The network is becoming the computer — or at least the backplane. On October 7, Arista Networks announced an open Ethernet rack-scale portfolio designed to power high-density scale-up and scale-out AI infrastructure, extending Ethernet into territory long held by proprietary interconnects.
The announcement, made in Santa Clara, was engineered in partnership with AMD, Arm, Broadcom, d-Matrix, Meta, Microsoft, and Qualcomm, and builds on the Ethernet for Scale-up Networks (ESUN) initiative. The designs support up to 144 accelerators per rack, with even more reachable through cross-rack architecture — liquid-cooled racks purpose-built for the extreme density, power, and thermal demands of modern AI clusters.
"The network in the AI era is no longer just an independent pipe between servers. It's now a tightly integrated rack-scale backplane that is essential to maximizing compute," said Tyson Lamoreaux, Senior Vice President of Cloud and AI Networking at Arista. The pitch is openness: a standards-based alternative to proprietary compute islands, letting cloud operators scale AI infrastructure with their preferred mix of accelerators rather than a single vendor's stack.
The scale-out side of the portfolio leans on Ultra Ethernet Consortium standards, Multipath Reliable Connection fabric resiliency, multi-plane routing, and intelligent dynamic and cluster load balancing to optimize job completion times. The rack designs are thorough pieces of engineering: liquid-cooled 7060EX7 switches with 102.4 terabits per second of capacity each, fiber patch panels, integrated liquid-cooling manifolds, drip trays with active leak detection, rack management controllers, and power shelves with backup battery units.
Perhaps the most striking number is optical. With next-generation optics such as XPO and open co-packaged and near-packaged optics, per-rack capacity jumps from 1.6 petabits per second on today's OSFP-based switches to 6.5 petabits per second — a fourfold densification that Arista says cuts the overall data center footprint nearly in half, with proportional savings in time and money.
Arista plans to demonstrate the rack-scale solutions at the OCP Global Summit, October 12–15, at booth E61. For enterprise architects, the message is that Ethernet — the same technology running office LANs — is now a credible fabric for the most demanding AI workloads on earth. The proprietary interconnect era isn't over, but its moat is getting shallower every quarter.