A Turning Point for High-Speed Optical Materials: The Triple Challenge of InP Substrate Expansion, Rare-Earth Policy and PD Packaging
As AI training scale and data center bandwidth demand soar, materials and packaging technologies for high-speed optical modules face unprecedented challenges. InP substrates, a key material for 200G PDs and high-speed LDs, are reshaping the global supply chain under the pressure of rare-earth policy and capacity transitions. This article takes a deep look at InP wafer makers' expansion strategies, the impact of rare-earth policy on III-V materials, and module maker and PD packaging trends, offering a panoramic view and decision-making insights for the coming 800G/1.6T optical module industry.
🔍 Key Takeaways
From InP substrates to OIO co-packaged optics, a triple challenge is reshaping the high-speed optical communications supply chain
Soaring AI data center demand for 800G/1.6T optical modules is making InP substrates, PD/LD packaging and rare-earth policy the industry's key turning points. This article breaks down:
Sumitomo's expansion strategy: the 4-inch → 6-inch transition timeline and the reshaping of global supply.
The rare-earth policy shock: China controls 70% of indium mining; how could export controls push PD costs up 10x?
Module maker demand dynamics: the different strategies of AWS, Microsoft and Oracle, and the capacity allocation challenge.
Packaging technology trends: how can ELS and PIC+EIC co-design break through the power bottleneck?
Future OIO technology: Micro LED vs. VCSEL, which will become the new standard for GPU interconnect?
1. Sumitomo's InP Substrate Expansion Strategy
1. Capacity Upgrade, Not Wholesale Expansion
Sumitomo's current main production line is xxx
Its expansion strategy leans toward xxx.
This strategy will lead to xxx
2. Upgrade Timeline and Capacity Planning
2024-2025:
2026:
2027:
Total 4-inch capacity will rise from x0k wafers/year to x0k wafers/year.
3. Demand-Driven Strategy
Long-term partners must provide x-year demand forecasts; general partners must provide x-year demand forecasts.
Customers that cannot lock in long-term demand in advance will face supply shortage risk.
4. Comparison with Competitors
JX (Archotech):
AXT:
Sumitomo:
5. Market Impact
2. The Impact of Rare-Earth Policy on III-V Materials
1. III-V Materials' Dependence on Rare Earths
Key materials:
InP (indium phosphide): the core substrate for high-speed optical modules and lasers.
GaAs (gallium arsenide): widely used in high-frequency RF devices and infrared lasers.
The role of rare earths:
Indium (In) and gallium (Ga) are core raw materials, and global supply is highly concentrated in China.
2. China's Rare-Earth Advantage
China controls about xx% of global indium mine output and imposes policy controls on exports.
Although xx and xx hold some mineral sources through xx and xx partnerships, they remain exposed to swings in Chinese policy.
3. Current State (2024-2025)
Policy impact:
Export controls are tightening, especially for high-purity InP/GaAs substrates.
This has caused some US and Taiwanese optical communications makers to suffer x-month delivery delays.
Supply chain adjustments:
International vendors are actively developing non-China sources (xx, xx).
xx is trying to move production lines overseas (xx) to reduce risk.
Cost changes:
InP substrate prices are up xx–xx%, and 200G PD costs have soared to xx times their original level.
4. Future Trends (2026-2030)
Supply diversification:
Sumitomo and JX will expand x-inch capacity and introduce x-inch, reducing dependence on xx.
AXT is expected to keep expanding and serve the xx domestic market.
Policy direction:
Technical responses:
5. Long-Term Impact
Positive:
Negative:
6. Comparison of the Three Major Vendors
Company | Strategy | Capacity Transition Timeline | Price Positioning | Market Position |
Sumitomo | ||||
AXT | ||||
JX (Archotech) |
3. Module Maker Capacity and End-Customer Demand
1. Supply and Expansion Dynamics
Elaser is supplied by both xx and xx:
Supply ratio of roughly 1:2 (xx vs. xx COS/month).
Expansion plan:
Capacity to grow from xx/month to xx/month by year-end (3–xx times growth).
