← research bank

Package Perimeter Arbitrage: Optics Suppliers Capture Value NVIDIA Cannot Retain

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

Package shoreline is a zero-sum allocation game NVIDIA plays but cannot win. Every millimetre ceded to optics or power escape is permanently lost to other functions, and the 25x gap between achieved and theoretical bandwidth density reveals the constraint binds hard today.
Rests on filed figures, not on modelled shares. 10 premises (4 entity, 3 edge, 3 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.

Exhibits

Exhibit 1Relative performance, indexed to 100How the names in this thesis have traded against SOXX.
86185285SOXX 22512mo, indexed to 100 at start · dashed = SOXX benchmark

Series available as data/gen-package-shoreline.csv

Exhibit 2What 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.
Package perimeter scarcity forces NVIDIA to pay rising optics ASPs as cluster topologies shift from scale-up to scale-outPackage shoreline (perimeter and cross-sect…88.0%Package shoreline (perimeter and cross-sect… †100.0%Package shoreline (perimeter and cross-sect… †100.0%Datacenter Ai92.0%COMPOSED (and)81.0%

† 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.

81% if the 2 gates are independent, 88% 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: Package shoreline (perimeter and cross-section contention between power, memory and data) at 0.88 — Constraint is described but cross-section TSV data and vertical power delivery relief are unquantified

Optics suppliers capture margin expansion that NVIDIA's platform dominance cannot preventCoherent Corp. †100.0%Lumentum Holdings †100.0%Lumentum Holdings — revenue growth $69 †100.0%NVIDIA Corporation — operating margin 64.0% †100.0%

† 4 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.

No gating premise. Every premise here is supporting evidence, so this conclusion states no necessary condition — it is asserted from cited data rather than derived from a claim that could fail. Read it as a summary, not as a falsifiable call.

Weakest link: Coherent Corp. at 1.00 — Coherent is a primary co-packaged optics supplier exposed to this constraint

Optics margin capture accelerates faster than NVIDIA's pricing power can offset itAi Capex Cycle supplies Package shoreline (… †100.0%Coherent Corp. — operating margin 12.1% †100.0%

† 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.

No gating premise. Every premise here is supporting evidence, so this conclusion states no necessary condition — it is asserted from cited data rather than derived from a claim that could fail. Read it as a summary, not as a falsifiable call.

Weakest link: Ai Capex Cycle supplies Package shoreline (perimeter and cross-section contention between power, memory and data) at 1.00 — AI capex directly funds the platforms subject to perimeter scarcity

The variant

Consensus

NVIDIA's accelerator dominance translates directly to sustained margin expansion. The GB200 and VR200 platforms exploit co-packaged optics and advanced memory, but these are cost-optimised supply-chain inputs that preserve NVIDIA's pricing power. Higher datacenter AI capex flows primarily to the platform integrator.

Variant

Package shoreline is a zero-sum allocation game NVIDIA plays but cannot win. Every millimetre ceded to optics or power escape is permanently lost to other functions, and the 25x gap between achieved and theoretical bandwidth density reveals the constraint binds hard today. Coherent and Lumentum extract rent at the perimeter because NVIDIA has no physical recourse: the margin transfer is structural, not negotiable, and it accelerates as training clusters scale horizontally rather than vertically.

Differentiator

Consensus treats co-packaged optics as a BOM line; the physical constraint makes it a tollgate. The optics suppliers sit outside the package, so they do not compete for the contested resource—they ARE the scarce resource, and scarcity accrues to the seller. NVIDIA's 64% operating margin coexists with Lumentum's 69% revenue growth because the platform winner is also the constrained buyer.

Falsifiers

Open questions

Reasoning chain

Package perimeter scarcity forces NVIDIA to pay rising optics ASPs as cluster topologies shift from scale-up to scale-out VALID
premises

Composition 0.81: the mechanism is clear and the exposure is direct, gated on training topology trends

Optics suppliers capture margin expansion that NVIDIA's platform dominance cannot prevent VALID
premises

Composition 1.0: the financials are filed and the structural transfer is the thesis claim itself

Optics margin capture accelerates faster than NVIDIA's pricing power can offset it VALID
premises

Composition 0.79: the rate claim depends on the bandwidth utilisation gap being real, not idealised

Write-up

Pre-filled skeleton: gen-package-shoreline.md