Semiconductor connectivity company Astera Labs and infrastructure manufacturer Hyve Solutions have expanded their collaboration to help cloud providers and enterprises deploy artificial intelligence inference capacity more quickly. The companies announced the agreement on November 16, outlining plans to integrate Astera Labs’ Orion rack-scale architecture into Hyve’s Apex platform.
The collaboration is designed to simplify the deployment of AI infrastructure by combining pre-validated connectivity components with server systems that can accommodate different accelerator and processor configurations.
A Modular Approach to AI Server Deployment
Under the agreement, Hyve Solutions will integrate Astera Labs’ Orion Compute Board, Fabric Tray and Link Board into its Apex platform. The components are designed to connect accelerators and memory across server racks while supporting different hardware configurations.
The companies say the approach will allow customers to deploy a rack infrastructure once and change accelerators as computing requirements evolve. This could reduce the need to rebuild entire systems when businesses upgrade their AI hardware.
The platform is intended for neocloud providers and enterprise customers seeking to expand inference capacity without committing to a single accelerator configuration for the full lifecycle of a data centre deployment.
Connectivity Becomes a Key Performance Factor
As AI workloads become more demanding, the ability to move data efficiently between processors, accelerators and memory has become increasingly important. Connectivity limitations can leave expensive computing hardware underutilised, increasing the cost of running AI services.
Astera Labs said its single-hop Scorpio fabric delivered up to 89% higher tokens per second in an eight-GPU configuration compared with a conventional two-switch baseline in its testing. The result is a company-reported benchmark and should be considered in the context of the specific configuration tested.
The companies also said the Apex systems would be air-cooled, allowing deployment in existing data centres without requiring liquid-cooling infrastructure for this configuration.
Commercial Availability Planned for 2027
Hyve Apex systems incorporating the Orion architecture are expected to become available in the first quarter of 2027. The partnership reflects growing demand for modular AI infrastructure as businesses seek to increase computing capacity while managing equipment costs and deployment timelines.
The broader significance lies in the shift towards rack-scale design, where connectivity, memory and accelerator integration are planned as one system rather than as separate components















Comments