PHOTONIC CHIPS FOR AI

Photonic interconnects that scale AI supercomputers.

Interconnects Built for AI Scale

Lightmatter's photonic chips form a complete interconnect platform. Passage™ interconnects and Guide® light engines scale networking for AI supercomputers.

Photonic Interconnect

Passage

Complete photonics roadmap from NPO and OBO to 2D and 3D CPO and 3D Interposers. Edgeless I/O architecture enabling unlimited bandwidth scaling.

Explore Passage →

VLSP Light Engine

Guide

Very Large Scale Photonics technology. The universal light source designed for the entire photonics industry.

Explore Guide →

Lightmatter's 8X Leap: World-First 16-Wavelength Bidirectional Optical Link for AI Data Centers

World-first 16-wavelength bidirectional optical DWDM Link on a strand of standard single mode fiber. This Lightmatter breakthrough represents an 8X leap in bidirectional fiber bandwidth density and a doubling of radix, paving the way for the next generation of AI data centers.

Passage M1000: Powering the Next 1000x in AI Performance

Explore the Passage M1000 Reference System, a groundbreaking platform built around the Passage M1000 3D Photonic Superchip. Discover how this revolutionary photonic interconnect technology is changing the game for Artificial Intelligence (AI) infrastructure by enabling massive scale-up bandwidth and radix, connecting GPUs, TPUs and data center switches in the largest AI model training clusters.

Lightmatter InterConnect Launch Event at OFC 2025

Lightmatter Founder and CEO Nick Harris Unveils Passage L200 3D Co-Packaged Optics and Passage M1000 3D Photonic Superchip

AI scaling is constrained by electrical interconnect

Electrical I/O has reached its limits

Training frontier AI models requires moving petabytes of data between GPUs. Current electrical SerDes technology is fundamentally bandwidth and reach limited—chips are running out of edge space for I/O and scale-up domains are confined to a single rack.

The result: Powerful GPU clusters sit idle waiting for data.

Infrastructure can’t keep up

Model parameters have grown 240x in 3 years. Cluster sizes have grown 10x. But interconnect bandwidth has only improved 2x.

The gap is widening: Without a breakthrough in interconnect technology, next-generation AI training becomes economically infeasible.

Passage is built by Lightmatter to solve the hardest problems in AI

01

Frontier Training

100,000+ GPU clusters for trillion-parameter models

02

Multi-Modal AI

Video, image, and audio foundation model training

03

Reasoning Models

Scalable tensor parallelism for inference

04

Inference at Scale

Serving foundation models to billions of users

Production Ready at Scale

Valuation (Oct 2024)

$4.4B

Total Funding

$850M

Production Deployment

2025

Manufacturing Partners

TSMC GlobalFoundries Tower Semiconductor Amkor ASE

Standards Bodies

OIF IEEE Advanced Photonics Coalition UALink Ultra Ethernet OCP Jedec UCIe Consortium

Ready to eliminate 
your AI infrastructure bottlenecks?

Connect with our team to see how Lightmatter can transform your AI networking and compute architecture.

Contact sales