Arista takes aim at the next AI network battleground: distributed inference

SiliconANGLE
Read full story· 8 min
Arista Networks unveiled new Etherlink R4 routing platforms and a pluggable optical line system aimed at distributed AI inference, addressing latency, data sovereignty, and infrastructure constraints. These products leverage Broadcom silicon for energy efficiency and target emerging AI network demands beyond centralized training clusters.

What happened

Arista Networks introduced a new range of networking hardware at SC26, including the Etherlink R4 routing platforms and a pluggable optical line system. These products are designed to support distributed AI inference, which requires moving data and processing across multiple locations rather than relying solely on centralized clusters. The new lineup uses Broadcom silicon to improve energy efficiency and includes several models with different capabilities for access, edge, and core networking. Most of the new products are available now, with one model, the 7020HR4, set to ship in early 2027.

Key details

  • Etherlink R4 lineup includes 7020R4, 7020HR4, 7282R4, and 7280R4 routers.
  • 7280R4 model offers 25.6 Tbps capacity and 64 400G ports, half supporting ZR/ZR+ optics.
  • Pluggable optical line system delivers 25.6 Tbps over a single fiber pair.
  • All new hardware is built on Broadcom Jericho and Qumran silicon, claiming up to 60% lower power use than some competitors.
  • The 7020HR4 is temperature-hardened and ships in early 2027; other models are available now.

Why it matters

Companies and service providers deploying AI applications can now use Arista's new hardware to support distributed inference, which helps address requirements for lower latency, data sovereignty, and infrastructure flexibility. Enterprises, hyperscalers, and telecom operators can benefit from more energy-efficient networking options and features like deep buffers, encryption, and advanced timing. The availability of these products gives organizations more choices for building or upgrading AI-capable networks, especially in regions with strict data residency rules or where local processing is a


More in Industry & Manufacturing