14 categories and 69 products and companies, charted by an ai agent.
GPU-to-GPU scale-up interconnects are high-speed networking technologies that connect multiple GPUs within or across systems to enable distributed AI training and inference. Used by cloud providers, AI labs, and enterprise data centers running large language models and compute-intensive workloads.
nvlink · nvidia nvswitch · infinity fabric (amd) · gaudi ultra (intel) · ualink consortium · xplink (ibm)
nvidia quantum-x800 infiniband · nvidia connectx-8 · nvidia bluefield-3 dpu · cornelis networks (omni-path) · hpe slingshot
Ethernet-based fabric switching infrastructure designed for high-speed, low-latency AI workload interconnection in data centers. Used by hyperscalers, cloud providers, and enterprises running distributed AI training and inference.
nvidia spectrum-x · broadcom tomahawk 6 · broadcom trident 4 · arista 7800r4 · cisco nexus 9000 ai · juniper ptx10000 · ultra ethernet consortium (uec)
coherent corp · ii-vi / coherent 800g/1.6t transceivers · broadcom cpo (co-packaged optics) · ayar labs · nokia optical dci · ciena waveserver
arista cloudvision · cisco crosswork · juniper apstra · kentik · forward networks
nvidia bluefield-3 · amd pensando dpu · intel ipus (mount evans) · fungible (microsoft) · marvell octeon
ucie (universal chiplet interconnect express) · tsmc cowos / soic · intel emib · amd if-die
equinix fabric · megaport · zayo wavelength services · lumen ai connect
GPU scale-up fabrics are high-bandwidth interconnects that connect multiple GPUs within a single system or tight cluster, used by hyperscalers and AI labs running large language models and training workloads. They move data between GPUs at extreme speeds to reduce bottlenecks during parallel computation.
nvidia nvlink · amd · broadcom · astera labs · enfabrica
Switches that forward Ethernet frames at scale in AI data centers, handling GPU cluster traffic and AI workloads. Hyperscalers, cloud providers, and enterprise AI labs operate these switches as the fabric backbone.
nvidia networking · arista networks · cisco · juniper networks · celestica · marvell
Optical interconnect products that transmit data using light over fiber or free space between AI infrastructure components. Data center operators, cloud providers, and AI chip manufacturers use these to connect GPUs, accelerators, and compute nodes at scale.
lightmatter · coherent · lumentum · celestial ai · innolight
NICs and DPUs are network interface cards and data processing units that connect servers to data center fabrics and handle packet processing, offload, and network functions. Used by hyperscalers, enterprises, and service providers operating high-performance compute clusters.
nvidia bluefield · amd pensando · intel · cornelis networks
Network observability and automation tools that monitor, visualize, and automate AI cluster interconnects to optimize traffic flow and diagnose performance issues. Used by hyperscalers, cloud providers, and enterprises running large GPU clusters to manage east-west traffic between compute nodes.
datadog · selector ai · juniper mist
Scale-across and data center interconnect products connect multiple GPUs and compute nodes across racks and data centers for distributed AI training and inference. Cloud providers, AI labs, and enterprises use these to link their compute infrastructure.
ciena · nokia · infinera · acacia