A Cross-Market Technology Spillover in Motion
The market has historically viewed FMCW (Frequency-Modulated Continuous-Wave) LiDAR for autonomous vehicles and CPO (Co-Packaged Optics) for AI data centers as two entirely separate technology roadmaps. The former focuses on coherent optical sensing in automotive environments, while the latter serves high-speed transmission inside high-density server racks. As the commercial deployment of L4/L5 autonomous vehicles slows down, some market analysts argue that the substantial R&D capital invested in FMCW LiDAR over recent years is hitting a growth bottleneck.
However, the real area of interest lies in the transfer and repurposing of Photonic Integrated Circuits (PIC) and high-precision optical packaging capabilities across different end markets.
This is not a narrative of one technology saving another, but a classic case of cross-domain technology spillover. As the generative AI compute boom pushes conventional optical communications toward physical limits, CPO faces bottlenecks in manufacturing, light source management, and packaging yield. These challenges closely align with the photonic integration, coherent receiving, and automated alignment experience accumulated during FMCW LiDAR’s development within the automotive supply chain.
Why Now? Soaring Cabinet Power Densities Hit Physical Limits
The recent surge in cross-industry discussions is driven primarily by dual pressures on power consumption and bandwidth within AI infrastructure:
- Power Density and Thermal Walls: As the power consumption of high-end AI compute modules and trays surpasses the kilowatt threshold, overall rack power density—such as NVIDIA’s NVL72 rack reaching approximately 120 kW—continues to climb. Consequently, AI infrastructure faces simultaneous challenges in operational heat dissipation and high-speed interconnect energy consumption.
- Electrical Link Losses: As SerDes rates and switch capacities increase, copper interconnects remain confined to ultra-short links (such as intra-rack short-distance cables). Longer board-level electrical traces encounter higher insertion loss, equalization power consumption, and thermal management pressure. This compels the location of optical-to-electrical conversion to move progressively closer to the switch ASIC.
- The Positioning of CPO: By placing the optical engine inside or directly adjacent to the switch ASIC package, the CPO architecture significantly shortens high-speed electrical transmission distances. It is widely considered one of the key long-term evolutionary paths for high-end switches and high-speed Optical I/O, evolving alongside competing routes such as NPO and LPO.
Where Are the Bottlenecks? Optical Coupling, Light Source Management, and Packaging Yield
In industry practice, the most frequently cited commercialization hurdles for CPO include:
- Optical Coupling and Mass Production Alignment: How to rapidly and reliably couple multi-channel optical fibers to photonic chips with micron to sub-micron tolerances using active or passive alignment, fiber alignment arrays (FAU), and pre-alignment structures.
- Testability & Repairability: Standardization for wafer-level and pre-packaging optical testing remains less mature than that for traditional electronic chips. In many CPO architectures, a failure in the optical engine can degrade the yield of the entire high-value packaged module or force a complete scrap, making repairability a major engineering hurdle.
- Laser Sources and Thermal Management: Silicon photonic chips are highly sensitive to temperature variations. The thermal output of high-power ASICs imposes stringent reliability demands on optical component stability and laser lifespan.
Several of these challenges—specifically PIC integration, light source management, high-precision coupling, and reliability engineering—overlap directly with the miniaturization and chip-level integration process of FMCW LiDAR.
To address the size, cost, and mass-production challenges of automotive LiDAR, companies such as Aeva and SiLC invested heavily in integrating laser control, beam splitting, modulation, and coherent reception onto single or few photonic chips. These developments required establishing reliability engineering suited to harsh automotive conditions involving temperature extremes, mechanical vibration, longevity, and high-volume consistency.
Supply Chain Realignment: Moving from Module Assembly to Heterogeneous Integration
As optical transmission transitions from pluggable modules to substrate-level co-packaging, the core capability of the supply chain is shifting from traditional independent optical module assembly toward semiconductor-level electro-photonic heterogeneous integration:
| Metrics / Domains | Traditional Optical Module Era | CPO Era Critical Requirements | Transferable FMCW LiDAR Experience |
|---|---|---|---|
| Light Source & Engine | Integrated within independent modules | External Laser Sources (ELSFP), High-Power Lasers/SOA, Thermal Isolation, and Repairability | FMCW light sources, SOA, thermal control, and PIC integration experience |
| Testing & Yield | Independent testing and replacement post-module assembly | Known-Good-Die (KGD), pre-packaging optical testing, and system-level repairable design | Electro-photonic co-testing and mass-production consistency engineering |
| Form Factor & Coupling | Placed on the front panel | Engines positioned adjacent to ASIC, requiring ultra-high-density heterogeneous integration | PIC miniaturization and automated optical coupling expertise |
Note: Although CPO and FMCW LiDAR share commonalities at the PIC and packaging levels, their end-use requirements differ. LiDAR operates in free-space optics and range measurement, whereas CPO targets maximum bandwidth and extremely low Bit Error Rates (BER) inside optical fiber channels.
