AUTO-UPDATED

Arista takes Ethernet into the rack, and the network becomes the backplane

Arista Networks has launched an open Ethernet rack-scale portfolio designed to integrate AI networking directly into server racks, supporting heterogeneous compute environments for large-scale artificial intelligence deployments.

Key Points

  • The new Etherlink SU-144 architecture connects up to 144 accelerators in a single-hop domain and up to 1,024 accelerators across multiple racks.
  • Arista is collaborating with an ecosystem including AMD, Arm, Broadcom, Microsoft, Meta, and Qualcomm to support non-Nvidia AI accelerator hardware.
  • The portfolio offers three physical rack form factors: orthogonal chassis, cabled backplane, and cross-rack designs, all supporting advanced liquid cooling.
  • New hardware includes 102.4 terabit-per-second switches and next-generation optics designed to increase rack capacity to 6.5 petabits per second.
  • The strategy shifts Arista from selling individual switches to providing validated, integrated rack reference architectures for hyperscalers and AI infrastructure providers.

Why it Matters

This shift positions the network as a tightly integrated backplane rather than a separate data pipe, which is essential for maximizing the performance of modern AI compute clusters. By providing a unified operating model across scale-up and scale-out fabrics, Arista aims to challenge proprietary ecosystems and offer enterprises more flexibility in building heterogeneous AI infrastructure.
SiliconANGLE News Published by Zeus Kerravala
Read original