Macrontology
Research bank
Theses
Power Caps the AI Buildout — the bottleneck moved from chips to megawattsThe ABF Chokepoint Does Not Pay Its Owners — Buy the Bottleneck Fails HereASML / China-DUV displacement is overpriced — the sell-off prices near-full substitution off a 5-machine data pointThe China chokepoint moves from lithography to bonding — and the substitution trade is priced in the wrong categoryCost Per Token Is Set Outside the ChipCustom Silicon Pays a Different MerchantEMIB Shifts the Packaging Bottleneck to SubstratesHBM moat vs. DRAM commoditization — is SK Hynix's HBM-mix 'miss' actually bullish?HBM4 Qualification Concentrates Share, Not Spreads ItSilicon cannot emit light — indium phosphide is the chokepoint CPO relocates but cannot removeThe Neocloud Rent Is Consumed By The Asset Before It Reaches EquityNVIDIA's Rent Compresses Through Software, Not SiliconOptical attach is set by topology, not by shipments — linear optical TAM models are mis-specifiedThe Megawatt Is the Unit of AccountThe Upstream Is Single-Sourced and UnpricedChipflation Is a Wafer Allocation, Not a Demand ShockMemory-demand durability — does algorithmic efficiency cap the memory super-cycle?Neocloud margin re-rating — do open-source deployment + rising rental rates break the thin-margin cap?Semicap service is an installed-base annuity, and the market still prices these names as capex cyclicalsABF substrate dual-gate arbitrage: substrate converters capture the spread the market assigns upstreamAccelink: The Valuation Anomaly in China's Optical Chip IndigenizationAdvanced Packaging: Equipment Suppliers Capture the Scarcity PremiumAlphabet: The Margin Inversion Nobody PricesAmazon: The Capex-Margin InversionApplied Optoelectronics: Hyperscaler Procurement Arbitrage Driving Negative SpreadsArista: Hyperscaler capex intensity conceals margin compression riskArm: The Royalty Stream That Scales With Everyone's Margin but Its OwnCo-packaged optics transfers margin to substrate integrators, not optics specialistsConstellation: Hyperscaler Monopsony Masks Nuclear Fuel TransmissionCoWoS Constraint Prices Into the Wrong CompaniesEaton: Margin expansion telegraphs demand scarcity more than growth can showElectrical Steel Asymmetry: GE Vernova Captures Transformer Scarcity Without Steel ExposureFoxconn: Margin Squeeze Masked by AI Server Revenue MixGlobalFoundries: Capex Intensity Signals Strategic Exclusion from AI, Not WeaknessGPU collateral decay gates buildout faster than physical supplyHBM5: The Signal Integrity Tax Inverts the Memory Supplier Power BalanceTransformer lead times gate hyperscaler returns, not start datesInnolight: Negative Operating Leverage Hiding in Hyperscaler ConcentrationInterconnection queue converts hyperscaler capex into generator rentLong-context inference: memory constraint compresses cloud margin before GPU-poor plays feel itMicrosoft: Burning Cash to Rent Margin It Cannot OwnMoE inference: converted miners capture memory scarcity, hyperscalers leak itMurata: The Margin That Doesn't MoveNAND Flash: Datacenter Crowding-Out Misread as Memory Cycle RecoveryTransceiver margin expansion is a mirage: buyer concentration at 1.6T resets pricing power the market prices as durableOracle: Monopsony Rent Capture Masked by Consolidated AccountingPackage Perimeter Arbitrage: Optics Suppliers Capture Value NVIDIA Cannot RetainPalantir: The Margin-Protected Infrastructure PlayQualcomm's low capex masks an IP-only future neither consensus nor bears have pricedRack Power Density: The Liquid Cooling Adoption MirageShin-Etsu: The Hidden Rent in the DenominatorSilicon capacitor adoption arbitrages package economics invisible to component analystsSilicon Wafer Duopoly: Record Downstream Margins Signal Pricing Power Collapse, Not TightnessSK Hynix: The Customer Concentration That Validates the MoatSMIC: Bottleneck Arbitrage Disguised as National ChampionTesla: Capex Surge Hides Margin Compression at Suppliers

GPU collateral decay gates buildout faster than physical supply

gen-gpu-collateral-value · conviction — · status open · horizon — · as of 2026-08-10

