Lightmatter is a photonic computing company founded in 2017 by Nicholas Harris, Darius Bunandar, and Thomas Graham, headquartered in Mountain View, California. The company builds optical interconnect and photonic computing technology designed to overcome the bandwidth and energy bottlenecks that constrain large-scale AI training and inference. Its core thesis is that traditional electrical input/output between chips has reached physical limits, and that light-based data movement is required to scale frontier AI clusters.
The company's flagship technology is the Passage family of photonic interconnects, including the L200 and L20, which deliver multi-terabit-per-second bandwidth to connect GPUs and accelerators at data-center scale. Lightmatter also develops the Guide light engine, a universal light source for the broader photonics ecosystem. Together these products target the wiring layer of AI infrastructure rather than the compute cores themselves.
Lightmatter has raised approximately $850 million in total funding and was valued at around $4.4 billion in late 2024, placing it among the most heavily capitalized hardware startups in the AI infrastructure space. Investors and partners are drawn to the prospect of photonics becoming a standard layer in next-generation data centers.
The primary customers are hyperscalers, AI labs, and data-center operators building large training clusters where interconnect bandwidth is the binding constraint. Because the technology operates at the silicon photonics and packaging level, adoption typically involves deep co-engineering with chip and system vendors.
Lightmatter occupies a frontier, capital-intensive segment where timelines are long and commercial deployment depends on the broader AI hardware roadmap. It is best understood as deep-tech infrastructure rather than a turnkey product, and its trajectory is closely tied to the continued scaling of frontier model training.