The optical component industry is heading toward a supply crisis that few outside the industry fully appreciate. The explosive growth of AI training and inference infrastructure is driving unprecedented demand for high-speed optical transceivers — and the precision components inside them. But while AI companies can rapidly deploy capital for new data centers, the specialized manufacturing capacity for optical components cannot be expanded at the same pace. A collision is coming.
The Demand Side: AI Changes Everything
The scale of optical component demand driven by AI infrastructure is unlike anything the industry has previously experienced. Consider the numbers:
A single AI training cluster like those being built by hyperscalers can contain 100,000+ GPUs, each connected by high-speed optical links. A 100,000-GPU cluster using 400G optics might require 200,000-400,000 optical transceivers. At 800G, the transceiver count may be lower per link, but each transceiver contains more complex and expensive components.
Microsoft, Google, Meta, Amazon, and Oracle have collectively announced over $300 billion in data center capital expenditure for 2025-2026. Even if optical components represent only 3-5% of total data center cost, this translates to $9-15 billion in optical component demand — a massive increase from the industry's historical run rate.
But raw dollar demand understates the problem. The issue is not just volume — it is the specific types of precision components required:
EML laser chips: Electro-absorption modulated lasers for 800G-DR8 and 1.6T transceivers are manufactured by a handful of companies (Lumentum, Coherent, Broadcom, and a few others) using specialized InP fabrication processes. Expanding EML production requires new epitaxial reactors, wafer fab capacity, and testing equipment with lead times of 18-24 months.
High-speed driver ICs and TIAs: The electronic components that interface with optical devices are manufactured on advanced semiconductor processes (SiGe BiCMOS, 7nm/5nm CMOS) at foundries already strained by AI chip demand.
Precision micro-optics: The lenses, isolators, beam combiners, and waveplates inside each transceiver are manufactured by specialized optical component companies, many of them small firms with limited capacity.
Ceramic and MT ferrules: Every transceiver has fiber connections, and every fiber connection needs ferrules. The ferrule supply chain — dominated by a few Japanese manufacturers — was already operating at high utilization before the AI boom.
The Supply Side: Precision Cannot Be Rushed
The fundamental constraint on optical component supply is that precision manufacturing cannot be rapidly scaled. Unlike semiconductor fabrication, where capacity can be (expensively) added by building new fabs with standardized equipment from ASML, Applied Materials, and others, optical component manufacturing relies heavily on specialized, often custom equipment and deeply accumulated process knowledge.
Consider the challenges of expanding capacity for a few representative component types:
#
Laser Chip Fabrication
InP-based laser chips are manufactured on 2-inch or 3-inch wafers (compared to 300mm/12-inch wafers in silicon fabs), using MOCVD (metal-organic chemical vapor deposition) reactors that cost $2-5 million each and require months to install and qualify. Growing the epitaxial layers for an EML structure requires precise control of dozens of parameters — gas flow rates, temperatures, growth times — accumulated over years of process development. A new MOCVD reactor does not produce good lasers on day one; it may take 6-12 months of process tuning to achieve production-grade yields.
#
Optical Assembly
Transceiver optical sub-assemblies (TOSAs and ROSAs) are assembled using active alignment stations that cost $1-3 million each. Each station includes precision motion stages (nanometer resolution), optical power meters, high-speed electrical test equipment, and UV-cure adhesive dispensing systems. The stations must be operated by trained technicians in clean room environments. Even when equipment is available, training new alignment operators takes months.
#
Testing Bottleneck
Every optical component and sub-assembly must be tested, and high-speed optical testing equipment is itself in short supply. Bit error rate testers (BERTs) capable of characterizing 100+ Gbaud signals cost $500,000-$1 million and have lead times stretching to 6 months or more. Without adequate test capacity, manufacturers cannot validate their products, creating a bottleneck that is often overlooked in supply chain analyses.
The Concentration Risk
The optical component supply chain is dangerously concentrated:
- Ferrules: ~70% from Japan (Kyocera, Adamant Namiki) - InP laser chips: ~80% from three companies (Lumentum, Coherent, Broadcom) - Specialty micro-optics: Often single-sourced from small companies in China, Japan, or Germany - Precision polishing equipment: Dominated by two or three manufacturers globally
This concentration means that a disruption at any single supplier — whether from natural disaster, geopolitical events, or simply inability to scale fast enough — ripples through the entire industry. The 2011 Thailand floods and the 2021 semiconductor shortage demonstrated how concentrated supply chains amplify disruptions; the optical component industry is no less vulnerable.
What Industry Leaders Are Doing
The most forward-looking companies in the optical supply chain are taking action:
Vertical integration: Major transceiver companies like Coherent and Lumentum are investing in upstream component capacity (laser chips, micro-optics) to secure supply and capture more value.
Geographic diversification: Some companies are establishing manufacturing in multiple regions — maintaining facilities in both Asia and North America — to reduce single-point-of-failure risk.
Capacity investment: Several major component suppliers have announced capacity expansions totaling billions of dollars. Coherent is expanding InP and silicon photonics capacity. Lumentum is investing in new laser chip fabrication. US Conec is adding MT ferrule production lines. But these investments take 18-36 months to yield production output.
Design simplification: System designers are working to reduce the number of precision optical components per transceiver. Silicon photonics approaches that integrate multiple functions on a single chip could reduce the BOM (bill of materials) complexity, but silicon photonics has its own capacity constraints.
The Timeline
The most acute supply constraints are likely to emerge in late 2026 through 2028, as the current wave of AI data center construction reaches the phase where optical interconnects are installed. Transceiver lead times, which were already stretching in early 2026, will likely extend further. Component prices — particularly for laser chips and precision micro-optics — will rise.
For companies planning AI infrastructure deployments, the implication is clear: secure optical component supply commitments early. For component manufacturers, the opportunity is significant but demands careful capital allocation — investing in capacity that will be needed in 18-24 months, while managing the risk that demand forecasts may shift.
The precision optical component shortage is not a prediction — it is a mathematical consequence of demand trajectories that are already in motion colliding with manufacturing realities that cannot be wished away. The companies that recognized this early and invested accordingly will be the ones that thrive.