Key Findings
Wiwynn, in collaboration with leading ecosystem partners including Ayar Labs, GUC, and TE Connectivity, will unveil groundbreaking co-packaged optics (CPO) interconnect innovations at Computex 2026. This demonstration is designed to accelerate optical scaling in hyperscale AI data centers, showcasing a comprehensive full-stack solution from chip-level CPO design to practical data center deployment. This advancement promises to deliver ultra-high bandwidth and power efficiency, essential for AI workloads, surpassing the limitations of traditional electrical interconnect technologies.
Technical / Clinical Details
The demonstration will prominently feature Ayar Labs’ TeraPHY optical I/O chiplet combined with its SuperNova multi-wavelength light source. This silicon photonics-based solution directly enables multi-terabit throughput from GPU, CPU, or other accelerator packages, offering 5-10x higher bandwidth, 10x lower latency, and 4-8x better power efficiency compared to conventional copper interconnects. The integration of GUC’s ASIC design expertise and TE Connectivity’s connectivity solutions brings several specific advantages to this CPO technology:
- Dramatic Bandwidth Increase: Achieving multi-terabit-scale bandwidth to meet the demanding data transfer requirements of AI clusters.
- Significant Power Reduction: Substantially cutting power consumption compared to traditional electrical interconnects, lowering operational costs and environmental impact for AI data centers.
- Low Latency: Eliminating signal transmission bottlenecks to enhance data exchange speeds between computing nodes.
- Scalability: Enabling the connection of thousands of GPUs into a single, unified cluster via optical fabric, facilitating the construction of large-scale AI systems.
- Heterogeneous Integration: Ayar Labs’ optical engines are built using standard CMOS manufacturing processes, allowing for seamless integration into existing semiconductor ecosystems.
Background & Context
AI data centers face significant challenges from explosive data traffic growth and GPU communication bottlenecks. NVIDIA has also announced plans for full adoption of silicon photonics and CPO in its AI data center platforms by 2026, solidifying optical interconnects as the next frontier for AI infrastructure. In this context, Wiwynn and its partners’ push for CPO technology underscores photonics’ critical role as a foundational scaling layer for AI-era computing. NVIDIA’s investment of over $6.5 billion in companies like Lumentum, Coherent, Marvell, Corning, and Ayar Labs further highlights the industry-wide commitment to optical interconnect solutions.
Strategic Significance & Outlook
Wiwynn and its ecosystem partners’ CPO technology demonstration provides a tangible vision for the future of AI data centers. This technology is an indispensable element for overcoming the physical limitations of copper interconnects and meeting the demanding requirements of AI workloads. CPO is positioned as the ‘ultimate goal’ for achieving low latency and low power consumption, following stages of pluggable optical modules and near-packaged optics (NPO), with explosive adoption projected between 2026 and 2027. This advancement promises to dramatically enhance AI computing capabilities, suggesting that heterogeneous integration and advanced packaging will become central to next-generation data center design and operation.
Comments