Silicon photonics for AI — photonic interconnects that move data at the speed of light
HQ: Boston, MA, United States | Founded: 2017 | Employees: ~400-500 (est.) | Stage: Late Stage VC ($4.4B valuation) | Website: https://lightmatter.co
Lightmatter was founded in 2017 by Nick Harris, Darius Bunandar, and Thomas Graham as a spinout from MIT's photonics research, with a contrarian thesis: as transistors stop scaling, the bottleneck in AI data centers shifts to moving data — and light, not copper, is the answer. The company initially pursued photonic AI compute (the Envise processor and Idiom software), then repositioned around interconnect, where the physics advantage is clearest.
Its Passage platform is a family of 3D-stacked silicon-photonic engines that provide fiber-level bandwidth at chip scale: Passage Guide brings optical links into racks for scale-up compute, Passage L200 (announced April 2025) is described as the fastest co-packaged optics engine for AI systems, and Passage M1000 is a photonic superchip unveiled in April 2025 as the foundation of switchless, wafer-scale AI interconnect. The chips are manufactured on GlobalFoundries' Fotonix silicon-photonics process, and in 2025 Lightmatter announced an integration with NVIDIA's NVLink Fusion architecture.
Capital has followed the roadmap: $850M raised across six rounds, including a $155M Series C-2 at a $1.2B valuation (2023) and a $400M Series D at $4.4B (October 2024) with participation from T. Rowe Price, Fidelity, and existing investors GV, Sutter Hill Ventures, and SIP Global Partners. As of mid-2026 Lightmatter remains private, with secondary-market pricing near its last-round valuation.
Lightmatter is the best-funded pure play in silicon photonics, betting that copper interconnects — not compute — are AI's next scaling wall. Its Passage engines (3D-stacked photonic chips co-packaged with XPUs) claim an order-of-magnitude bandwidth jump, and its 2025 roadmap — Passage L200 co-packaged optics, the M1000 photonic superchip, and an NVIDIA NVLink Fusion integration — aims at becoming the optical fabric of AI supercomputers. With $850M raised, a $4.4B valuation, Google as a repeat investor, and manufacturing on GlobalFoundries' silicon-photonics process, it has real silicon and real partners. The open questions are hyperscaler adoption timing, whether NVIDIA co-opts co-packaged optics in-house, and execution on 3D-stacking yields — the classic deep-tech gap between demo and deployment.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki