Supermicro will build, sell and support AI inference systems that pair SambaNova's SN50 chips with Nvidia GPUs. In the racks, Nvidia GPUs handle prompt reading and SN50 chips handle writing answers. The companies will start with neoclouds and sovereign clouds, and initially plan for 10,000 SN50 chips.
Supermicro will build and sell AI inference systems that pair SambaNova’s SN50 chips with Nvidia GPUs, the companies said Tuesday. The collaboration initially plans for 10,000 SN50 chips, according to SambaNova’s announcement.
The racks will join Supermicro’s Data Center Building Block Solutions, the ready-made data center systems the company sells to customers.
How do Nvidia GPUs and SN50 chips split AI inference work?
The design splits the work between two kinds of hardware. Running an AI model happens in two stages. First, the system reads the prompt. Then it writes the answer piece by piece. In the new racks, Nvidia GPUs handle the reading and SN50 chips handle the writing, SambaNova said. The company argues that AI agents work nonstop, so how fast they act depends on how fast the model responds.
“Agentic AI is the next wave of inference, and it has to run everywhere,” said Rodrigo Liang, co-founder and chief executive of SambaNova. “Now customers can order SN50 RDUs alongside GPUs, from a world-leading partner they trust.”
Supermicro will assemble, test, deliver and support the systems, including with on-site service. SambaNova engineers will get AI models running and tuned on SN50 and help with deployments.
Which customers will get Supermicro SN50 systems first?
The companies will start with neoclouds, the cloud providers built specifically for AI, and with sovereign clouds, which keep data and computing inside one country. They then plan to sell to companies that run AI in their own data centers.
“Supermicro is committed to delivering flexible, high-performance AI infrastructure that supports the evolving requirements of our customers,” said Charles Liang, president and chief executive of Supermicro.
How much power does a SambaRack SN50 draw?
SN50 is SambaNova’s fifth-generation Reconfigurable Dataflow Unit, or RDU, a chip designed for inference. The company says it can serve the largest models without requiring extreme power density in each rack. A SambaRack SN50 draws an average of 20 kilowatts, SambaNova said last week.
That matters for operators, according to the companies. SambaNova says customers can add inference capacity in data centers they already operate instead of waiting for new builds held back by power and cooling. It also says they can keep using their existing GPUs for longer. The two companies say the arrangement will let them deliver much more inference capacity in 2027.
The Supermicro deal follows another agreement with Oracle. At its AI World conference on October 27, Oracle said it plans to deploy SN50 chips in its cloud data centers and offer that capacity to AI model builders. Two days later, SambaNova signed Emergys, a digital transformation services company, to deploy and run its platform for enterprise customers.
SambaNova was founded in 2017 and is based in San Jose. It raised new funding at an $11 billion valuation in July. In June, Foxconn and Intel joined the company to build rack-scale AI infrastructure.
This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.