Supermicro will build, sell and support artificial intelligence inference systems that combine NVIDIA GPUs with SambaNova SN50 chips. The announcement was made on November 3, 2026, and brings SambaNova's accelerators into Supermicro's Data Center Building Block Solutions (DCBBS) portfolio.
The collaboration starts with an initially planned batch of 10,000 SN50 chips. Distribution will roll out in phases: first to NeoClouds and sovereign clouds, then to enterprises running their own data centers.
Why inference is split between GPU and RDU
The architecture splits inference work into two phases. In the prefill phase, the NVIDIA GPU processes the prompt — a compute-bound task. In the decode phase, the SN50 chip generates the response one token at a time — a memory-bound task.
According to SambaNova, AI agents can make thousands of calls to language models and each call can take minutes, and agent workloads spend most of their time in decode. The company therefore argues that an accelerator specialized in that stage can make agent inference faster and more efficient than using only one type of hardware.
What the SN50 is
The SN50 is SambaNova's fifth-generation reconfigurable dataflow unit (RDU), designed for high-performance inference. Its official datasheet lists a 5-nanometer TSMC process, 432 MB of on-chip SRAM and 64 GB of HBM2E, with DDR5 as a third memory tier.
SambaNova says the SN50 can scale to as many as 256 RDUs across multiple racks and was designed for interactive agents, coding assistants and other multi-step inference workloads, including applications that switch among several models. The company also says the RDUs do not require extremely high rack power densities, which would make deployment easier in existing data centers.
How delivery to customers will work
Under the announced division of roles, Supermicro will handle integration, validation, delivery and support for the systems, including on-site service. SambaNova engineers will work alongside Supermicro on deployments and on tuning models for the SN50.
“Supermicro builds AI infrastructure for data centers globally,” said Rodrigo Liang, SambaNova co-founder and CEO. “Now customers can order SN50 RDUs alongside GPUs, from a world-leading partner they trust.”
“Supermicro is committed to delivering flexible, high-performance AI infrastructure,” said Charles Liang, president and CEO of Supermicro, noting that the collaboration expands the DCBBS portfolio with heterogeneous architectures combining GPU computing with specialized inference acceleration.
Geekiki context: Supermicro was previously covered in October for a 256-CPU NVIDIA Vera rack for agentic AI at scale (read the previous article).
Editorial classification: CONFIRMED — this is an official announcement from both companies, published on November 3, 2026. The expectation of “substantially more high-performance inference capacity in 2027” is a company projection, not a measured result.
Image credit: The featured image shows SambaNova's SN50 chip, in earlier material from The Next Platform.
Sources
- SambaNova — Supermicro Will Build and Sell SambaNova-Powered Inference Systems (anúncio oficial, 3 de novembro de 2026)
- Unite.AI — Supermicro to Build and Sell SambaNova-Powered AI Inference Systems
Transparency note: this article was produced with AI assistance and checked against official and online sources, including SambaNova's announcement and Unite.AI's coverage. Readers are encouraged to consult the original sources listed below.
Muse
