Lightmatter is a Photonics & Interconnects company founded in 2017 and based in Boston, USA. It has raised $667M in total funding, most recently a Series D in 2024 at a $4.4B valuation.
| Date | Stage | Amount | Valuation | Lead investors |
|---|---|---|---|---|
| Oct 16, 2024 | Series D | $400M | $4.4B | T. Rowe Price |
| May 1, 2023 | Series C | $154M | — |
| GV |
| May 6, 2021 | Series B | $80M | — | Viking Global Investors |
| Jan 1, 2019 | Series A | $33M | — | GV |

Lightmatter CEO argues optical interconnects are the key to scaling AI systems beyond GPU compute limits.

As AI shifts from training to inference, optical interconnect makers are moving silicon photonics closer to the chip to reduce power and latency.
Lightmatter's flagship 3D photonic interconnect, described as the world's first photonic interposer: a wafer-scale active substrate that stacks compute chiplets on top while routing terabits of data optically across its surface. The M1000 delivers multi-terabit-per-second per-port bandwidth to connect GPUs, TPUs, and switches in the largest AI training clusters, attacking the interconnect bandwidth wall that limits how big AI systems can scale.
A chiplet-style co-packaged optics product that brings optical I/O directly next to an accelerator or switch ASIC, replacing power-hungry electrical SerDes with light. The L200 dramatically raises the bandwidth that can escape a package while cutting energy per bit, enabling denser, more efficient scale-up and scale-out fabrics for AI data centers.
Envise is Lightmatter's photonic compute engine that performs neural-network matrix operations using light for high throughput at low power, paired with Idiom, a software stack that lets standard ML models (PyTorch/TensorFlow) compile down to the photonic hardware. Together they target energy-efficient AI inference, complementing Lightmatter's photonic interconnect roadmap.
35 patents on file, but none with both an extractable figure and an abstract on Google Patents yet.