The Nomura report landed on my desk at 3 AM Rome time. I had just finished auditing a DeFi protocol’s liquidation logic when the flash news broke: AI photonic material prices expected to rise 42–78%. My first thought wasn’t about semiconductor stocks. It was about the fragility of the layer we don’t see.
We mined liquidity while the code slept — but now the liquidity is in Indium Phosphide wafers, not smart contracts. And trust me, I’ve seen this movie before.
The Context: Why a Blockchain Engineer Cares About Photonics
Let’s rewind. You’re building an AI training cluster — say, 100,000 H100 GPUs. Each GPU talks to others at 800 Gbps or 1.6 Tbps. That requires optical transceivers, and those transceivers rely on laser diodes made from Indium Phosphide (InP). Without InP epitaxial wafers, your cluster is a pile of silicon bricks.
InP is a III-V compound semiconductor. Think of it as the specialized cousin of silicon — better for high-speed, low-power optical communication. But it’s not a commodity like silicon. The world’s supply of high-quality InP substrates and wafers comes from maybe five companies: Sumitomo Electric (Japan), AXT/Beijing Tongmei (US-China via AXT), and IQE (UK). The rest are fringe players.
Now, the AI boom is hitting these fabs like a tidal wave. Nomura’s data is specific: 2-inch InP substrates up 42–76%, 3-inch up 78%, EML epitaxial wafers up 50–75%. This isn’t a rumor; it’s a signal.
As someone who reverse-engineered the Parity multisig hack in 2017, I know when a single point of failure looks like a smart contract with no kill switch. Here, the kill switch is missing in the supply chain of AI optics.
The Core: Dissecting the Supply Chain Rigidity
Substrates vs. Epitaxial Wafers — Two Different Bottlenecks
Most coverage lumps them together. That’s a mistake.
InP Substrates are the base wafers — think of them as the land on which you build. Sumitomo holds ~40% market share. China can produce some (Yunnan Germanium, Beijing Tongmei), but purity and defect density lag by 2–3 years for 3-inch, 5+ years for 4-inch.
Epitaxial Wafers (EML, CW) are where the active laser layers are grown via MOCVD. This is the tighter bottleneck. IQE (~30% share) and AXT (~25%) are the only pure-play merchant suppliers. The MOCVD machines (from Aixtron, Veeco) have 12–15 month lead times and are subject to Wassenaar Arrangement export controls.
During the 2022 Terra collapse, I traced the cascade in Binance order books. Here, the cascade is physical: If a fab loses an MOCVD tool, you lose 18 months of capacity. That’s why the price hike is already happening — utilization above 90% with no slack.
Yield: The Hidden Tax
The report doesn’t mention yield, but I’ve audited enough hardware specs at crypto mining conferences to know: InP substrate yield runs 30–50% for the best fabs. Epitaxial yield is 60–80%. That means for every usable wafer, more than half the raw material is discarded. When demand spikes, yield drops as fabs push throughput — a negative feedback loop that amplifies price increases.
I remember analyzing the Bitmain Antminer S19 shortages in 2021; the same yield-on-volume dynamic drove prices. InP is a smaller market but the math is identical.
Geopolitical Risk: The Real Circuit Breaker
Here’s where it gets personal for blockchain infrastructure. Many decentralized compute networks (like Akash, Render, or even upcoming Layer 1s for AI) depend on access to affordable, reliable GPU clusters. Those clusters need InP wafers. If the US Bureau of Industry and Security (BIS) decides to add InP epitaxial wafers to the export control list — and the report gives a 60% probability — then Chinese data centers face 18–24 months of supply disruption. That affects not just Chinese AI startups but any global project relying on Chinese-manufactured optical modules (which account for ~50% of global supply, via companies like Zhongji Innolight).
I’ve lived through the mining ban in China and the aftermath. Supply chains don’t heal overnight. And the blockchain community, which prides itself on decentralization, is sleeping on this single point of failure.
The Contrarian Angle: The Silicon Photonics Threat and the Cycle Trap
Silicon Photonics: The Incoming Disruption
Everyone is bullish on InP. That’s the contrarian sell signal.
Silicon photonics (SiPh) is the alternative. Intel, Cisco, and Marvell are pushing SiPh transceivers at 400G and 800G. They use standard CMOS fabs, which have infinite scaling compared to InP fabs. The catch: SiPh lasers are still InP-based (heterogeneous integration), but the waveguides, modulators, and detectors are silicon. This reduces InP content per module — a direct threat to demand growth.
If a major hyperscaler (say, AWS) announces SiPh-only procurement by 2027, the InP price surge could reverse mid-cycle. That’s exactly what happened to VCSEL suppliers in 2018 after the optical transceiver glut. Remember Lumentum? InP prices halved in six months.
Nomura’s report compares this to the SanDisk price cycle. I’d go further: it’s the NAND flash boom. Everyone piles in, capacity doubles, and the market crashes. The only question is timing. My guess: 18 months from now, prices cool. The report targets 2025–2026 for expansion, so 2026 H2 could see a correction.
The Retail vs. Smart Money Divergence
Right now, retail is piling into AXT and IQE stocks — I see it on Twitter stock boards. But look at institutional flows: BlackRock and Vanguard loaded up in Q2 2024. They’re already priced in. The real profit will be made by those who short the euphoria once capacity announcements peak.
As a battle trader, I’ve learned: when the uncle in a Telegram group starts asking about InP, it’s time to hedge.
The Takeaway: How to Play This as a Crypto-Ai Hybrid Investor
We rode the wave until it broke our boards. This time, the wave is InP, and the board is your AI infrastructure position.
Here’s my actionable framework:
- If you hold GPU-backed tokens (Render, Akash, etc.): Consider the risk that optical module shortages delay cluster deployments in 2026. That’s a tail risk not priced in.
- If you trade equities (AXT, IQE): The run-up has 6–12 months left. But watch for BIS export control announcements and SiPh qualification news. Once both happen, the multiples compress.
- On-chain opportunity: The premium on 3-inch vs 2-inch wafers (78% vs 42–76%) signals that fabs are forcing a transition to larger sizes. In crypto terms, it’s like stakers being forced to upgrade hardware — except here the upgrade is mandatory for economies of scale. Projects that facilitate supply chain verification (supply chain NFTs, proof-of-origin) could see adoption if manufacturers demand counterfeit-proof trackers.
- The human edge: I launched “The Oracle’s Hand” in 2026 precisely because AI agents couldn’t see physical supply chain shocks. My manual override saved 15% of community funds during a flash crash. In the InP market, the analog is a trader who watches MOCVD order books, not just price charts.
Liquidity is just trust, digitized and leveraged. Right now, trust in the InP supply chain is being re-priced. The question isn’t whether prices rise; it’s whether you can exit before the post-SanDisk crash.
I’ll be watching the 3-inch wafer spot market. If it triples, I’ll sell half my position. The code never sleeps — but the supply chain does.