Astera Labs unveils modular Orion architecture for AI inference

what happened

Astera Labs has introduced Orion, a modular rack-scale reference architecture for AI infrastructure, GlobeNewswire reports. The design targets neocloud providers and enterprises, allowing them to select GPUs and CPUs according to availability, lead time, cost and workload requirements.

Orion includes six production-grade reference designs covering the rack, compute tray, fabric tray, memory tray, link board and compute board. Its standards-based PCIe connectivity is designed to scale from 8 to 160 accelerators per pod and help increase tokens per second from each accelerator across inference workloads.

The architecture is integrated with COSMOS Rack Manager and COSMOS Fabric Manager for management, fleet-wide telemetry, diagnostics and infrastructure intelligence. Astera Labs plans to extend the roadmap to other open interconnects, including UALink. Orion rack and system designs are expected to become available through OEM and ODM partners in Q1 2027. Astera Labs will demonstrate a powered rack at the OCP Global Summit in San Jose, California, on October 12–15, 2026.

why it matters

The announcement matters for the AI-infrastructure market because Orion separates rack-scale design from the choice of accelerators and CPUs. Astera Labs says this can help operators expand inference capacity while adapting to changes in hardware availability and economics.

instruments