SK Hynix — First-Principles Forecasting Model

A layered build: bit output (capacity-checked) × ASP (shortage-driven) → revenue → operating leverage → profit → EPS → fair value. Every driver is a slider — flex the assumptions and watch profit and valuation move. Defaults are aggressive-but-grounded, not conservative. ~1,350 KRW/$. Current ≈₩2.15M.

Why this exists: the market prices linear growth; memory is currently in a supply-constrained, price-inflating regime that compounds (volume × price × margin = three multiplicative vectors). This model makes each vector explicit so you can pressure-test where the exponential comes from — and where it would break.

The Layers (drag to flex)

Layer 1 · Annual bit (volume) growth24%

Reality check: 1c DRAM to 190K wpm, M15X live, Yongin ~400K wpm, ASML EUV fleet doubling by end-2027 → supports ~22-25%/yr.

Layer 2 · 2026 ASP (price) growth150%

KB: DRAM contract +194% / NAND +244% for 2026. TrendForce Q2 +58-63% QoQ.

Layer 2 · 2027 ASP growth40%

Shortage persists (Goldman/KB/chairman to 2030) → further increases off a high base.

Layer 2 · 2028 ASP growth12%

The swing year: new capacity arrives → prices plateau or soften. Push negative to model a glut.

Layer 3 · Operating margin76%

Q1'26 actual 72%; Goldman 70-80%; KB 78.1%. Price flows to OP on a fixed cost base (operating leverage).

Layer 4 · Exit P/E on 2028 EPS10x

Trades ~7x 12mo-fwd today; Korean semis avg ~23x. Cyclical memory historically 6-12x mid-cycle.

Model Output

2028 Op Profit

₩—

$—

2028 EPS

₩—

$—

Fair Value / share

₩—

$—

Upside vs ≈₩2.15M

—%

at exit multiple

	2025A	2026E	2027E	2028E
Revenue ₩T	97	—	—	—
Revenue $B	72	—	—	—
Op profit ₩T	47	—	—	—
Op profit $B	35	—	—	—
EPS ₩K	59	—	—	—

The Layered Cake — how revenue builds (₩T)

Prior-year base + Volume (bits) + Price (ASP)

Each year stacks volume growth (capacity) then price growth (shortage) onto the prior year. The orange band is the part of the story the market struggles with — pricing power in a structural shortage.

Supply Reality Check (can it physically be made?)

2026 capex	₩40T (~$30B)
ASML EUV	~30 machines, fleet doubling by end-2027
1c DRAM	190K wpm by end-2026, >50% of DRAM
Yongin	~400K wpm over time (Ph.1 ₩31T)
P&T7 packaging	₩19T — world's largest HBM assembly

—

Cross-check vs the Street (2028E Op Profit)

This model yours	₩—
Goldman (Jun 1)	₩454T ~$336B
2026 anchor: Goldman	₩271T ~$201B
2027 anchor: Goldman	₩401T ~$297B
Consensus 2026 rev	₩318T ~$235B

—

Key finding from the model: to reproduce Goldman's decelerating OP path (₩271T → ₩401T → ₩454T, growth slowing to +13%), the implied price path is roughly +195% (2026), +20% (2027), then −10% (2028). In other words, Goldman's ₩454T already bakes in a 2028 price decline, offset by ~25% volume growth. So ₩454T is the moderate case, not the aggressive one — if the shortage holds and 2028 prices stay flat-to-up (the "First-principles bull" preset), 2028 OP lands nearer ₩500-550T. That gap is precisely the exponential the market hasn't priced.

First-principles conclusion. —

Model assumptions: net profit ≈ op profit × 0.85 (calibrated so default 2026 EPS ≈ consensus ₩281K); 712M shares (post ₩12.2T cancellation); base FY2025 revenue ₩97.1T (actual). Revenue_t = Revenue_t-1 × (1+bit) × (1+ASP_t). This is a scenario tool, not company guidance. Forward figures are estimates; verify against filings each quarter.