Lightmatter has joined the NVIDIA NVLink Fusion ecosystem, linking the photonics company to NVIDIA's expanding AI infrastructure network.
Under the arrangement, Lightmatter will supply co-packaged optics and near-packaged optics products designed to work with NVIDIA optical and SerDes technologies. It is adapting its bi-directional optical link architecture to NVIDIA's optical and electrical systems to support semi-custom AI deployments.
The tie-up focuses on connections between specialised processors and NVIDIA switch silicon. Lightmatter said its products will enable semi-custom XPUs from different suppliers in the NVLink Fusion ecosystem to connect through a common optical interconnect platform.
It added that the design would cut fibre and connector requirements by 50%, helping simplify physical network layouts inside large AI systems. That matters as operators build denser AI clusters and try to manage the growing complexity of moving data between processors, switches, and memory resources.
Optical links
Co-packaged optics and near-packaged optics are drawing close attention across the data centre sector as chipmakers and infrastructure suppliers look for alternatives to conventional electrical interconnects. The technology places optical connectivity closer to switching and compute silicon to ease bandwidth constraints and improve power efficiency as AI systems grow.
Lightmatter has focused its business on silicon photonics for AI data centres. Its product roadmap includes the Passage platform and the Guide light engine, which it positions as tools for linking very large numbers of processors inside advanced computing systems.
NVIDIA has been expanding NVLink Fusion as it pushes its interconnect technology deeper into the market for custom and semi-custom AI systems. The ecosystem approach allows companies outside NVIDIA's own processor portfolio to build infrastructure that still connects to NVIDIA switching technology.
That creates an option for cloud operators and other large-scale buyers that want to combine in-house or third-party accelerators with NVIDIA networking. For suppliers such as Lightmatter, inclusion in that framework broadens the range of systems their hardware can address.
Nick Harris, Founder and Chief Executive Officer, Lightmatter, described the partnership as a marker of where AI system design is heading.
"This is what the next era of AI infrastructure looks like. By integrating Passage CPO solutions with NVIDIA's NVLink Fusion architecture, we are combining the industry's most advanced AI platform and the world's leading interconnect to unleash generations of leading frontier AI models," said Harris.
Ashish Karandikar, Vice President of Engineering, NVIDIA, framed the announcement in the context of changes in data centre design as AI workloads spread across more computing environments.
"AI is being fused into every computing platform, requiring fundamentally re-architected data centers. Integrating Lightmatter's advanced photonic engines into the NVLink Fusion ecosystem provides our partners and hyperscale customers with more choice and flexibility to build specialized, energy-efficient AI infrastructure at unprecedented scale," said Karandikar.
Market signal
The announcement also drew attention from industry analysts who track optical networking and AI infrastructure spending. The main commercial question is whether co-packaged optics can move from pilot projects and roadmaps into broader deployment in mainstream AI cluster design.
Alan Weckel, Co-Founder and Analyst, 650 Group, said Lightmatter's addition to the ecosystem pointed to a broader market shift.
"The addition of Lightmatter to the NVLink Fusion ecosystem marks a critical milestone in the maturation of co-packaged optics. By making its Passage 3D photonic roadmap compatible with NVIDIA's high-speed interconnect, Lightmatter is significantly expanding the addressable market for its CPO products. This partnership provides a validated blueprint for hyperscalers to overcome traditional I/O bottlenecks and scale AI clusters to the levels required for next-generation intelligence," said Weckel.
The backdrop is a surge in demand for AI computing infrastructure, which has increased pressure on the links between chips as much as on the chips themselves. As systems expand from racks to clusters containing thousands of processors, moving data quickly and efficiently has become a central engineering and commercial challenge.
Vendors across the sector are now positioning optical interconnects as a practical answer to those constraints. Lightmatter's inclusion in NVLink Fusion shows how that effort is moving closer to the core of mainstream AI system architecture rather than remaining a specialist networking bet.