Drivers: 800G optical module demand and 1.6T development.
2. Key Customer Analysis
AWS:
xx's largest customer, with monthly shipments of xx from Taiwan and xx from Houston, US.
Microsoft:
Mainly xx; highly automated, low margin, but takes up a large share of capacity.
Oracle:
Taking a wait-and-see stance on xx modules; once orders land, they will affect capacity allocation.
Centera:
Has an existing allocation of xx/month, which may expand as xx matures.
3. Market Trends and Variables
800G → 1.6T module demand is driving capacity to double.
Diverging customer strategies:
Variables:
4. Packaging and PD/LD Supply Status
1. The PD Supply Bottleneck
200G PD prices have reached xx times their original level, with the market highly concentrated in xx and xx.
2. Advanced Packaging and Module Design
ELS and PIC+EIC integration are the core trends.
Two architectural approaches:
CW laser splitting: one high-power laser → multiple channels.
Distributed EMLs: an independent EML on each layer, offering high flexibility and good fault tolerance.
3. The NVIDIA Case
144-channel MPO → 18 silicon photonics chips → 8 dies each.
A layered, modular design that supports ultra-high 1.6T bandwidth and suits multi-spec assembly.
4. Industry Challenges and Opportunities
Challenges: shortages of high-speed PDs/LDs, packaging yield and cost pressure.
Opportunities: PIC+EIC co-design and ELS packaging bring efficiency and cost advantages for winning the 1.6T generation.
5. Future Technology Focus: From Optical Transceivers to Co-Packaged Optical I/O (OIO)
With the rise of AI computing and hyperscale data centers, demand for GPU-to-GPU connectivity is climbing rapidly. Conventional high-speed copper can no longer keep up with TB/s-class bandwidth and brings power, latency and EMI (electromagnetic interference) bottlenecks. Optical I/O (OIO) is seen as the key next-generation solution, promising to integrate optical transceiver functions directly at the chip or package level and fundamentally change high-speed interconnect architecture.
1. A First Look at Optical In/Out (OIO)
(1) Technical Goals
Directly replace high-speed copper links between GPUs, shortening signal paths and lowering power.
Shift optical signals from conventional board-edge connections (pluggable) to chip-level packaging (chiplet/interposer).
(2) Candidate Technologies Compared
VCSEL Array (vertical-cavity surface-emitting laser array)
Strengths:
Highly mature, with a track record from 25Gbps to 200Gbps.
Can be integrated with standard CMOS processes, keeping packaging cost under control.
Challenges:
High packaging complexity, requiring precise optical coupling and thermal design.
As arrays grow, beam quality and alignment precision become bottlenecks.
Micro LED (micro light-emitting diode)
Strengths:
Low cost and low power; many channels can be stacked over short reach to achieve high aggregate bandwidth.
Its cost structure suits large-scale GPU-to-GPU connectivity.
Challenges:
Per-channel rate is only 1~2Gbps, requiring massive channel parallelism.
Process yield and packaging density control still need breakthroughs.
(3) Market Status
Both technologies are still being explored:
VCSEL Array is used for.
Micro LED focuses on.
2. Future Integration Vision and Timeline
(1) Integration Vision
The ultimate goal is xxxxxx
Typical architecture:
Light source xxxxx
A light-guiding layer etched inside the package or PCB for short-reach optical transmission.
An integrated PD on the receive side, co-packaged with control circuitry.
(2) Technology Roadmap
2026:
2027:
2028-2030:
3. What End Customers Require of Suppliers
(1) Integration Requirements
End customers (such as NVIDIA, AWS and Microsoft) are no longer satisfied with simply buying lasers or PDs; they require:
Light source + packaging + control circuitry as an integrated solution.
Modular designs that can be embedded directly inside GPU or switch ASIC packages.
(2) Performance Metrics
xxx
(3) Reliability and Process Consistency
xxx
(4) Vertical Integration Capability
xxxx
4. Industry Opportunities and Challenges
xxx
Comments