Photonics & Quantum.
The fifth tollbooth of the Physical Layer of AI. Silicon photonics, polymer electro-optic modulators, infrared optics, and photonic-quantum compute — the cohort that prices the bandwidth between transistors. 22.3% of the book. Live receipts below.
Four names. Four cost bases. Live P&L.
Layer 4 sector tape — owned + watchlist.
4 names I own (gold). 4 names I watch (silver). When the gold and silver move together on different catalysts, the layer is the trade — not the ticker.
| Status | Ticker | Last | Day | Week | Volume | Theme |
|---|---|---|---|---|---|---|
| Loading cohort… | ||||||
The weekly Layer 4 brief is for Founding 50 only.
Cohort moves with cause-of-move citations. Trim-trigger alerts on POET, LWLG, LPTH, XNDU. Pre-publish Moat Audits before they hit the free Substack. Free during dev for early members, but it may not stay free. Early access — get in before Pro launches.
Claim a Founding 50 seat →Why I promoted "Physics Frontier" into formal infrastructure.
Receipt 1 — Same customer base.
POET sells to hyperscalers and to the chip vendors hyperscalers buy from. XNDU sells to research labs and chip vendors. LWLG licenses modulator IP to the optical transceiver companies that ship into hyperscaler backbones. LPTH sells optics into the same transceiver supply chain. Every one of these companies' demand curves bends to a single number: how many megawatts of AI compute get racked in the next 36 months. That is the same curve as MU and CEG. If you accept Power and Compute as real layers, the photonic layer pencils with identical logic.
Receipt 2 — Same physics bottleneck.
Moore's Law is cooked. HBM is bandwidth-bound by copper interconnect. TSMC's Q1 2026 print showed 66.2% gross margin against a 63–65% guide — the fab is selling chips faster than it can build them. The bottleneck moved from transistor count to how fast bits move between transistors. Silicon photonics (POET) lifts the ceiling. Polymer EO (LWLG) lifts it further. Photonic-quantum (XNDU) eliminates the need for the ceiling. One bottleneck, three answers at three timescales.
Receipt 3 — Same kill vectors.
A real layer fails for layer-level reasons. If Microsoft and Meta cut data center capex 30%, every Layer 4 name drops together. If Nvidia announces in-house silicon photonics, POET and LWLG drop together while XNDU survives. If a quantum-winter narrative takes hold, XNDU drops and POET survives. The kill vectors are categorical, not idiosyncratic. That is the technical definition of a sector — and a sector that maps to AI infrastructure is a layer.
The five things that would make me rip up the map.
Get the weekly Photonics & Quantum cohort report.
Friday close numbers. Cause-of-move citations. Trim-trigger watch on POET, XNDU, LWLG, LPTH. Free, weekly, no spam. Pro tier launches Friday May 1 with daily alerts.
Live data: Yahoo Finance via /api/prices · Position cost basis from etrade · Updated on every page load · Not financial advice · No 1-10 AI scores · Receipts not predictions.
Sources: TSMC Q1 2026 IR · Motley Fool POET coverage · Quinn's research desk