Silicon Photonics and CPO Emerge as Potential Responses to AI Data-Center Copper Constraints
DataNews Editorial Team
An industry report presents silicon photonics (SiPh) and co-packaged optics (CPO) as potential responses to copper interconnect constraints in large AI clusters targeting 800G and 1.6T connectivity. The report attributes to NVIDIA a view that 2026 will mark the commercialisation of silicon photonics, while also claiming the company has invested US$2 billion each in Lumentum and Coherent.
- The report positions SiPh and CPO as potential responses to copper interconnect constraints at 800G and 1.6T.
- AI clusters described at the 100,000-accelerator scale increase the importance of chip-to-chip bandwidth, power and thermal management.
- NVIDIA is reported to view 2026 as a commercialisation year for silicon photonics.
- The report claims NVIDIA invested US$2 billion each in Lumentum and Coherent.
- Taiwanese supply-chain participants 台積電 and 日月光 are cited in connection with SiPh/CPO and advanced packaging.
The next constraint in AI infrastructure may increasingly sit between accelerators rather than within them. According to the industry report, AI clusters approaching the 100,000-accelerator scale are driving sharply higher chip-to-chip data-transfer requirements, exposing the bandwidth, power, reach and cooling limitations of conventional copper interconnects.
The report identifies 800G and 1.6T as the relevant high-speed connectivity targets and presents silicon photonics and co-packaged optics as technologies intended to address those limits. In the report’s framing, SiPh and CPO are potential approaches to improving performance and thermal characteristics relative to copper-based links at higher speeds.
The report attributes to NVIDIA the position that 2026 will be the year of silicon-photonics commercialisation. It also claims that NVIDIA has invested US$2 billion in Lumentum and a further US$2 billion in Coherent. These claims should be treated as reported rather than independently confirmed.
Taiwanese supply-chain participants including TSMC and ASE are identified as relevant to the SiPh/CPO transition and to advanced packaging. This points to the importance of packaging and photonics supply chains as AI-system architecture moves beyond GPUs and other compute silicon.
For data-center operators and AI-infrastructure investors, the potential implication is a shift in the bottleneck from compute availability toward high-speed interconnects and advanced packaging. If CPO reaches commercial deployment on the reported timeline, value could extend beyond accelerators into optical components, photonic platforms and advanced-packaging services. The pace of adoption will depend on manufacturing readiness, supply availability and whether CPO delivers a sufficiently compelling operational and economic case at scale.
As AI clusters scale toward hundreds of thousands of accelerators, interconnect bandwidth, power, reach and thermal management could become a more consequential infrastructure constraint alongside compute. If CPO commercialises on the timeline cited in the report, procurement and investment could increasingly shift toward optical interconnects, photonic platforms and advanced packaging; however, supply-chain maturity and the practical economics of CPO will determine adoption rates.
- TechNews 科技新報 / ChineseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.