The market is overreacting again.
SK Hynix’s target price dropped 33% overnight. Mirae Asset sliced it from 420,000 won to 280,000 won. Headlines screamed "memory slowdown." Retail panic.
But here’s the data that got buried: DRAM spot prices just broke all-time highs. And Mirae Asset, the same firm that cut the target, maintained their "Buy" rating.
I spent the last 48 hours dissecting their full report. Not the summary. The raw analysis. Seven dimensions. Code-level granularity.
Let’s cut through the noise.
Context: Why the market blinked
The sell-off narrative was clean: HBM hype peaked. Samsung is catching up. Huawei’s China localisation threatens supply. CXMT (ChangXin Memory Technologies) might IPO.
All true. All surface-level.
What the market missed was the structural shift: SK Hynix is no longer a cyclical memory stock. It has become an AI infrastructure compounder. But valuation frameworks haven’t caught up.
Core: The real architecture under the hood
Let’s talk about why the rating holds despite a 33% cut.
1. HBM3E supply lock
SK Hynix owns ~50% of the HBM3E market today. NVIDIA’s Blackwell ramp is HBM3E-hungry. Every B200 GPU needs ~144GB of stacked memory. There is no alternative source at scale in 2025.
This isn’t a commodity business. It’s a high-barrier, high-margin franchise. Gross margins are running at 40-50%. Comparable to TSMC. Not Micron.
2. Spot prices just broke
DRAM spot prices hit new 2024 highs. Not because of AI demand—AI runs on contract. The spot move signals that the analog memory market (phones, PCs, legacy servers) is also recovering.
This is a dual-engine cycle. HBM pulls the flywheel. Legacy supports the floor.
3. The capex paradox
Critics point to negative free cash flow. They’re right—on paper. But the capex is building HBM advanced packaging lines. These assets have 3-5 year payback periods at current ASPs.
Code that doesn’t break is boring. A balance sheet that only looks good in a bear market is cheap. SK Hynix’s capex is structural investment, not speculative burn.
Contrarian: What the big shorts are missing
The contrarian angle isn’t bullish. It’s structural skepticism applied to market fear.
Mirae Asset’s 33% cut is a valuation reset, not a fundamental downgrade.
They didn’t revise HBM revenue. They didn’t change NVIDIA supply estimates. They adjusted the multiple because the market is pricing in two valid risks:
- Customer concentration: NVIDIA is 30-50% of Hynix’s revenue. One bad contract negotiation could crater margins.
- China competition: CXMT and Huawei’s local mature-node DRAM will erode legacy pricing by 2027.
Both are real. But neither is new. The market just woke up to them.
The hidden signal: Mirae Asset explicitly flagged "early shareholder returns" as a factor to watch. Translation: management knows the capex load is visible. They’re preparing a capital allocation pivot.
That’s not a distress signal. That’s execution maturity.
Takeaway: Where the market resets next
Most analysts will tell you to wait for the next ER. I’ll give you a tighter signal.
Watch the July HBM4 pre-mass-production announcements.
If Hynix confirms their 12-stack HBM3E yield is above 60% and HBM4 tape-out is on schedule, the current valuation discount will vanish within two trading sessions.
If Samsung announces an HBM3E performance parity win with NVIDIA, the discount deepens.
Vulnerabilities aren’t bugs. They’re bounded risks with defined test vectors.
Patience isn’t about holding through fear. It’s about letting the architecture prove itself. SK Hynix is still the cleanest execution in memory right now.
The market overcorrected. But the correction gave you a better entry.
— Grace Lee