AI Open Rack

Overview

The AI Open Rack is Astera Labs' open rack architecture concept for AI Infrastructure 2.0, designed to deliver open-standards scale-up connectivity across any CPU, GPU, or XPU. It consists of three integrated components: the Compute Platform (any CPU/GPU/XPU plus Astera Labs connectivity silicon), the Scale-Up Switch Platform (Scorpio X-Series PCIe/UALink fabric switches with Aries and Taurus retimers), and the Active Cabled Backplane (active signal conditioning for extended reach and scale-up across and between racks).

The architecture is built collaboratively with Open Rack partners including AMD, Amphenol, ARM, Aspeed, Cadence, Ingrasys, Insyde, Molex, Synopsys, TE Connectivity, Wistron, and Wiwynn, plus Interop Collaboration partners Arista, Micron, and Samsung — spanning IP vendors, CPU/GPU partners, cable and connector vendors, ODMs, and BMC vendors. The physical reference implementation fits within an OCP ORv3 rack and is demonstrated at Astera Labs' booth at OCP summits.