Memory 4-Way: SK +42% vs Micron +6%, Who Wins?

"AI memory supercycle, ranked across revenue growth, P/S normalization, and SOTP"

Item	Value
Top Pick	SK Hynix (KRX: 000660)
2nd / 3rd / 4th	Samsung / SanDisk / Micron
FX Assumption	USD/KRW = 1,462 (2026-05-11)
Horizon	6 to 12 months (FY26 to FY27)
Models	(1) Revenue growth (2) P/S normalization (3) SOTP overlay
Cycle Temperature	72°C / 100 (peak approaching)
Attractiveness	SK Hynix ★★★★★ / Samsung ★★★★☆ / SanDisk ★★★☆☆ / Micron ★★☆☆☆

⚠️ Hierarchy of models: §6 P/S normalization and §7 SOTP are the substantive models; §5 revenue growth is a supporting sanity check. Although §5 takes a per-share form, the alpha is reverse-engineered from SK Hynix consensus, which makes it mathematically equivalent to "hold each company's current P/S constant and grow revenue into FY27." All final decisions should anchor on §6 and §7.

1. Executive Summary

Share price is just a function of share count, so what matters is not per-share price but market cap relative to revenue (P/S). SanDisk trades around ₩2.32M per share, the highest of the four in KRW terms, yet its market cap is only one-fifth of Samsung's. Flipping to P/S inverts the rankings.

Memory-only P/S ranking after SOTP adjustment: Samsung memory implied 5.19x < SK Hynix 6.62x < SanDisk 11.4x < Micron 16.7x.

Blended Three-Model Targets

⚠ §6 mean-revert scenarios assume convergence; Samsung's +200% is a blended-P/S optical illusion driven by non-memory units, which §7 SOTP corrects to +16%.

Key takeaways:

• Top Pick: SK Hynix, consistent +40s% across all three models. 72% operating margin, HBM lead. Buy here
• 2nd: Samsung, memory implied P/S 5.19x is a -22% discount to SK. HBM4 NVIDIA qualification or a DS spin-off could add +30% on top
• 3rd: SanDisk, $42B backlog is real, but +4,109% over one year is overextended. Size small
• 4th: Micron, P/S 16.7x is the most expensive of the four. NVIDIA Vera Rubin premium is already in the price

2. Price Snapshot (2026-05-11)

All four are at record highs, but SanDisk operates in a different volatility regime. A +4,109% one-year run is effectively crypto-like, with realized beta above 5 and a one-quarter miss capable of producing 30 to 50% drawdowns.

3. Fundamentals

Latest Quarter

SK Hynix's 72% operating margin is unprecedented in manufacturing. Normal memory margins sit in the 20 to 30% range, and even cycle peaks rarely exceed 40 to 50%. The 72% print reflects supercycle dynamics layered with HBM mix premium.

Next-Quarter Guidance / FY Consensus

HBM / Core Product Positioning

Earnings Call Tone

• Samsung: Explicitly used "AI memory supercycle"; DS operating income ₩53.7T represents 94% of total. Mobile and display drag
• SK Hynix: First-ever ₩50T quarter, all-time record. HBM customer pre-contracts as base. Highlighted the 72% operating margin
• Micron: Framed Q3 guidance as "exceeding the entirety of FY24 revenue." FY26 capex >$25B, FY27 +$10B incremental
• SanDisk: Data center +645% YoY, $42B backlog plus $11B financial guarantees. NAND supply shortage guided through 2028. Sell-side reaction is mixed; Seeking Alpha framed it as "The Call Was A Warning"

Supplementary Valuation Metrics

How to Read These Metrics

EV/EBITDA (enterprise value over pre-D&A profit): memory is a heavy-capex industry, so depreciation can distort P/E. EV/EBITDA gives a cleaner view of operating cash generation.

• Normal cycle average: 5 to 8x
• Cycle peak: 10 to 15x
• All four at 12 to 36x now, all in peak territory, with SanDisk's 35.7x at a level the industry has never seen

Forward P/E (next-12-month earnings): looks artificially low because earnings are spiking.

