AXTI is InP substrate bottleneck for optical transceiver supply chain (800G→1.6T→3.2T); analyst trimmed position but keeping exposure; demand exceeds supply with 2x capacity coming by end 2026; CPO inflection 2027+
AI capex → GPUs → optical interconnect → transceivers (800G/1.6T/3.2T) → EML/DFB lasers → InP substrates (AXTI supply). Supply-constrained (demand > supply), fresh capex on 6-inch InP R&D underway, QoQ earnings improvements expected as backlog converts. Analyst trimmed last week to rotate gains elsewhere.
Everything's gone quiet around $AXTI suddenly? They're still a vital InP substrate bottleneck: AI capex ↓ More GPUs ↓ Optical interconnect ↓ Optical transceivers (800G → 1.6T → 3.2T) ↓ Each transceiver needs EML/DFB/CW lasers ↓ Those are fabbed on InP substrates ↓ $AXTI (and Sumitomo) supply the substrates Near term story is 800G/1.6T pluggables. CPO inflection will be "late 2027 and beyond" (management guidance). With 2X InP capacity by end of 2026. I.e. demand > supply. And fresh capex in 6-inch InP R&D for next-gen EML/SiPho: Larger diameter = more die per wafer = lower unit cost + is required by some advanced device roadmaps. But ofc, a ton of pricing-in has happened already: Probably looking further ahead to 6-inch InP ramp for 1.6T and probably even CPO (2027+ story). Personally don't see anything wrong with trimming at these levels (I did last week to rotate some gains elsewhere) But keeping some exposure makes sense imo since a ton of capex flows direct to them. So we should see a QoQ improvement in reported earnings as backlog converts etc.
| Horizon | Raw move | vs SPY |
|---|---|---|
| 1 day | -0.7% | -1.0% |
| 1 week | -20.1% | -16.2% |
| 1 month | -40.5% | -40.1% |
| 3 months | — | — |
| 6 months | — | — |