Ayar Labs has joined the NVIDIA NVLink Fusion ecosystem, with the companies working to make Ayar Labs’ co-packaged optics (CPO) products optically and electrically compatible with NVIDIA optical and SerDes technologies. The goal is to let hyperscalers and system builders design optically connected, rack-scale AI infrastructure around NVLink Fusion.
Ayar Labs is positioning CPO as a way to push high-bandwidth connectivity while reducing the power and distance constraints that show up when systems rely on copper links. The company describes its CPO approach as “putting optics where it matters most” to extend scaling headroom as bandwidth requirements increase and electrical constraints tighten in rack-scale architectures.
For data center engineers, the practical takeaway is that optical links packaged closer to the compute complex are being treated as an architectural option for scale-up fabrics, not just for longer-reach networking. If CPO can deliver on its power-per-bit and latency targets in real deployments, it could change how rack-scale designs budget power for interconnect, and how far designers can stretch bandwidth before signal integrity and thermal limits force compromises.
“By joining the NVIDIA NVLink Fusion ecosystem, we’re introducing co-packaged optics as a foundational building block for customers deploying heterogeneous compute in NVIDIA AI factories,” said Mark Wade, CEO at Ayar Labs.
NVIDIA framed the move as expanding options for heterogeneous systems. “By expanding the NVLink Fusion ecosystem with Ayar Labs’ optical connectivity, innovators can scale bandwidth and bring heterogeneous AI infrastructure to market faster,” said Ashish Karandikar, Vice President at NVIDIA.
The companies said the collaboration is intended to help customers deploy heterogeneous compute, including custom silicon, within NVIDIA rack-scale platforms while preserving NVLink-based architecture investments. They also said Ayar Labs will work with customers and ecosystem partners on CPO integration tied to NVLink Fusion deployments, including system architecture, validation requirements, and platform timelines.
Source: Ayar Labs