• SK Hynix at 5.25x means "1 year and 5 months of earnings recoups market cap," which seems cheap in absolute terms, but cyclicals have one-off denominators

PEG (P/E divided by growth) ⚠ a trap for memory:

• Generally PEG below 1.0 is attractive
• All four sit at 0.03 to 0.15, which looks "extremely attractive" for non-cyclical stocks
• But memory EPS growth is one-time. The next downcycle can take EPS down 50 to 90%, flipping a 0.03 PEG to 5.0 overnight
• Bottom line: low PEG in memory does not equal buy signal. Useful for stable growers like SaaS or healthcare, near-useless for cyclicals. We do not use PEG in our decision framework

Cross-Read

• EV/EBITDA reaches the same conclusion as P/S: SK fair → Samsung → SanDisk → Micron (most expensive)
• Forward P/E is a cycle trap: 5 to 9x looks cheap, but in a downcycle P/E can go negative or balloon to 50x
• Do not use PEG for memory: do not let the 0.03 to 0.15 numbers fool you
• The three-model hierarchy (§5 revenue growth, §6 P/S normalization, §7 SOTP) remains the spine. EV/EBITDA, forward P/E, and PEG are sanity checks only

4. One-Paragraph Briefs

Samsung (KRX: 005930), ₩285,500

1Q26 revenue ₩133.9T and operating income ₩57.2T (+756% YoY) are all-time records. DS contributes ₩53.7T (94% of total). HBM4 production start partially clears the late-mover discount. But mobile and foundry weakness mask the memory print, leaving the memory implied P/S at 5.19x, a -22% discount to SK Hynix at 6.62x. Strength: cheapest memory on SOTP, HBM4 production. Weakness: HBM share trails SK. Risk: another discount if NVIDIA HBM4 qualification slips.

SK Hynix (KRX: 000660), ₩1,908,000, Top Pick

1Q26 revenue ₩52.58T and operating income ₩37.61T at 72% margin. Revenue +198% YoY, operating income +406% YoY. HBM3E and HBM4 share leader. FY26 operating income consensus ₩247 to 280T, FY27 ₩338 to 378T. PTs: Nomura ₩2.34M, SK Securities ₩3.00M. Strength: pure memory exposure, dominant margins, HBM #1. Weakness: already +904% over one year. Risk: NAND price weakness (NAND mix ~30%).

Micron (NASDAQ: MU), $783.52 (~₩1,145,500)

FY26 Q2 revenue consensus $19.07B (+450% YoY), Q3 guide >$25B. HBM4 36GB 12H is exclusive to NVIDIA Vera Rubin. Calendar 2026 HBM sold out. Aggressive capex expansion (FY26 $25B+, FY27 +$10B). 38 analysts at $517.50 median, $600 high. Strength: HBM exclusive, highest US-market beta. Weakness: P/S 16.7x is the most expensive of the four. Risk: capex overshoot, NVIDIA concentration.

SanDisk (NASDAQ: SNDK), $1,588.48 (~₩2,322,400)

Q3 FY26 revenue $5.95B (+97% QoQ, +251% YoY). GM 78.4%. Q4 guide $7.75 to 8.25B (GM 79 to 81%, EPS $30 to 33). Five multi-year contracts cover one-third of FY27 bits, $42B backlog and $11B guarantees. NAND supply shortage guided through 2028. Strength: locked-in backlog, NAND supercycle leverage. Weakness: +4,109% one-year run, extreme absolute price volatility. Risk: post-peak NAND ASP mean reversion.

5. Revenue-Growth Model (Supporting Sanity Check)

⚠ The surface form of this model is "current price × (1 + α × revenue growth)", but α = 1.38 was back-solved from SK Hynix consensus, which makes the model mathematically equivalent to "hold each company's current P/S constant and grow revenue into FY27." It assumes the 5.3x gap in P/S between Samsung (3.16x) and Micron (16.71x) persists forever, an unrealistic premise. Treat as a sanity check, not a primary model.

Scenario Targets

Probabilities: Bear 25 / Base 50 / Bull 25.

Probability-Weighted Expected Return

Risk-adjusted attractiveness is roughly comparable, but SanDisk's absolute volatility is much higher and demands a smaller position. The decisive limitation is the "current P/S constant" assumption baked into all of these numbers. Micron's +55% assumes P/S stays at 16.71x; in §6.2 we show that reversion to the 9.47x peer average inverts the answer to -43%.

