Astera Labs has introduced Orion, a modular, rack-scale reference architecture for AI infrastructure. Its compute-agnostic design lets neocloud and enterprise customers choose their preferred mix of GPUs and CPUs and scale AI inference capacity without being locked into one hardware vendor.
Built on open standards, Orion uses a high-radix PCIe switch fabric that can connect up to 160 accelerators per pod. Astera Labs will offer production-grade reference designs through its OEM and ODM partners rather than manufacture the racks itself. The company also plans to add support for other open interconnect standards, including UALink.
Astera Labs is showcasing a powered Orion rack at the 2026 Open Compute Project (OCP) Global Summit. Partner systems based on the reference designs are expected to become available starting in the first quarter of 2027.