Lightmatter is showcasing the M1000 at the Optical Fiber Conference in San Francisco from April 1-3, 2025, highlighting its potential impact on the industry.

The M1000 includes a reconfigurable waveguide network that integrates 256 optical fibers, delivering significantly higher bandwidth compared to traditional Co-Packaged Optics (CPO).

Lightmatter's innovations come at a time of increasing demand for high-performance computing solutions in AI, as traditional semiconductor scaling reaches its limits.

The Passage M1000, a large reference platform measuring over 4,000 square millimeters, is engineered for next-generation XPUs and switches, accommodating extensive die complexes in a compact 3D package.

The Passage L200, expected to launch in 2026, will feature data transfer rates of 32 Tbps and 64 Tbps, significantly enhancing the training speeds of advanced AI models.

With a total I/O bandwidth exceeding 200 Tbps per chip package, the L200 enables up to eight times faster training for AI models while incorporating edgeless I/O capabilities.

Companies that adopt these optical innovations are poised to lead advancements in AI and reshape the competitive landscape of technology stocks.

The M1000 overcomes traditional bandwidth limitations by allowing electro-optical I/O connections across its entire surface, facilitated by an extensive waveguide network.

Nick Harris, CEO of Lightmatter, emphasized that the Passage M1000 represents a significant advancement in photonics and semiconductor packaging for AI infrastructure.

Lightmatter has unveiled two groundbreaking technologies designed to enhance AI chip connectivity through optical connections and silicon photonics, joining major players like AMD and Nvidia in this innovative space.

This superchip, with an overall package size of 7,735 mm², can provide up to 1,500W of power to its chiplets, aligning with the demands of next-generation AI processors.

Following an $850 million funding round, Lightmatter is valued at $4.4 billion, positioning itself as a leader in the optical computing sector.