6. P/S Normalization (§6 Substantive Model)

6.1 Current P/S Matrix

6.2 Scenario A, Converge to Peer Average (9.47x)

6.3 Scenario B, Converge to SK Hynix (6.62x)

6.4 Scenario C, Converge to Micron (16.71x), Bull/Peak, 5 to 10% Probability

6.5 Conclusions

1. Micron expensive across the board, downside of -21 to -60% in every scenario
2. SK Hynix fairly valued, near the peer average with +40 to 152% in every other path
3. SanDisk bimodal, ranging from -42% to +58% by scenario
4. Samsung requires SOTP, half of the 3.16x optic is non-memory drag

7. SOTP for Samsung (§7 Substantive Model)

7.1 Why SOTP

Samsung's 3.16x blends memory (fair ~6.6x), system LSI / foundry (~7x), MX (~1x), and CE / SDC (~0.5 to 0.7x). A single multiple cannot be compared to a pure-play like SK Hynix.

7.2 FY26E Segment Buildup

7.3 Memory P/S Sensitivity

7.4 Samsung Memory Implied P/S (Apples-to-Apples)

Samsung memory implied Mcap = 1,704T - (454T + 151T + 91T) = 1,009T
Samsung memory implied P/S = 1,009T / 194T = 5.19x

Samsung memory trades at roughly a -22% discount to SK Hynix. Pure memory re-rating would unlock about +22% on this segment alone.

8. Blended Targets and Recommended Weights

When §5 and §6 conflict, we prioritize §6. Micron's +55% vs -43% is the biggest gap, and the blended +6% is a conservative midpoint. SanDisk's +124% vs -17% leans toward §5 if the NAND cycle extends another 1 to 2 years, or §6 if multiples compress.

Recommended Portfolio (100% Memory Allocation Assumed)

9. Cycle Peak Diagnostic, 10 Signals / 72°C

Current score: 7/10 = 72°C, "peak approaching, no traversal signal yet". If any of signals 8, 9, or 10 flip, cut exposure by 30 to 50%. The next earnings cycle (Micron Q3 FY26 in late June, SK and Samsung 2Q in late July, SanDisk Q4 FY26 on 8/13) is the decisive read.

10. Key Risks

1. HBM oversupply: Samsung, SK, and Micron all expanding capex simultaneously; 2H 2027 oversupply possible
2. NAND price crash: a 30% post-peak ASP drop hits SanDisk directly
3. CXMT / YMTC catch-up: China's memory rise pressures DRAM and NAND with discount pricing on a 3 to 5 year horizon
4. Hyperscaler capex slowdown: HBM demand softens immediately if NVIDIA / MSFT / AMZN / Meta cut guidance
5. FX: USD/KRW assumed at 1,462; a ±5% move shifts US-name targets in KRW by ±5%
6. SanDisk volatility: +4,109% one-year, +569% YTD; a single quarterly miss could mean -30 to -50%
7. NVIDIA concentration: Micron's HBM4 exclusivity exposes it to AMD MI400 share shifts

11. Recommendation Summary

Conclusion

The four memory names are at all-time-record one-year returns (+420 to +4,109%) on the back of the AI supercycle and HBM premium. But once you switch from per-share price to market cap divided by revenue (P/S), the rankings clarify quickly. SK Hynix delivers a consistent +40s% across all three models and is a buy now. Samsung's memory implied P/S of 5.19x is a -22% discount to SK, but non-memory drag holds the blended upside at +23%. SanDisk's fundamentals (revenue +251% YoY, $42B backlog) are real, but +4,109% in one year argues for diversification. Micron at 16.7x P/S is the most expensive of the four, with only +6% blended upside.

The cycle thermometer reads 72°C, peak approaching. Seven of ten peak signals are on; hyperscaler capex, DRAM spot pricing, and Chinese price offensives are the three signals still quiet. If any of those flips, cut by 30 to 50% immediately. The next earnings cycle (June Micron, July SK and Samsung, August SanDisk) is the deciding read.

For the plain-English version, see the memory four-way beginner guide. For comparison with the standout single name of the season, see the SanDisk +3,685% analyst report and SanDisk beginner guide, the gap between NAND pure-play beta and a four-name diversified basket becomes obvious.

This report is for market analysis purposes and does not constitute investment advice. All three models rely on assumptions; actual prices may diverge meaningfully due to cycle transitions, macro, FX, and company-specific events. USD/KRW is assumed flat at 1,462, so US-name targets in KRW move proportionally with FX. SanDisk's +124% base scenario depends on the NAND supercycle extending through 2028; cycle signals warrant immediate re-evaluation. All investment decisions remain the responsibility of the reader.