SambaNova, Super Micro plan AI inference systems built on 10,000 SN50 chips
In brief
- SambaNova and Super Micro plan inference systems with 10,000 SN50 chips, Crypto Briefing reported.
- 10,000-chip effort has no disclosed deployment timeline.
- SN50 is up to 5x faster than Nvidia's B200 GPUs, SambaNova claims.
- SN50 systems are set to ship in the second half of 2026, with SoftBank lined up in Japan.
What the SN50 is built to do
SambaNova says the SN50 can handle models with up to 10 trillion parameters. It's also making a more pointed claim. The company says the chip delivers up to 5x greater speed and up to 3x lower total cost of ownership than Nvidia's B200 GPUs (those are SambaNova's own figures, not established fact).
The rack-level specs are more concrete. SambaRack SN50 systems typically hold 8 to 16 RDUs per rack and are designed for an average power draw of roughly 20 kW per rack. They can scale to clusters of up to 256 RDUs linked by multi-terabyte-per-second interconnects, and they're designed to fit into existing data center infrastructure without requiring liquid cooling.
How big is 10,000 chips?
Crypto Briefing put 10,000 chips at somewhere between 625 and 1,250 racks, depending on configuration. The outlet described that as a rough illustration rather than a disclosed build plan, so it isn't a number either company has put forward.
There's still no public timeline for the 10,000-chip build.
Where SambaNova's other deals stand
The broader SN50 rollout does have dates attached. SN50 systems are set to ship in the second half of 2026, according to the report, and SoftBank (the telecom giant) is lined up for initial SN50 deployments in Japan.
SambaNova also has a multi-year collaboration with Intel focused on integrating with Intel's Xeon processors. Oracle Cloud Infrastructure is on the partner list too, with that partnership aimed at expanding high-performance inference capabilities across cloud environments.
Money hasn't been the bottleneck lately. SambaNova raised more than $350 million in a Series E round in February 2026, with Intel Capital among the participants, which was the same month the SN50 launched.
What's missing from the Super Micro plan is just as telling as what's in it. Crypto Briefing's report doesn't link a press release, and it doesn't say when the 10,000 chips will ship or where they'll go. Until one of the companies puts those details out, the scale of the project is the main thing on record.
Frequently asked questions
What is SambaNova's SN50 chip?
The SN50 is the latest generation of SambaNova's Reconfigurable Dataflow Unit (RDU), the company's custom AI processor. SambaNova launched it in February 2026 for AI inference workloads, and the company says it can handle models with up to 10 trillion parameters.
When will the 10,000-chip SambaNova and Super Micro systems be deployed?
No deployment timeline for the 10,000-chip effort has been disclosed. Separately, SN50 systems in general are set to ship in the second half of 2026, and SoftBank is lined up for initial SN50 deployments in Japan.
Do SambaNova SN50 racks need liquid cooling?
No. SambaRack SN50 systems are designed to fit into existing data center infrastructure without liquid cooling. They typically hold 8 to 16 RDUs per rack, with an average design power draw of roughly 20 kW per rack.
