The Pulse
Hyve to offer Astera Labs' Orion components in AI inference racks
Hyve Solutions will integrate Astera Labs’ Orion Compute Board, Fabric Tray and Link Board into its Apex platform for rack-scale AI inference. The companies say Apex systems featuring Orion are on track for availability in the first quarter of 2027.

Hyve Solutions will offer Astera Labs’ Orion Compute Board, Fabric Tray and Link Board as part of its Apex platform at rack scale. The companies said the pre-validated components let customers choose accelerators and CPUs to fit their cost, lead-time and workload needs, and can shorten integration and qualification cycles.
The collaboration is aimed at neoclouds and enterprises seeking to bring AI inference capacity online faster. The Orion Compute Board scales up to eight PCIe accelerators and peripherals under a single fabric switch. The Fabric Tray provides a common scale-up fabric for accelerators and memory pooling, while the Link Board extends connectivity from tray to rack scale.
Hyve’s Apex systems are air-cooled, allowing deployment in data centers without liquid-cooling infrastructure. The platform’s chassis is designed to support higher-power accelerators as they become available.
“We design, integrate, validate, and deploy at rack scale. Pairing Hyve Apex with Astera Labs’ Orion gives customers scale-up AI that they can deploy once and evolve over time, swapping accelerators as their workloads change without rebuilding the system,” said Dr. Dimitris Ziakas, Hyve Solutions’ chief technology officer and vice president of architecture.
Performance and availability
In an 8-GPU configuration, one Scorpio X-Series 320-lane switch delivered up to 89% higher tokens per second than a baseline using two 144-lane switches. The test measured output across four large language models and three workload profiles, at concurrency levels from 1 to 512 requests. Astera Labs and Hyve noted that gains vary by model and workload and that the results are not indicative of performance in all customer environments.
Hyve Apex systems featuring Orion are on track for availability in the first quarter of 2027. Customers can contact Hyve Solutions or Astera Labs for commercial inquiries.
Source
Related stories
- Cerebras puts split AI inference into production with AMD GPUs
- WEKA-backed scorecard challenges GPU use as an AI inference metric
- AMD closes Taalas deal to add inference silicon to its accelerator roadmap
- AMD reports faster inference glue kernels using AITER config pins
- Nvidia claims Vera Rubin inference lead in provisional MLPerf results