Supermicro launches SN50 AI inference racks

Super Micro Computer has introduced new SN50 AI inference racks through a strategic collaboration with SambaNova, adding high-performance inference capabilities to its data center portfolio. The systems use SambaNova’s SN50 chips and are designed as full AI inference racks for enterprise and cloud data centers.

The collaboration is initially planning for 10,000 SN50 Reconfigurable Dataflow Unit chips, which will be built into disaggregated AI inference systems that combine SambaNova RDUs with NVIDIA (NVDA) GPUs. These racks expand Supermicro’s focus on configurable, full-rack infrastructure rather than custom one-off designs.

The SN50 inference racks are being added to Supermicro’s Data Center Building Block Solutions portfolio, operating alongside its existing NVIDIA-based systems and Vera CPU racks. This creates a broader mix of accelerators, offering buyers configurations that can include GPUs, SambaNova RDUs and Vera CPUs within the same infrastructure framework.

In these heterogeneous systems, NVIDIA GPUs handle the prefill phase of AI workloads, while SambaNova SN50 chips are built to run the decode phase quickly. This division of labor is intended to increase inference throughput and efficiency, allowing operators to add capacity without requiring extreme rack power density or next-generation data center build-outs.

Supermicro will integrate, test, deliver and support the new inference racks globally, including providing on-site service for customers. SambaNova engineers will tune models for SN50 and support deployments, helping ensure that workloads are optimized for the new chips.

Initial deployment targets include neoclouds and sovereign clouds, with plans to extend to enterprises that run AI in their own data centers. By pairing NVIDIA GPUs that read the prompt with SN50 chips that write responses, the companies aim to address emerging agentic AI workloads that require continuous, fast model responses.

The SN50 racks sit alongside Supermicro’s broader AI portfolio and support its push toward higher-margin, integrated AI data center solutions. The company continues to build modular server and storage platforms that can be configured with different accelerators inside standardized rack designs.

Super Micro Computer shares traded higher on Tuesday as the market responded to the announcement of the SambaNova collaboration and the planned SN50 inference racks. Trading also came ahead of Supermicro’s scheduled fiscal first-quarter earnings report, which is due after the market close with a call at 5 p.m. ET.