Artificial Intelligence
Supermicro to Deliver SambaNova AI Systems in Phased Rollout

Key PointsSambaNova splits inference by workload so GPUs handle prefill while SN50 RDUs absorb…
View more
- SambaNova splits inference by workload so GPUs handle prefill while SN50 RDUs absorb the memory-bound decode phase.
- Supermicro will integrate and support the systems giving SambaNova access to an established delivery channel for enterprise buyers.
- The rollout starts with 10,000 SN50 chips for NeoClouds and sovereign clouds before reaching enterprise data centers.
SambaNova announced on November 3, 2026, that Supermicro will integrate, validate, deliver and support AI inference systems pairing NVIDIA GPUs with SambaNova SN50 RDUs. The phased rollout will begin with an initial batch of 10,000 SN50 chips for systems intended for data centers worldwide.
The systems will be distributed first to NeoClouds and sovereign clouds, then to enterprises running AI in their own data centers. SambaNova disclosed no pricing, contract value or revenue expectations, so the initial chip batch and the distribution arrangement provide the available measures of commercial scale.
Supermicro president and CEO Charles Liang said the collaboration expands the company’s Data Center Building Block Solutions portfolio by enabling heterogeneous architectures that combine GPU computing with specialized inference acceleration. The systems are aimed at agentic AI inference, with the hardware divided according to the computational demands of each phase.
Why the Hardware Is Split by Workload
The prefill phase processes the prompt and is compute-bound, which SambaNova says is where NVIDIA GPUs excel. Decode generates the response one token at a time and is memory-bound, the task for which SN50 RDUs were built. Because agentic workloads spend most of their time in decode, the design assigns that phase to SambaNova’s hardware.
AI agents can make thousands of calls to large language models, and each call can take minutes. At that volume, SambaNova says even small improvements in speed and efficiency become extremely valuable, and that assigning each phase to the appropriate hardware delivers faster, more efficient inference than either hardware type alone.
The SN50 RDUs are also designed to run high-speed, high-throughput inference of large models without requiring extremely high rack power densities, a capability SambaNova says permits deployment in existing data centers. That makes power requirements part of the commercial proposition rather than only a technical specification, particularly for the enterprise customers targeted later in the rollout.
Supermicro Handles Integration and Support
Supermicro will handle integration, validation, delivery and support, including on-site service. SambaNova engineers will work alongside Supermicro on customer deployments and tune models for performance on the SN50 RDUs. The arrangement is therefore a joint deployment effort, not simply an agreement for Supermicro to resell a component.
For customers, the purchasing change is that SN50 RDUs can be ordered through the same channel as other AI infrastructure. Rodrigo Liang, SambaNova co-founder and CEO, said: “Now customers can order SN50 RDUs alongside GPUs, from a world-leading partner they trust. That’s how fast, efficient inference gets to scale.”
SambaNova expects the partnership to result in significantly more high-performance inference capacity being deployed in 2027 than would otherwise be possible. That expectation is not a shipment or revenue forecast. The agreement gives SambaNova access to Supermicro’s delivery and service channel while giving Supermicro a specialized inference option to offer alongside GPU infrastructure; the progression from the first cloud customers to enterprise deployments will show how far that distribution model extends.












