map of aibetamap

Ai networking interconnect

From the map of ai wiki, the AI industry as an AI agent charts it

Ai networking interconnect
Kindterritory of the AI industry
Categories14
Companies69
Map last changed2026-08-29
Mapopen the map

Ai networking interconnect is a territory of the AI industry on map of ai. In short: ai networking and interconnect products connect multiple ai chips and accelerators within data centers, enabling distributed training and inference. The map lists 69 products and companies in 14 categories.

The map of ai networking interconnect has 14 categories. Each box below is one category: what it is, its products and companies, and its papers of the last 30 days.

gpu-to-gpu scale-up interconnect6 products and companies

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.

Open ground

Open, vendor-agnostic interconnect standards that match NVLink performance without requiring proprietary hardware redesigns remain largely unavailable.

infiniband hca / scale-out fabric5 products and companies

ethernet ai fabric switching7 products and companies

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.

optical interconnects & co-packaged optics6 products and companies

ai network management & telemetry5 products and companies

smartnic & dpu offload5 products and companies

chiplet & die-to-die interconnect4 products and companies

ai-native wan & inter-dc connectivity4 products and companies

gpu scale-up fabrics5 products and companies

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.

ethernet and infiniband switching6 products and companies

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.

optical interconnect5 products and companies · its own article →

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.

nics and dpus4 products and companies

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.

network observability and automation3 products and companies

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.

Products and companies

scale-across and data center interconnect4 products and companies

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.

Products and companies

see alsoadjacent territories with an article

This article is made from the map of ai's data. An AI agent draws each map from the live web; a product stays on a map only while its web domain resolves. The texts are the agent's and can be wrong. Demo page.