Supermicro to Buy 10,000 SambaNova SN50 Chips for AI Systems

Supermicro has agreed to purchase 10,000 SambaNova SN50 RDU chips to build and sell AI disaggregated inference systems, the two companies announced.

The deal puts SambaNova's inference silicon inside systems from a major server vendor. Supermicro will handle the full delivery chain: integration, testing, deployment and support, including on-site service for customers. The systems will combine Nvidia GPUs with SambaNova RDUs in a single chassis.

SambaNova's own role is narrower. Its engineers will tune AI models to run on the SN50 chips and support Supermicro during deployments. The rollout starts with neoclouds and sovereign cloud providers. Enterprises running AI in their own data centers follow later.

What is the SN50?

SambaNova unveiled the SN50 AI chip in February 2026. The company designed it for agentic AI workloads. It claims the RDUs are 5x faster than "competitive chips" and deliver 3x lower total cost of ownership. SambaNova did not name the comparison hardware.

The chip is built on SambaNova's Reconfigurable Data Unit (RDU) architecture. Key specifications:

• Three-tier memory design supporting ten-trillion-parameter models
• Capacity for "thousands of simultaneous AI sessions with consistent high performance"
• 4x more network bandwidth than the previous generation
• Up to 256 accelerators linked over a multi-terabyte-per-second interconnect

The architecture points at one workload class: many parallel inference sessions running large models at once. That is the agentic AI pattern, where software agents make repeated model calls.

What does this change for buyers?

"Agentic AI is the next wave of inference, and it has to run everywhere," said Rodrigo Liang, co-founder and CEO of SambaNova. "Supermicro builds AI infrastructure for data centers globally. Now customers can order SN50 RDUs alongside GPUs, from a world-leading partner they trust. That's how fast, efficient inference gets to scale."

In practice, buyers get a single vendor route to heterogeneous hardware. They can order GPU and RDU compute together, with one company responsible for integration and support, rather than assembling inference infrastructure from separate suppliers.

Charles Liang, president and CEO of Supermicro, said: "Supermicro is committed to delivering flexible, high-performance AI infrastructure that supports the evolving requirements of our customers. Our collaboration with SambaNova expands our broad DCBBS portfolio by enabling heterogeneous architectures that integrate GPU computing with specialized inference acceleration, helping customers deploy efficient, scalable AI solutions."

The mention of the DCBBS portfolio signals where the chips sit in Supermicro's product strategy: as inference accelerators inside a broader data center building block lineup, not as GPU replacements.

A company still changing shape

SambaNova, founded in 2017, started with training workloads. In 2025 it pivoted to become an AI cloud services provider. Its Suite product lets companies train generative AI models based on popular foundation models, on-premises or in the cloud.

The company launched SambaNova Cloud in September 2024. That service provides cloud-based inference using the SN40L chip, launched in September 2023, which runs models with up to five trillion parameters.

The Supermicro deal extends that trajectory from cloud services back into hardware distribution. A 10,000-unit initial order gives SambaNova volume deployment through a channel it lacked on its own, and the actual demand signal will emerge as the first neocloud systems ship.