← research bank

Microsoft: Burning Cash to Rent Margin It Cannot Own

gen-microsoft · conviction — · status open · horizon — · as of 2026-08-07

Microsoft is destroying free cash flow—$16B trailing versus $116B capex—to lease compute from suppliers capturing 42-44% operating margins while its own blended margin cannot exceed software-era levels.
Rests on filed figures, not on modelled shares. 12 premises (7 field, 5 entity); no derived cell is involved, so the undisclosed supply weights that put a range on other pages in this bank cannot move this one.

Exhibits

Exhibit 1Relative performance, indexed to 100How the names in this thesis have traded against SOXX.
52170287SOXX 225MSFT 9712mo, indexed to 100 at start · dashed = SOXX benchmark

Series available as data/gen-microsoft.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.

Exhibit 3What the conviction is actually made ofEach premise and the number it composes to. A conjunction of plausible premises is far weaker than any of them.
Microsoft's free cash flow collapsed to $16B trailing despite $305B revenue and 69% gross margin because $116B capex finances suppMicrosoft Co… — free cash flow (TTM) $16.36… †100.0%Microsoft Corporatio… — capex (TTM) $115.90… †100.0%Broadcom Inc.88.0%Broadcom Inc. — operating margin 44.2% †100.0%COMPOSED (and)88.0%

† 3 premises marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.

One gating premise, so the conclusion is exactly as strong as it. The figure is an ordering device, not a calibrated probability — see how the numbers are made.

Weakest link: Broadcom Inc. at 0.88 — Custom silicon and switching supplier; capex flows to its NRE and capacity, not Microsoft's balance sheet.

Supplier operating margins—Broadcom 44%, Arista 42%, Marvell 16%—exceed what Microsoft can extract from hyperscaler competition, sArista Networks85.0%Arista Networks — operating margin 42.0% †100.0%Marvell Technology82.0%Microsoft Corporati… — operating margin 45.… †100.0%COMPOSED (and)69.7%

† 2 premises marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.

70% if the 2 gates are independent, 82% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.

Weakest link: Marvell Technology at 0.82 — Custom and merchant SerDes/PAM4 supplier; physical-layer constraints prevent substitution or integration.

Hyperscaler purchasing power is neutralized because Google, Amazon, and Meta face identical supplier oligopoly and power/space conMicrosoft Corporation79.0%Broadcom Inc. — revenue growth $48 †100.0%Arista Networks †83.0%Arista Networks — revenue growth $29 †100.0%COMPOSED (and)79.0%

† 3 premises marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.

One gating premise, so the conclusion is exactly as strong as it. The figure is an ordering device, not a calibrated probability — see how the numbers are made.

Weakest link: Microsoft Corporation at 0.79 — One of four hyperscalers bidding for constrained supply; competitive intensity removes monopsony leverage that defined prior capex cycles.

The variant

Consensus

Microsoft's $116B capex run-rate funds Azure AI leadership and locks in OpenAI exclusivity, justifying the 22× forward multiple. The market assumes capital intensity converts to durable margin via enterprise software lock-in and hyperscaler economies of scale. Azure's 45% revenue share and 69% gross margin frame this as cloud infrastructure that happens to buy GPUs, not a structurally different business.

Variant

Microsoft is destroying free cash flow—$16B trailing versus $116B capex—to lease compute from suppliers capturing 42-44% operating margins while its own blended margin cannot exceed software-era levels. The supply chain reveals merchant silicon and custom-silicon paths converge on the same economic outcome: Microsoft finances supplier R&D at zero return, because hyperscaler purchasing power means nothing when three firms control every alternative and physical constraints bind all buyers equally. The $190B 2026 plan accelerates wealth transfer, not moat deepening.

Differentiator

Earnings analysts see capex as investment in owned infrastructure; supply-chain structure shows Microsoft rents margin from Broadcom, Arista, and Marvell—who grow faster and earn more—while competing hyperscalers face identical constraints, eliminating the bargaining asymmetry that made Azure a margin vehicle in the software era.

Falsifiers

Open questions

Reasoning chain

Microsoft's free cash flow collapsed to $16B trailing despite $305B revenue and 69% gross margin because $116B capex finances supplier expansion, not owned infrastructure. VALID
premises

Cash generation fell 75% as capex septupled, yet the capital buys no owned margin—it funds Broadcom's 44% operating income and supplier capacity expansion.

Supplier operating margins—Broadcom 44%, Arista 42%, Marvell 16%—exceed what Microsoft can extract from hyperscaler competition, so every incremental capex dollar cements a margin structure Microsoft cannot arbitrage. VALID
premises

Microsoft's 45% blended margin reflects legacy software; AI capex buys 42-44% supplier margin that hyperscaler scale cannot compress, so the incremental business is margin-destructive at the consolidated level.

Hyperscaler purchasing power is neutralized because Google, Amazon, and Meta face identical supplier oligopoly and power/space constraints, eliminating the demand asymmetry that previously let Microsoft extract terms. VALID
premises

All hyperscalers bid for the same constrained capacity from three suppliers, so Microsoft's $116B buys queue priority and OpenAI exclusivity but zero margin capture—bargaining power requires demand asymmetry, and 2026 AI capex erased it.

Write-up

Pre-filled skeleton: gen-microsoft.md