←The mapResearch bank
Theses
Solid-state transformer: unrelieved constraint carrying semiconductor revenue as rent, not as scalePackage Perimeter as Hyperscaler Negotiating Leverage Against NVIDIACoWoS Gatekeeper Paradox: TSMC's Margin Reflects Crowding-Out, Not ConversionNVLink's Demand Is a Fragile Coalition of Neoclouds Under Collateral PressureWolfspeed: High-Voltage Moat Evaporates Under Thermal ScrutinySK Hynix: The Memory Supplier Suffering Its Own Customer's ConstraintAccelink: The Valuation Anomaly in China's Optical Chip IndigenizationCoWoS Binds the Buyer, Not the LandlordHBM5: The Transmission-Line Bottleneck Transfers Margin to the Memory SuppliersArm: The Revenue-Mix Pivot Consensus Hasn't PricedInnolight: The Hyperscaler Pricing Ratchet Disguised as GrowthLong-Context Inference: Infrastructure Debt Becomes the LeverageInterconnection queue converts datacenter operator debt into utility equity premiumAmazon: The Capex-Margin InversionAixtron: Customer Capex Discipline Masks Downstream GrowthShin-Etsu: Hidden Exposure to Datacenter Grid FailuresTransformer scarcity is pricing power for industrial conglomerates, not a hyperscaler cost problemMarvell's custom-ASIC rent accrues upstream, not in the fabless P&LKLA: the annuity thesis the equipment sell-side ignoresLG Innotek: Substrate Ambition Subsidised by the Wrong CustomerElectrical Steel Asymmetry: GE Vernova Captures Transformer Scarcity Without Steel ExposureHBM4: equipment leverage swamps memory maker concentrationABF substrate dual-gate arbitrage: substrate converters capture the spread the market assigns upstreamNAND Flash: Consumer Demand Collapse Disguises Enterprise Pricing PowerTesla: AI Capex Subsidy Disguised as Automotive TurnaroundGas Turbine Bottleneck Inverts the Natural Gas ExposureUALink: Consortium Activity Masks Structural IrrelevanceCo-packaged optics is a packaging play being misfiled as an optics transitionApplied Optoelectronics: Loss-Funded Market Access Captures Transceiver Supply ScarcityMicrosoft: Free Cash Flow Collapse Telegraphs Margin Compression Before It Hits the P&LRack Power Density: The Liquid Cooling Adoption MirageGPU collateral decay transmits to NVIDIA demand before physical constraints clearxAI: Vertical Integration Theater Masks Structural GPU DependencyAdvanced Packaging: The Scarcity Premium Flows Upstream, Not DownSilicon Wafer: Duopoly Ships into Oligopoly Margin Explosion, Captures NoneEaton: Margin expansion telegraphs demand scarcity more than growth can showMoE inference: stranded-power miners own the scarcest input, hyperscalers rent itApplied Materials: The Hidden Margin Trap in a Structural UpswingMurata: The Margin the Bottleneck Hasn't ReachedAlphabet: The Capex Ratchet Liquidity Mispriced as Platform OptionalityLam's service annuity masks exposure to memory's margin conversionEUV scarcity is priced into ASML, invisible in AlphabetConstellation: Hyperscaler Monopsony Masks Nuclear Fuel TransmissionOracle: Monopsony Rent Capture Masked by Consolidated AccountingFoxconn: Customer Concentration Absorbs AI Margin Before It Hits the P&LThe Conventional DRAM Squeeze: HBM Conversion Creates a Consumer Margin Crisis Through 2028Palantir: The Margin-Protected Infrastructure PlayDISCO: Memory oligopoly capex collapses the monopolistCXMT: The Supplier-Margin Windfall Hiding Inside the Subsidy StorySMIC: Subsidy Converts to Capacity Under Obscured Margin PressureQualcomm's low capex masks an IP-only future neither consensus nor bears have pricedTransceiver margin expansion is a mirage: buyer concentration at 1.6T resets pricing power the market prices as durableTokyo Electron: Memory Rent Disguised as Equipment MarginSilicon capacitors expose Intel's package productivity deficitArista: Hyperscaler capex intensity conceals margin compression riskGlobalFoundries: Customer Concentration Masks Structural Insulation from AI Capex Whiplash
Mechanisms
ABF substrate and build-up film supplyCoWoS advanced-packaging capacityConventional DRAM and NAND supply (HBM crowding-out)EUV tool capacity — the lowest rungGPU residual value as loan collateralGrid interconnection queue positionHeavy-duty gas turbine delivery slotsLarge power transformer lead timesSamsung memory long-term agreementsBuyer concentration tighteningHBF consortiumRent converting into capacityRent migrating upstreamRent not being competed awayUALink ConsortiumUltra Ethernet Consortiumco-packaged-optics displaces copperco-packaged-optics displaces optical-transceiveremib displaces cowosglass-substrate displaces abf-substratehybrid-bonding displaces euvsilicon-capacitor displaces mlccChip designDatacenter mathEfficiency arrives in steps, not trendsInference shapeMemory economicsPhotonicsTau scalingToken mathTokenomicsCapacity arriving — CoWoS (Chip-on-Wafer-on-Substrate)Capacity arriving — HBM4Capacity arriving — Silicon Wafer

CoWoS Gatekeeper Paradox: TSMC's Margin Reflects Crowding-Out, Not Conversion

gen-cowos · conviction computed 0.87 · status open · horizon — · as of 2026-08-19

