Astera Labs introduces Orion rack-scale AI design

Astera Labs has announced Orion, a compute-agnostic rack-scale reference design for AI infrastructure that combines modularity with production-grade connectivity. The architecture is designed to let customers select their own GPU and CPU while building racks based on open standards. Astera Labs plans to make Orion available through its ODM and OEM partners.

The design targets neocloud and enterprise customers that need to expand inference capacity while maintaining predictable, low-latency performance across complex infrastructure. Astera Labs frames Orion as a way to reduce the cost and risk of committing early to a single accelerator platform before availability or pricing is clear. The company also positions the reference designs as a way for ODMs and OEMs to avoid repeated integration, validation and software work for each new accelerator platform, helping speed AI infrastructure deployments when inference demand is high.