Diverse Approaches and Key Industry Strategies
Key stakeholders across the supply chain are demonstrating distinct strategic postures regarding the evolution of Optical I/O and CPO:
- NVIDIA (System Architecture Push): Advancing commercial CPO through system architectures. NVIDIA introduced the Spectrum-X Photonics and Quantum-X Photonics switch platforms for Ethernet and InfiniBand networks, co-packaging silicon photonic engines with switch ASICs. NVIDIA is building a comprehensive silicon photonics ecosystem spanning foundry manufacturing, advanced packaging, external lasers, optical fibers, and connectors through partnerships with TSMC, Coherent, Lumentum, Corning, Foxconn, and SENKO.
- Broadcom (High-Speed Ethernet & PAM4 Continuity): Broadcom continues to focus on the parallel evolution of high-speed SerDes, PAM4 DSPs, switch ASICs, and CPO to balance power consumption, cost, and existing data center network ecosystems.
- TSMC COUPE (Foundry Heterogeneous Integration): Utilizing wafer-level or 3D heterogeneous integration of Electronic Integrated Circuits (EIC) and Photonic Integrated Circuits (PIC) to shorten electrical path lengths between electronic and photonic dies, lowering interface loss and footprint. External optical fiber coupling and production alignment still require integration with downstream optical packaging processes.
- FMCW LiDAR Vendors & Startups (Cross-Market Validation): In early 2026, Aeva announced a high-power semiconductor optical amplifier (SOA), explicitly extending its long-term capabilities in silicon photonics design, wafer fabrication, and photonic integration into the AI data center optical interconnect market—spanning Co-Packaged Optics (CPO) and External Laser Small Form-Factor Pluggable (ELSFP) architectures. This move highlights how photonic technologies developed for FMCW LiDAR are spilling over from sensing applications into high-speed optical communications and AI infrastructure, illustrating the cross-industry utility of silicon photonics platforms.
- OSAT Packaging (Advanced Packaging Execution): Outsourced Semiconductor Assembly and Test (OSAT) leaders such as ASE continue to showcase CPO packaging capabilities. By leveraging specialized pre-alignment structures and automated assembly, they achieve micron to sub-micron coupling tolerances. Their engineering experience accumulated in automotive optics and high-precision packaging serves as an important foundation for supporting commercial CPO production lines.
Key Milestones to Watch for Market Adoption
To objectively determine whether FMCW LiDAR’s technology spillover evolves from isolated business expansions into an industry-wide supply chain transition, two future milestones warrant monitoring:
- Milestone 1: Scaled Validation of External Lasers (ELSFP): If ELSFP modules complete large-scale deployment validation covering high power output, thermal management, hot-pluggability, and long-term reliability, it will indicate that CPO has cleared major hurdles in light source repairability and thermal isolation. If suppliers attribute their component platforms or reliability methodologies to past FMCW LiDAR R&D, it will serve as direct evidence of technology spillover.
- Milestone 2: High-Volume Production Signals from OSAT Leaders: A critical signal will be whether capital expenditures by major OSAT providers like ASE for CPO automated alignment and optical testing lines shift from R&D trial runs to commercial mass production. Concurrent improvements in yields, capacity utilization, and unit packaging costs will be essential to demonstrate the economic viability of optical packaging.
Unlocking Cross-Market Value from R&D Capital
While the market has generally treated FMCW LiDAR and CPO as independent sectors, recent moves by companies like Aeva indicate that light source, coherent optical, and photonic integration capabilities developed for FMCW LiDAR are beginning to spill over into AI optical interconnects.
Substantial R&D investments in the semiconductor and photonics industries rarely vanish due to timeline adjustments in a single end market. The technical assets accumulated by FMCW LiDAR developers over years of meeting stringent automotive requirements are relocating to commercial opportunities within the generative AI compute wave. Driven by the cross-industry movement of talent, component platforms, and packaging know-how, these capabilities are emerging as a notable contributor to the evolution of next-generation Optical I/O.
SELECTED REFERENCES
- Aeva Technologies, Inc. (2026, January 15). Aeva Unveils Industry-Leading High-Power Semiconductor Optical Amplifier for AI Data Center and Physical AI Applications. Aeva Investor Relations.
- NVIDIA Corporation (2025). NVIDIA Announces Spectrum-X Photonics and Quantum-X Photonics Networking Switches to Scale AI Factories to Millions of GPUs. NVIDIA Investor Relations.
- NVIDIA Corporation (2025–2026). NVIDIA Spectrum-X Ethernet Technical Whitepaper. NVIDIA.
- NVIDIA Corporation (2025–2026). NVIDIA Spectrum-X Ethernet Datasheet. NVIDIA.
- ASE Technology Holding Co., Ltd. (2025–2026). Silicon Photonics.
- TSMC Research Team. (2024). High Bandwidth and Energy Efficient Electrical-Optical System Integration Using COUPE Technology. Proceedings of the IEEE Electronic Components and Technology Conference (ECTC).
- Optical Internetworking Forum (OIF). (2025). OIF-ELSFP-02.0: External Laser Small Form Factor Pluggable (ELSFP) Implementation Agreement. OIF Implementation Agreement.