Hook
SK Hynix just posted the fattest quarterly profit in its history. Net income hit a record high, driven by the AI boom’s insatiable hunger for its HBM3E memory chips. The numbers are jaw-dropping: operating profit surpassed 5 trillion Korean won for the first time. Panic sells. I just watch.
But here’s the kicker: the stock dropped. The market yawned and sold. “Revenue beat, but earnings missed expectations,” the analysts whispered. The chart lies. The volume speaks. Something deeper is happening beneath the surface.
Context
Sk Hynix is the world’s second-largest memory chip maker, but in the booming HBM (High Bandwidth Memory) market, it’s the undisputed king—holding roughly 50% share, ahead of Samsung and Micron. HBM is the turbocharger for AI GPUs like NVIDIA’s H100 and B200. Every one of those GPUs needs up to eight stacks of HBM3E memory, tightly integrated via advanced packaging (TSV, micro-bumps, hybrid bonding) and co-designed with logic foundries like TSMC.
For over a year, demand from hyperscalers (Amazon, Microsoft, Google, Meta) has outstripped supply. SK Hynix has ridden this wave to a dramatic turnaround: from a loss-making 2023 to a record-breaking 2024. The company’s MR-MUF packaging technology gave it a crucial yield advantage over its rivals, allowing it to secure exclusive or near-exclusive deals with NVIDIA.
Core
But here’s the nuance the headlines missed. The profit was record-high, but the rate of profit growth slowed.
Sequentially, operating profit grew by about 15-18%. Market whispers had expected 20-25% growth. That gap—a few percentage points—triggered the sell-off. Alpha doesn’t wait for permission. The market is pricing in two fears that the naive bulls ignore:
Fear #1: The Capex Trap.
Sk Hynix is spending like a drunken sailor on shore leave. Its 2024 capital expenditure is estimated at over 12 trillion Korean won. That’s a record. They are building new HBM-dedicated lines (M15X in Cheongju), converting existing DRAM factories, and planning a mega-cluster in Yongin, with total investment exceeding 120 trillion won over the next decade.
What the income statement giveth, the balance sheet taketh away.
The result? Free cash flow is negative. Estimated operating cash flow for the first half of 2024 is around 8-9 trillion won, but capex is 12 trillion. That’s a 3-4 trillion won hole. They are spending every dime they earn—and more—just to stay in the game.
Depreciation is coming for those gross margins.
These massive new factories won’t just produce chips; they’ll produce decades of depreciation charges. For every trillion won spent on equipment, expect annual depreciation of roughly 100-150 billion won for 7-10 years. That alone could shave 5-8 percentage points off gross margins over the next two years.
Fear #2: The Customer Concentration Sword of Damocles.
NVIDIA is not just a customer; they are a lifeline. Estimates suggest that NVIDIA alone accounts for over 60-70% of SK Hynix’s total HBM revenue. That is a terrifying amount of single-client risk.
What happens if NVIDIA decides to dual-source more aggressively? Or worse, if they embrace Samsung’s HBM4 and reduce orders? A 20% drop in HBM orders from NVIDIA could wipe out the entire incremental profit growth that the market hoped for.
The market is not stupid. It knows that a monopoly on today’s technology does not guarantee tomorrow’s revenue.
Contrarian Angle
You will hear analysts say: “But demand is structural! AI capex is only going up! This is a super-cycle!”
They are right about the demand. They are wrong about the valuation.
The core issue is that the market is trying to value SK Hynix like a growth stock (think NVIDIA: expanding margins, durable moat, high free cash flow yield) when it is still, at its heart, a capital-intensive cyclical commodity manufacturer.
Look at the numbers:
- ROIC (Return on Invested Capital): Estimated at 12-18%. This is good, but compare it to NVIDIA’s 50%+ ROIC. Ski Hynix needs to invest a mountain of capital to generate those profits.
- WACC (Weighted Average Cost of Capital): Around 8-10%. The spread is positive, but razor-thin. A price war (from Samsung or Micron) or a demand hiccup could easily flip it negative.
Market expectations have morphed. In the past, memory cycle peaks were priced at 5-7x earnings. Now, due to the AI narrative, SK Hynix trades at 10-12x forward earnings. That multiple is earned only if the growth is both high and sustainable. Last quarter’s “miss”—even on record profit—signals that the sustainability is in doubt.
The volume speaks: big money rotated out after the earnings call. Panic sells. I just watch. The chart lies. The volume speaks.
There’s also an unspoken geopolitical layer. SK Hynix operates a massive DRAM fab in Wuxi, China. The US export controls on advanced chip equipment have put that factory in a vulnerable position. Upgrading it to EUV level is tough. Losing that fab to stricter rules could slash 10-15% of their DRAM capacity. The market’s “disappointment” partly reflects the quiet fear that the China tail risk is rising.
Takeaway
Alpha doesn’t wait for permission. The “record profit but stock down” paradox is a loud signal. It says: the easy money in the HBM cycle has been made. From here, the stock moves on execution, margin protection, and free cash flow generation—not just revenue growth.
Will SK Hynix manage the capex-to-cash conversion? Will NVIDIA stay faithful? Or will the market re-rate it back to a 7x PE cyclical, ignoring the AI story?
The next two quarterly reports will answer that. I’m watching the volume, not the headlines.