not the story. The order flow in Korean memory stocks tells me that the smart money is hedging, not chasing.
The HBM Monopoly Myth
The prevailing narrative is that HBM is a structural monopoly locked in by NVIDIA's design wins. But order flow analysis suggests something different. Look at the options market on SK Hynix: put skew has been rising even as the stock rallies. This is a classic “poor man’s hedge” — institutional investors are buying the stock for the narrative but hedging against the risk that HBM margins compress due to faster-than-expected commoditization.
Why? Because HBM manufacturing is not a one-time design winner. It is a tournament. The first mover advantage (SK Hynix) is real, but the time-to-market gap is shrinking. Samsung is investing aggressively to close the 3-6 month gap in HBM3E yield rates. If Samsung’s yield rate on HBM3E climbs to 70%+ by Q4 2024, the pricing premium on HBM will erode. The market is pricing in a scarcity premium. Scarcity is always temporary.
The Depreciation Trap
Here is the nuance the headlines miss. Both companies are in a high-capex cycle. Samsung’s semiconductor capital expenditure in 2024 is expected to exceed $50 billion. SK Hynix is spending over $15 billion. These are not growth investments that yield immediate returns. They are long-term gambles that lock in depreciation for 5-7 years.
Let me run the math. Assume Samsung’s HBM-related capital expenditure is $20 billion, depreciated over 5 years on a straight-line basis. That adds $4 billion in annual depreciation. If HBM revenue grows to $30 billion, the gross margin drag from depreciation alone is 13 points. To sustain attractive net margins, Samsung needs not just high HBM revenue, but high margins on a large scale. And scale itself is a risk: if HBM demand softens or if the technology shifts to HBM4 before the HBM3E capacity is fully utilized, you are left with stranded assets.
The Traditional Memory Anchor
Do not forget the traditional business. DRAM and NAND are still the majority of revenue for both companies. And those markets are in a slow recovery from a deep trough. Inventory levels for DDR4 and NAND are “healthy” (6-8 weeks), but pricing recovery has been modest — 10-20% from trough. That is not a boom. It is a normalization. The market is treating it as a boom because the baseline was so low. This is a statistical illusion, not a structural improvement.
When you combine HBM’s high but fragile margins with traditional memory’s cyclical but stable margins, you get a blended profile that is not as compelling as the stock price suggests.
Contrarian: The Blind Spots the Market Is Ignoring
The Korean Won (KRW) Signal
The KOSPI rally is happening against the backdrop of a weakening Korean won. The won has depreciated roughly 5-6% against the USD in 2024. For an export-heavy economy like Korea, a weaker currency is a double-edged sword. It boosts export competitiveness in the short term (revenues are in USD, costs are in KRW). But it also increases the cost of imported capital equipment and materials. Given that Samsung and SK Hynix are importing multi-billion dollar lithography machines from ASML and materials from Japan, the depreciation raises their capex bill.
More critically, a depreciating currency is often a signal of capital flight or macro instability. Foreign investors repatriating capital from Korean equities is a known pattern during global risk-off events. The KOSPI rally may already be partially driven by local retail investors chasing momentum, not by smart institutional accumulation. The second derivative — the change in foreign investment flow — is what you should watch, not the absolute KOSPI level.
The Geopolitical Fragility
Korea is a “fragile giant." It has dominant positions in memory, but its upstream dependencies (EUV from ASML, advanced chemicals from Japan, EDA from Synopsys/Cadence) create structural vulnerabilities. The US export controls on China have indirectly benefited Korean firms by restricting Chinese competitors. But the risk is asymmetric: if the US expands controls to include advanced packaging equipment that is critical for HBM, or if China retaliates by restricting rare earth exports (gallium, germanium), the supply chain for Korean MEMORY goes from “tight” to “broken." This is not priced in.
The implied probability of a black swan event — like a full-scale semiconductor trade war between the US and China that collateralizes Korea — is near zero in the option market. That is the blind spot. Trust is a variable I solve for, never assume. The market is assuming geopolitical stability. History suggests that is a dangerous assumption.
The Customer Concentration Risk
SK Hynix generates over 80% of its HBM revenue from its top two customers (NVIDIA and AMD). Samsung has a slightly more diversified customer base, but still heavily tilted toward the same pool. If NVIDIA decides to dual-source aggressively (as it historically has done) to reduce dependency on SK Hynix, or if NVIDIA starts developing in-house HBM controller IP that reduces the brand moat, the pricing power of both Korean firms diminishes.
This is not a hypothetical. NVIDIA has already expressed interest in supplier diversification for HBM4. The market is treating this as a “growth insurance” for smaller players like Micron. But for SK Hynix, it is a risk to its premium pricing. If NVIDIA’s purchasing power demands a 10% price cut on HBM3E in 2025, SK Hynix’s margin profile changes dramatically.
The Cycle of the Narrative
The market is pricing in a “structural super-cycle" for HBM. I am skeptical. Super-cycles are rare. What we are seeing is a classic technology S-curve: early adopters (data centers) paying high prices for scarce supply, followed by rapid capacity expansion, followed by oversupply, followed by margin compression. The “super-cycle” narrative is a marketing tool used to justify higher valuations. It is not a mechanical reality.
Speculation is gambling with a spreadsheet. The spreadsheet right now shows a P/E ratio for SK Hynix of 15-20x on forward earnings, and for Samsung of 8-10x. The Samsung multiple is low because the market accurately perceives its cyclicality. The SK Hynix multiple is higher because of its HBM premium. But if HBM becomes more commoditized, the PE ratio will compress. The gap between the two companies’ valuations is itself a signal that the market is overextrapolating recent HBM success.
Takeaway: The Trade, Not the Story
I am not saying the rally is wrong. I am saying the structure is weak. The next true test is the Q2 2024 earnings season. If Samsung and SK Hynix report earnings that confirm HBM revenue is translating into net profit margins above 40%, the rally has legs. But if earnings beat is driven entirely by revenue growth with flat or declining margins, the market will start discounting the capex cycle.
My personal positioning: I would not chase this rally. I would wait for the earnings confirmation. If the data confirms the structural case, there will be better entry points on pullbacks. If it does not, the downside is asymmetric. The market does not owe you an exit, only a price.
Final Note: Security is not a feature; it is the foundation. The security of this rally — its structural integrity — depends on HBM margins holding up against capacity expansion and commoditization. Watch the yield curve of HBM industry margins, not the revenue curve. That is where the truth lives.