Over $20bn of GPU-backed debt assumes 5-6 year residuals. H100 rental rates fell 60-70% in eighteen months, tightening advance rates and refinancing terms before next-gen deployment. The financing channel closes ahead of the physical one, stranding neoclouds with negative operating margins and debt maturities they cannot roll.
Rests on filed figures, not on modelled shares. 12 premises (6 entity, 4 field, 2 edge); 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. 903 of 1,848 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.
25157290index100 = startSOXX 221MSFT 97CRWV 6412mo, indexed to 100 at start · dashed = SOXX benchmark

Series available as data/gen-gpu-collateral-value.csv

Exhibit 2Who pays GPU residual value as loan collateral, and who keeps the moneyCapturers average 54.6% operating margin against payers' -2.1% — the owners of the scarce thing capture the rent, as expected.
NVIDIA Corporation64.0%Microsoft Corporation45.2%CoreWeave, Inc.-2.1%

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

Neocloud capacity scales with chip supply and hyperscaler budgets. CoreWeave, Lambda, and Crusoe grew triple-digit revenue by converting rental demand into leverage. The buildout is constrained by NVIDIA allocation and power, not capital structure.

Variant

Over $20bn of GPU-backed debt assumes 5-6 year residuals. H100 rental rates fell 60-70% in eighteen months, tightening advance rates and refinancing terms before next-gen deployment. The financing channel closes ahead of the physical one, stranding neoclouds with negative operating margins and debt maturities they cannot roll. Microsoft and hyperscalers own chips outright; CoreWeave and Lambda rent balance sheets collateralised by depreciating inventory.

Differentiator

Equity analysts model GPU deployment as capex-gated; credit prices collateral decay first. Rental rate collapse is observable, but the refinancing squeeze precedes the revenue miss by two quarters, and only the levered operators feel it.

Falsifiers

Open questions

Reasoning chain

CoreWeave cannot sustain negative operating margins through a refinancing cycle with falling collateral values VALID
premises
  • GPU residual value as loan collateral0.82 strong
    Constraint exists and binds levered operators; range reflects sparse public LTV data
  • CoreWeave, Inc. — operating margin -2.1%1.00 strong
    Filed; no uncertainty
  • GPU residual value as loan collateral supplies CoreWeave, Inc.1.00 strong
    Graph edge confirmed
  • CoreWeave, Inc.1.00 strong
    Name exists and trades

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 GPU residual value as loan collateral at 0.82

Negative margin plus collateral-dependent refinancing makes CoreWeave the cleanest short against improving hyperscaler rental economics

Lambda's revenue growth decouples from CoreWeave's as residual assumptions diverge by 2027 VALID
premises
  • GPU residual value as loan collateral0.78 strong
    Weaker if operators share identical capital structures; strengthens if Lambda holds less leverage
  • Lambda, Inc. — revenue growth $791.00 strong
    Filed growth rate
  • CoreWeave, Inc. — revenue growth $2001.00 strong
    Filed growth rate
  • Lambda, Inc.1.00 strong
    Name exists

Composed 0.78 via and over 1 gating premise · 3 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link GPU residual value as loan collateral at 0.78

Lower growth Lambda likely carries lighter debt load; growth spread widens as collateral terms tighten

Microsoft benefits structurally as neocloud refinancing risk shifts rental demand back to owned-outright hyperscaler capacity VALID
premises
  • GPU residual value as loan collateral0.75 strong
    Presumes customers treat neocloud and hyperscaler GPU as substitutes when neocloud credit spreads widen
  • Microsoft Corporation — operating margin 45.2%1.00 strong
    Filed margin
  • GPU residual value as loan collateral supplies Microsoft Corporation1.00 strong
    Graph edge confirmed
  • Microsoft Corporation1.00 strong
    Name exists

Composed 0.75 via and over 1 gating premise · 3 supporting premises shown but not multiplied in — citing a filed figure should not cost conviction · Weakest link GPU residual value as loan collateral at 0.75

Positive operating margin plus balance-sheet ownership insulates Microsoft from refinancing risk; neocloud distress is hyperscaler opportunity

Sources

Write-up

Pre-filled skeleton: gen-gpu-collateral-value.md