TSMC is capturing rent from the constraint, but its 56% margin sits far below the 64% input-cost layer above it—meaning TSMC is the margin floor, not ceiling, in this stack. The constraint is binding severely on NVIDIA and AMD while leaving TSMC itself unconstrained, which is the signature of crowding-out rather than conversion.
Rests on filed figures, not on modelled shares. 8 premises (6 entity, 2 field); no derived cell is involved, so the undisclosed supply weights that put a range on other pages in this bank cannot move this one.
How 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

Exhibit 1Relative performance, indexed to 100How the names in this thesis have traded against SOXX.
70220369index100 = startAMD 280SOXX 213TSM 179NVDA 12512mo, indexed to 100 at start · dashed = SOXX benchmark

Series available as data/gen-cowos.csv

Exhibit 2Who pays CoWoS advanced-packaging capacity, and who keeps the moneyCapturers average 47.1% operating margin against payers' 47.0% — the owners of the scarce thing capture the rent, as expected.
Taiwan Semiconductor Manufac56.1%Analog Devices, Inc.38.1%SK Hynix68.0%NVIDIA Corporation64.0%Broadcom Inc.44.2%Advanced Micro Devices11.8%

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

CoWoS capacity is the AI-accelerator bottleneck, and TSMC's 56% operating margin reflects monopoly pricing power on the scarce resource. The market treats packaging allocation as the rent and expects TSMC to convert that rent into capacity as it has historically done with wafer fabs.

Variant

TSMC is capturing rent from the constraint, but its 56% margin sits far below the 64% input-cost layer above it—meaning TSMC is the margin floor, not ceiling, in this stack. The constraint is binding severely on NVIDIA and AMD while leaving TSMC itself unconstrained, which is the signature of crowding-out rather than conversion. Supply response is absent: no percentile data exists because the technology layer lacks historicals to rank against. The 25% capex-intensity decline signals capital is being withheld, not deployed.

Differentiator

Earnings models see TSMC's margin and infer pricing power. Supply-chain structure reveals TSMC is being out-earned by its own suppliers at 64% gross margin, and the 25-point capex collapse says the rent is not rotating into future relief—it is being banked.

Open questions

Falsifiers

unstructured TSMC capex intensity stays below 26% through year-end 2026
Customer-weighted capex_pull remains under 26% in the December 2026 computation
settles confirmed by 2027-01-15 · no machine-readable clauses yet
unstructured Input-cost pressure (supplier gross margin) stays above TSMC operating margin through Q1-2027
Derived input_cost_pressure exceeds field:tsmc.financials.operating_margin by at least 4 points in Q1-2027 data
settles confirmed by 2027-05-01 · no machine-readable clauses yet
unstructured AMD accelerator revenue growth decelerates more than NVIDIA's in the next two quarters
AMD Data Center GPU revenue growth QoQ falls below NVIDIA Data Center growth QoQ by >10 points in either Q3 or Q4-2026 earnings
settles confirmed by 2027-02-28 · no machine-readable clauses yet

Reasoning chain

TSMC sits below its suppliers in realized margin, making it the transmission mechanism for rent extraction rather than the primary beneficiary VALID
premises
  • Taiwan Semiconductor Manufacturing Company0.95 strong
    90%
    TSMC is the monopoly provider at 90% share
  • Taiwan Semiconductor Manufacturing Company — operating margin 56.1%1.00 strong
    Filed figure, Q2-2026
  • ABF substrate and build-up film supply0.88 strong
    Every package needs substrate; binding severely on both NVIDIA and AMD per constraint records

Composed 0.88 via and over 1 gating premise · 2 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link ABF substrate and build-up film supply at 0.88

Input-cost pressure at 64% exceeds TSMC's 56% operating margin with zero band spread. The monopolist is being out-earned by its suppliers, which inverts the expected rent distribution. This is structural: substrate and film makers capture the scarcity premium while TSMC converts capacity and competes on cycle time.

Capital intensity is collapsing rather than rising, indicating rent is being harvested instead of rotated into capacity additions VALID
premises
  • CoWoS advanced-packaging capacity0.93 strong
    Binding severely on NVIDIA and AMD; TSMC itself unconstrained per records
  • Taiwan Semiconductor Manufacturing Company — revenue growth $361.00 strong
    Filed growth rate, context for capex denominator

Composed 0.93 via and over 1 gating premise · 1 supporting premise shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link CoWoS advanced-packaging capacity at 0.93

Customer capex intensity fell 25% while the constraint remains severe and TSMC revenue grew 36%. Normally a binding capacity constraint with 36% revenue growth would pull forward capital deployment. The 25-point drop with common-mode flag says capital is being withheld across the customer base, not just by one hyperscaler. This is distribution, not reinvestment.

The constraint binds customers severely while leaving the supplier unconstrained, the hallmark of crowding-out allocation rather than technical bottleneck relief VALID
premises
  • NVIDIA Corporation0.96 strong
    72%
    Largest accelerator shipper, named in 72% of constraint-binding records
  • Advanced Micro Devices0.94 strong
    Second accelerator name, smaller allocation amplifies constraint severity per records
  • CoWoS advanced-packaging capacity0.92 strong
    Severe on NVIDIA/AMD, none on TSMC; allocation IS the forecast per constraint detail

Composed 0.92 via and over 1 gating premise · 2 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link CoWoS advanced-packaging capacity at 0.92

Severity=severe on customers, severity=none on the supplier. That asymmetry is allocation discipline, not a technical ceiling being raised. TSMC rations by price and favored-customer sequencing; customers cannot bid capacity free because substrate supply binds in series. The margin and capex pattern confirms the rent is being held, not cycled.

Sources