Hook: The Price Signal That Breaks The Pattern
Serenity just dropped a statement: AI photonic material prices are expected to surge. Nomura's research backs it – 2-inch InP substrates up 42-76%, 3-inch jumping 78%. But this isn't just another semiconductor shortage story. The structural liquidity of the entire AI infrastructure narrative is being rewired at the substrate level. And most analysts are still looking at the wrong charts.
When a single material node – Indium Phosphide – becomes the bottleneck for 800G/1.6T optical modules, the price signal isn't just about supply and demand. It's a liquidity event. The market is pricing in a scarcity that doesn't yet exist in the physical world. The question is: how much of this is structural, and how much is narrative leverage?
Context: The Optical Interconnect Economy
Every DGX GB200 server racks up dozens of 800G modules. Each module needs at least eight EML lasers – each laser starts as an InP epitaxial wafer. The math is brutal: 2023 saw ~2 million EML epi-wafer units. By 2026, that number hits 5 million. But the global supply base? Maybe 5-7 qualified fabs. Sumitomo Electric, AXT/Beijing Tongmei, IQE. That's it.
This isn't a slow-cycle commodity like DRAM. InP substrate growth cycles take 4-6 weeks for a single crystal boule. MOCVD tool delivery is 12-15 months. The capacity elasticity is near zero. And the market is treating this as a transient blip. That's the mispricing.
Core: The Structural Liquidity Trap in Photonic Materials
Let me break down the numbers I've been modeling since early 2024. I built a custom Python script during my 2020 Curve days to track liquidity congestion – this is the same principle applied to physical material flows.
Current utilization rates for InP substrate fabs: 90-95%. Nominal capacity is saturated. The price elasticity of supply is essentially zero in the short term. But here's the kicker: demand elasticity is also low. Optical module makers can't switch to silicon photonics overnight. The switching cost is 12-18 months of re-certification. So we're looking at a 12-18 month window where both sides are locked in.
AXT's gross margin sits at 40-45% today. If 2-inch InP prices rise 70% and costs stay flat, that margin jumps to 55%+. IQE's ~30-35% margin could hit 45%. The market is pricing in this margin expansion, but it's not discounting the volatility. NAND flash cycles taught us that single-material price spikes are mean-reverting within 18 months. The SanDisk analogy embedded in the Nomura report is telling – this is a cyclical boom, not a permanent shift.
But the deeper structural issue is the geographical concentration. 40% of InP substrates come from Japan (Sumitomo). 25% from AXT, a US-China hybrid. IQE holds ~30% of the epi-wafer market from the UK. Any single-point disruption – a US export control update, a Japan trade policy shift, a UK regulatory move – shatters the security assumption. And security is the narrative that's propping up the valuation premiums.
Contrarian: The Real Arbitrage Is in the Narrative Mispricing
Everyone's hunting the price play on AXT and IQE. But the asymmetric bet is on the narrative structure itself. The market is treating InP as an AI infrastructure bottleneck, which implies a premium that should persist. But silicon photonics is the 800-pound gorilla in the room. Intel, Cisco, and Huawei are pushing SiPh for 800G/1.6T. A single major hyperscaler adoption announcement would collapse the InP demand growth from 25% CAGR to below 10%.
I've modeled this out. If SiPh captures 20% of the 800G market by 2026, the InP epi-wafer shortage disappears. Prices revert. And the stocks – currently trading at 30-40x PE – face a 50%+ drawdown. That's the contrarian trade: not shorting InP, but shorting the assumption that the current price structure is stable.
There's also the regulatory arbi. The US is likely to tighten InP export controls to China in H2 2025. That creates a bifurcation: a US/Europe supply chain that trades at a premium vs. a China-dependent supply chain that faces a discount from geopolitical risk. AXT – with its dual US-China footprint – is uniquely positioned for this arbi. But most analysts are ignoring the structural split.
Takeaway: The Next Narrative Shift
Prices rising 78% on 3-inch InP isn't the story. The story is that the narrative of 'AI infrastructure security' is being built on a single, fragile material node. When that narrative cracks – either through SiPh displacement or regulatory disruption – the liquidity in the narrative will drain faster than the physical supply ever could. The alpha was always in the noise, not the hype.
