Astera Labs has introduced Orion, a modular reference design for rack-scale AI infrastructure. The architecture is compute-agnostic, allowing customers to select their preferred CPU and GPU configurations for their deployments.
The company aims to assist neocloud and enterprise clients in accelerating the deployment of AI infrastructure. By utilizing open standards, the Orion design is intended to improve operational efficiency and token throughput for large-scale inference tasks.
The reference designs will be available through the company's network of OEM and ODM partners. This approach addresses the challenges faced by manufacturers who currently must manage complex integration and validation processes for each new accelerator platform.
