Hyve’s custom-designed network switches deliver high-throughput, low-latency connectivity required for AI and HPC environments where consistent data movement is crucial to workload performance.
Data center switch platforms leverage the NVIDIA Spectrum and Broadcom Tomahawk silicon that support hardware-level adaptive routing and dynamic load balancing operating at line rate to optimize GPU-to-GPU communication across diverse topologies.
Features like Linear Pluggable Optics (LPO), Co-Packaged Optics (CPO) and Co-Packaged Copper (CPC) reduces power consumption for air-cooled and liquid-cooled environments.
Switch designs emphasize modularity, so configurations align with specific customer workload requirements without unnecessary cost or complexity.
Hyve offers flexibility with switching silicon from Nvidia, Broadcom and Marvell enabling customers to select the architecture that best fits their environment.
Every switch is rack and network OS ready. Whether deploying a single unit, in an aggregation rack, or a cluster at scale, Hyve validates the switch as a fully integrated data center component. This dock-to-deployment model compresses provisioning timelines and eliminates common deployment disruptions. All switches support open source and custom OS and are maintained by Hyve’s Unified Global Services (H.U.G.S.) for ongoing operational continuity.
Hyve NextLink is a 2RU, air-cooled switch built for AI and high-performance computing (HPC) fabrics. It delivers 51.2 Tbps of non-blocking switching capacity across 64 × 800GbE OSFP ports, ready for open NOS deployment including ONIE and SONiC. Features like Linear Pluggable Optics (LPO), Co-Packaged Optics (CPO) and Co-Packaged Copper (CPC) reduces power consumption for air-cooled and liquid-cooled environments.
AI Infrastructure Platforms
Modular Approach to Rack-Scale Data Centers
