Lightmatter Achieves 1.6 Tbps Fiber Speed, Challenging AI Interconnect Limits

Event summary

  • Lightmatter demonstrated a Passage™ Co-Packaged Optics (CPO) chiplet achieving 1.6 Tbps throughput per fiber, an 8x increase over existing solutions.
  • The technology combines Lightmatter's photonic engine with Qualcomm Technologies’ 112G PAM4 optical SerDes chiplet.
  • The architecture utilizes a 16-wavelength DWDM (dense wavelength division multiplexing) at 112G per SerDes lane.
  • Evaluation kits are now available for lead customer testing, with plans to deliver 100 Tbps and beyond for next-generation XPUs and switches.
  • The announcement follows Lightmatter’s initiative with Alphawave Semi, now part of Qualcomm Technologies.

The big picture

The announcement highlights the growing bottleneck of interconnect bandwidth in AI data centers, a challenge that is increasingly limiting the scaling of frontier AI models. Lightmatter’s solution, leveraging co-packaged optics and a dense wavelength division multiplexing architecture, represents a significant step towards overcoming this limitation and enabling the next generation of AI infrastructure. The collaboration with Qualcomm underscores the increasing importance of photonic interconnects in high-performance computing.

What we're watching

Scale Limits
The ability of Lightmatter’s CPO architecture to sustain its performance advantage as AI models and cluster sizes continue to expand will be a key determinant of its long-term success.
Competitive Response
How Qualcomm and other SerDes providers will integrate photonic interconnects into their offerings will shape the competitive landscape and influence Lightmatter’s market share.
Adoption Rate
The pace at which hyperscalers adopt Lightmatter's Passage L-Series technology, beyond initial evaluation kits, will dictate the company’s revenue trajectory and overall impact on the AI infrastructure market.

Related topics

Semiconductors