Lightmatter and Qualcomm Collaborate to Develop 1.6 Tbps/Fiber CPO Chiplet, Resolving AI Training’s Electrical Bottleneck

August 9, 2026

alchedek USA

Overview

Lightmatter, in strategic collaboration with Qualcomm Technologies, has achieved 1.6 Tbps/fiber unidirectional throughput with its Passage L-Series Co-Packaged Optics (CPO) chiplet. This breakthrough chiplet leverages a 16-wavelength DWDM architecture, delivering approximately 8 times the bandwidth of existing Near-Packaged Optics (NPO) and CPO solutions. This marks a critical advancement in Lightmatter’s mission to address the electrical bottleneck in AI training through innovative optical interconnects.

In Depth

Key Findings

Lightmatter, through its strategic collaboration with Qualcomm Technologies, has announced a breakthrough achievement aimed at resolving the electrical bottleneck in AI training. The company has demonstrated an impressive 1.6 Terabits per second (Tbps) per fiber unidirectional throughput with its Passage L-Series Co-Packaged Optics (CPO) chiplet. This innovative chiplet incorporates a 16-wavelength DWDM (Dense Wavelength Division Multiplexing) architecture, delivering approximately 8 times the bandwidth compared to existing Near-Packaged Optics (NPO) and CPO solutions. This achievement represents a significant step towards realizing high-speed and power-efficient optical interconnect technology essential for accommodating the exponential growth in AI workloads.

Technical & Business Details

Performance of Passage L-Series CPO Chiplet: Lightmatter’s Passage L-Series CPO chiplet minimizes the electrical signal path by integrating the optical engine in very close proximity to the processor or switch ASIC. This integration significantly improves signal integrity and drastically reduces power consumption. The 1.6 Tbps/fiber unidirectional throughput is critical performance for eliminating data movement bottlenecks between chips and boards, particularly in AI model training and inference.
16-Wavelength DWDM Architecture: The 16-wavelength DWDM architecture employed in this chiplet allows for the simultaneous transmission of multiple optical signals at different wavelengths over a single optical fiber. This dramatically increases the effective data transmission capacity, enabling ultra-high-bandwidth connectivity while limiting the increase in physical fiber count. This contributes to optimizing space and cost within data centers and AI factories.
Collaboration with Qualcomm Technologies: The strategic cooperation with Qualcomm Technologies facilitates the integration of Qualcomm’s deep expertise in mobile and computing with Lightmatter’s optical computing and photonics technology. This partnership is a crucial step towards embedding optical technology into next-generation AI chipsets and networking solutions, potentially accelerating broader market adoption.

Background & Context

The increasing complexity of AI models and the explosion of data volumes are exposing the limitations of current electrical interconnect technologies. Specifically, data movement between GPUs and AI accelerators leads to poor power efficiency and increased latency, which are major bottlenecks restricting the overall performance of AI systems. To address this challenge, CPO technology, which integrates optical engines into chip packages, has gained significant attention, with leading semiconductor companies and startups engaged in intense development competition. Lightmatter’s achievement underscores the superiority of optical technology in high-performance computing for the AI era.

Strategic Significance & Outlook

The success of Lightmatter’s Passage L-Series CPO chiplet, in collaboration with Qualcomm Technologies, holds the potential to significantly transform the future of optical interconnects in AI infrastructure. An eightfold increase in bandwidth compared to existing solutions will dramatically accelerate AI training speeds, enabling the development and deployment of larger and more complex AI models. This technology is expected to have a major impact on data center design, power management, and operational costs, and further commercialization and adoption are anticipated as a key technological innovation for realizing a sustainable and high-performance AI ecosystem.

Source: https://www.alchedek.com/news/articleView.html?idxno=751

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.