Lightmatter

Type: Company (Photonic AI Infrastructure)

Lightmatter is the leading commercial player in photonic interconnects for AI data centers. The company's Passage product line implements co-packaged optics (CPO) that places optical transceivers directly adjacent to — or beneath — compute dies, solving the bandwidth bottleneck that limits distributed AI training at scale.

Lightmatter's Passage portfolio now spans two integration topologies. The Passage L200 is a co-packaged-optics module that achieved a record 1.6 Tbps per single fiber and over 200 Tbps per package, enabling fully non-blocking communication between AI accelerators (Lightmatter claims up to 8x faster AI model training from eliminating distributed-training communication bottlenecks). The Passage M1000 goes further architecturally: it is a >4,000 mm² multi-reticle active photonic interposer — described as the world's first 3D-stacked silicon photonics engine — that the compute die complex stacks onto, delivering 114 Tbps total optical bandwidth across 256 fibers (448 Gbps/fiber, 8-wavelength WDM, 1,024 SerDes, 1.5 kW delivery) with the first on-package solid-state optical circuit switching. The M1000 is built on GlobalFoundries' GF Fotonix platform and packaged by Amkor.

Commercially, the company is moving from demonstration to a hyperscaler supply path. Its January 2026 partnership with GUC (Global Unichip Corp) co-designs Passage 3D CPO into GUC's ASIC and advanced-packaging flow for AI hyperscalers and HPC — the interconnect analogue of Ayar Labs' route through TSMC COUPE. An earlier partnership with Qualcomm signals expansion beyond specialized AI infrastructure. Lightmatter is also developing Envise, a photonic compute chip that performs matrix multiplications optically — positioning the company across both the interconnect and compute layers of the photonic stack.

Key Contributions

• Passage L200: Record 1.6 Tbps/fiber, 200+ Tbps/package CPO (Lightmatter Passage L200)
• Passage M1000: 114 Tbps total optical bandwidth, 256 fibers, >4,000 mm² 3D active photonic interposer, world-first on-package solid-state optical circuit switching (GF Fotonix + Amkor) (Lightmatter Passage M1000)
• Up to 8x faster AI training through reduced communication bottlenecks (Lightmatter Passage L200)
• GUC partnership (Jan 2026) — commercial Passage 3D CPO for AI hyperscalers, scale-up across multiple racks (Lightmatter–GUC Partnership)
• Qualcomm partnership expanding CPO beyond AI niche (Lightmatter Passage L200)
• Envise photonic compute chip for optical matrix multiplication (Lightmatter Passage L200)

Mentioned In

• Photonic Interconnects — Primary commercial implementation
• Co-Packaged Optics — Active-interposer CPO topology
• Photonic Neural Networks — Envise photonic compute chip
• Photonic Tensor Cores — Compute architecture context

Related Entities

• GUC (Global Unichip Corp) — Commercial 3D-CPO ASIC/packaging partner
• Ayar Labs — Competing CPO approach (chiplet-style via TSMC COUPE)
• TSMC (Photonics) — Foundry path for the competing chiplet-CPO ecosystem

Changelog

• 2026-06-24 — Added the Passage M1000 active-photonic-interposer (114 Tbps / 256 fibers / on-package optical circuit switching) and the Lightmatter–GUC commercialization partnership; expanded the portfolio framing to two CPO topologies (module vs. active interposer). 1 → 3 sources.