Lightmatter CEO: AI’s Next Scaling Challenge Is Interconnect, Not Compute

2026-06-26 · Source: Big Data & AI News - EE Times · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Cloud Computing & IT Infrastructure · Depth: Advanced, extended

Summary

Lightmatter CEO Nick Harris highlights silicon photonics as the critical solution for AI's escalating interconnect challenges, replacing traditional copper. Founded in 2017, Lightmatter initially focused on photonic compute but pivoted to interconnects as AI workloads, particularly large Transformer-based LLMs, became memory and interconnect-bound rather than compute-bound. Copper's physical limitations, such as a 1-meter range and immense weight in racks, are driving industry leaders like NVIDIA towards Co-Packaged Optics (CPO). Lightmatter's Passage platform offers Near Package Optics (NPO), CPO, and advanced photonic interposers, holding 400 patents and enabling I/O scaling with area rather than just perimeter. The company also introduced Guide, an integrated laser module capable of up to 512 lasers on a 2cm x 2cm die, powering tens to hundreds of terabits per second. GuideOne (128 lasers, 50 Tbps IO) and GuideDR are designed for DWDM and 224 gig systems, respectively, offering features like self-healing and a new liquid-cooled Laser Network Interface Card form factor for OCP racks, reducing server faceplate space by 4x.

Key takeaway

For AI Architects designing next-generation data centers, recognize that interconnect, not compute, is the primary scaling bottleneck for large LLMs. Your strategy must prioritize silicon photonics solutions like Near Package Optics (NPO) and Co-Packaged Optics (CPO) to achieve petabit-scale bandwidth and coherent computing domains. Consider integrated laser modules and OCP-compatible liquid-cooled form factors to enhance system reliability, power efficiency, and serviceability, ensuring your infrastructure can support future AI demands.

Key insights

AI's scaling challenge has shifted from compute to interconnect, making silicon photonics essential for future data center infrastructure.

Principles

• Interconnect bandwidth dictates AI workload partitioning.
• IO scaling with area ($L^2$) surpasses perimeter ($L$).
• Integrated photonics enhance reliability and efficiency.

Method

Lightmatter integrates up to 512 lasers on a single 2cm x 2cm chip using 300mm CMOS foundry processes, enabling high-bandwidth optical interconnects and co-packaged optics (CPO) with features like self-healing.

In practice

• Adopt NPO/CPO for GPU/switch interconnects.
• Utilize integrated laser modules for power efficiency.
• Implement detachable fiber solutions for serviceability.

Topics

• Silicon Photonics
• AI Infrastructure
• Optical Interconnects
• Co-Packaged Optics
• Integrated Lasers
• Data Center Networking

Best for: CTO, VP of Engineering/Data, Investor, AI Hardware Engineer, AI Architect, Director of AI/ML

Related on AIssential

• This Will Power Everything — Optical interconnects are critical for scaling AI data centers beyond copper's physical and thermal limitations.
• Scaling Power-Efficient AI Factories with NVIDIA Spectrum-X Ethernet Photonics — Co-packaged optics in Ethernet switches significantly enhance AI factory scalability, power efficiency, and network resilience.
• Photonics: A Foundational Scaling Layer for AI-Era Computing — Photonics is becoming the foundational layer for scaling AI computing by addressing system-level data movement and memory bottlenecks.
• Coherent Breaks Ground on Expanded Texas Facility, Scaling AI’s Optical Backbone — Optical interconnects, leveraging 6-inch indium phosphide wafers, are fundamental for scaling high-performance AI infrastructure efficiently.
• Why XPO Resonated at OFC 2026 — XPO offers a scalable, power-efficient, and high-density optical interconnect solution for next-generation AI infrastructure.
• [Server] The Bottleneck of Compute: How Humanity Weaves the Soul of Memory with Silicon — HBM overcomes the "Memory Wall" by prioritizing wide parallel data paths and 3D stacking over frequency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.