A UK semiconductor company lost a quarter of its value in a single day this week. The market assumed Huawei exposure was catastrophic. The Huawei exposure is under five percent of revenue.
The panic was about the wrong thing.
Where Indium Phosphide Fits in the AI Stack
Every 1.6T optical transceiver shipping into an AI cluster relies on lasers made from indium phosphide (InP). Silicon cannot lase. InP does.
The supply chain:
Raw indium (zinc byproduct)
→ InP boule → sliced into substrates (AXT Inc)
→ Epitaxial wafer growth (IQE)
→ Laser fabrication (Lumentum, Coherent, MACOM)
→ Transceiver assembly
→ AI GPU clusters
IQE sits at the third step. It is the only independent large-scale epiwafer foundry with qualified InP capacity globally.
Why Switching Is Hard
Qualifying a new epiwafer supplier takes 12-24 months. Sample testing, device fabrication trials, reliability qualification. Once qualified, switching costs are enormous. There are approximately five qualified InP epiwafer suppliers worldwide. IQE is the only independent one at scale.
The MACOM Signal
In April 2026, MACOM invested £81M into IQE at 19.8p per share. MACOM received 11.5% ownership and two board seats, plus a long-term strategic supply agreement. The message: they cannot secure InP capacity without IQE.
Why Silicon Photonics Does Not Kill This
The common objection: silicon photonics will replace InP. This misunderstands the technology. Even silicon photonic ICs need InP lasers, integrated heterogeneously. Monolithic lasers-on-silicon are not expected before 2030. InP is needed regardless of which platform wins.
The Tension
IQE is not a clean story. Gross margins were under 4% last year on £118M revenue because reactors run at roughly half capacity. The wireless segment (57% of revenue) is declining. Three CEOs in five years. The current CEO also serves as CFO.
But the underlying structural thesis is clean: the InP bottleneck is real and tightening. AXT Inc is doubling capacity for 2027. Lumentum reported 90% revenue growth. The optical interconnect buildout for AI is a capital expenditure cycle backed by the largest technology companies in the world.
The pattern has historical precedent. As Marc Levinson documents in The Box, the shipping container created a modular interface that collapsed freight costs — the same structural pattern happening now in optical interconnects.
The market panicked about Huawei and ignored the structural story underneath. That is the pattern: value migrates upward as lower layers commoditise, and the market is slow to update on where the new bottleneck sits.
Originally published on Telegraph
Top comments (0)