HBM4: equipment leverage swamps memory maker concentration
gen-hbm4 · conviction computed 0.76 · status open · horizon — · as of 2026-08-19
hbm4.demand_pull>=83.8124 — 68.61 to 90.9How to read the numbers on this page
A range instead of a point. 980 of 1,929 supply weights are not disclosed by anyone. Where redrawing them across their plausible range moves a figure by more than 25%, the figure is shown as a range and marked. A tight number is ground you can stand on; a wide one is not.
Not every premise is scored. A premise citing something the model verifies on every rebuild — a filed figure, a graph edge, a computed cell — is a PRECONDITION, not a risk. It gates the conclusion but contributes no uncertainty, because charging a conclusion for being verifiable made well-evidenced arguments look weaker than vague ones.
Composed two ways. Where several premises gate a conclusion, the figure is given as "X% if independent, Y% if they move together". They are claims about one industry, so the truth is between and nobody can say where. Treat it as an ordering device, not a calibrated probability.
Dated. Each figure is computed from facts, and the page states the age of the oldest one beneath it. The full arithmetic runs from a published model config to company revenue exposure.
Exhibits
Series available as data/gen-hbm4.csv
Green/blue = model marks it as CAPTURING the rent (unbound and supplies the scarce good); faded = PAYING it (bound severe or moderate). Operating margin, live.
The variant
Consensus
HBM4 supply concentration drives the thesis—SK Hynix (55% share) and Samsung (25%) command pricing power, with Micron trailing at 20%. The narrative is memory-supplier margin expansion into a capacity-constrained substrate/packaging bottleneck. Investors long the memory makers, short the hyperscalers paying up.
Variant
The equipment suppliers enabling HBM4—DISCO, Hanmi, ASMPT—exhibit structural leverage the memory consensus ignores: 42%, 44%, and 0.8% operating margins respectively, feeding into a layer growing 84% YoY with tight constraint load (1.0). SK Hynix's 68% margin is already pricing what it can extract; DISCO's 42% margin on 27% growth into the same wave is mispriced relative to the durability of the gate it controls. Equipment names capture HBM4 margin expansion with lower embedded expectations and tighter positional control than the memory suppliers trading at scarcity valuations.
Differentiator
Earnings models anchor on disclosed supply shares and miss that equipment margin is levered to *all three* memory makers simultaneously, while each memory supplier competes away its own rent. Supply-chain structure implies concentration at the tool layer, not the output layer.
Open questions
- What is DISCO's captive vs merchant tool mix for HBM dicing—if SK Hynix or Samsung vertically integrate tooling, thesis duration shortens sharply
- When does die-to-wafer hybrid bonding displace wire bonding in HBM assembly, and what share does Hanmi retain vs Besi/AMAT
- CoWoS allocation transparency: if TSMC discloses customer-specific packaging slots, Micron's excess supply becomes measurable and thesis timing tightens
Falsifiers
Reasoning chain
DISCO Corporation0.82 strongDicing saws are non-substitutable in HBM stack separation; DISCO's position is structural, not cyclicalDISCO Corporation — operating margin 42.3%1.00 strongFiled margin demonstrates rent capture already embeddedHanmi Semiconductor — operating margin 43.6%1.00 strongDie bonder margin shows similar structural positioningHBM4 — demand pull at least 83.8124 [68.6–90.9 depending on shares nobody discloses]1.00 strongCustomer growth pulls through all three memory suppliers equally
Composed 0.82 via and over 1 gating premise · 3 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link DISCO Corporation at 0.82
Equipment layer margin (42-44%) into 84% demand growth with non-substitutable position durably captures HBM4 expansion, while memory maker margin (68% SK Hynix) already prices scarcity and invites competitive response
SK Hynix — operating margin 68.0%1.00 strongFiled operating margin, extreme even within memory sectorSK Hynix0.75 strongAssume margin converts to capacity adds within 18 months—standard memory sector playbook, visible in Samsung's planned 80 kwpm and Micron's 20 kwpm adds by 2026-Q4HBM4 — si margin at least 1.06981.00 strongSupply interval / overshoot period >1 signals capacity is being monetized into expansion
Composed 0.75 via and over 1 gating premise · 2 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link SK Hynix at 0.75
Extreme margin with visible capacity response (Samsung +80k, Micron +20k wafer starts by Q4 2026) means SK Hynix's rent is already being converted into competing supply—the standard way good margins stop being good
Micron Technology0.70 moderate20%65%20% HBM4 share with 65% margin at $1.1T market cap prices sustained allocation; Samsung's 80 kwpm add (vs Micron's 20 kwpm) and SK Hynix's 50 kwpm add compress that share by mid-2027Micron Technology — revenue growth $1671.00 strongYoY growth from small HBM3E base creates optics that obscure share loss velocityHBM4 — constraint breadth at least 31.00 strongThree constraints binding customers (CoWoS, ABF, EUV) mean supply adds relieve customer gates, not create new demand for Micron's incremental output
Composed 0.70 via and over 1 gating premise · 2 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link Micron Technology at 0.70
Micron's absolute growth masks share compression as SK Hynix and Samsung add 130 kwpm combined by Q4 2026 into customers constrained by packaging/substrate, not memory—new Micron supply arrives into a queue, not a bid
Sources
- KRW 44.2B SK Hynix TC bonder order for HBM4 ramp, June 2026 — link acc 2026-07-23
- Counterpoint Research global DRAM and HBM market share dataset — link acc 2026-07-21
- TTM revenue 132.08T KRW (~$95B), gross margin 68.34%, operating margin 58.58%, forward P/E 6.27, TTM capex 28.89T KRW (~$20.77B) — link acc 2026-07-21
- Market cap $1.235T USD (June 2026) — link acc 2026-07-21
- FQ3 2026 revenue $41.46B (vs $9.30B year-ago), gross margin 84.6% GAAP, operating margin 80.4% GAAP, capex $7.1B (net); nine-month revenue $78.96B, capex $19.6B gross — link acc 2026-07-21
- Market cap ~$1.10T, forward P/E 6.03, TTM revenue $90.27B — link acc 2026-07-21
- >70% global share dicing saws and grinders; record FY2025 shipments on AI demand — link acc 2026-07-23
- HBM4 supplier allocation (notably NVIDIA HBM4): SK hynix ~mid-50%, Samsung ~mid-20%, Micron ~20% — Samsung more competitive on the newer node than in HBM3E — link acc 2026-07-21