chokepoints.ai
SUBSCRIBE
10 layers580 nodes2,376 dependencies9 chokepoints112 bottlenecks6,500+ companiesnode size = companies identified

Networking and interconnect hardware

BOTTLENECK

NVIDIA controls GPU-to-GPU fabric and InfiniBand, and no qualified second source can match its scale-up bandwidth.

Hardware connecting AI accelerators within and between servers, racks, and pods. Bandwidth and latency here determine how large a model cluster can train as a single unit. NVIDIA controls GPU-to-GPU fabric and InfiniBand; Broadcom dominates Ethernet switch silicon.

All hardware connecting AI accelerators within and between servers, racks, and pods; includes scale-up GPU-to-GPU fabric, InfiniBand, Ethernet switching, optical transceivers, optical components, cable assemblies, and co-packaged optics.

What the evidence shows

Cisco commanded more than a quarter of the global network hardware market in 2024.

statista.com

800G switches accounted for 35.8% of data center switching revenue in the quarter.

datacenterknowledge.com
RESCORED JUL 2026oligopolyscaling370 companies

Who controls it

CiscoBroadcomCredo+367 more tracked

No independently verified market-size figure is published for this node yet.

Where it sits in the stack

Takes in: Switch ASICs, laser chips, PIC wafers, DSP chips, copper and fibre cable materials

Sends on: High-speed lossless interconnect fabric for AI cluster communication

view in atlas

Related nodes

AI accelerators and siliconHost server CPUsNetwork interface cards, SmartNICs, and DPUsMemory subsystemsStorage subsystemsServers, systems, accelerator boards, and racks

Full scorecard, owner shares, supply edges and the full tracked roster are in the desk letter.

GET THE BRIEFING