Specifications
System
SambaRack SN50
Vendor
SambaNova Systems
Chip Generation
Fifth-generation SambaNova RDU
RDUs per Rack
16x SN50 RDU
Cooling
Air-cooled - no liquid cooling loop required
Purpose
Cloud-scale agentic AI inference
Scale-Out
Inference workloads scaled out over a dozen racks
Software Platform
SambaStack full-stack enterprise AI platform
Architecture
Dataflow execution model with SambaFlow
Deployment Options
On-premises or in the cloud
Efficiency Position
Energy-efficient inference versus GPU-based baselines
Rack Integration
Purpose-built, factory-integrated rack
Target Customers
Neoclouds, sovereign AI programmes, government and public sector, enterprises
Product Family
SambaCloud, SambaStack, SambaRack and RDU
Overview
The SambaNova SambaRack SN50 is a purpose-built, air-cooled AI rack for agentic inference. It combines sixteen fifth-generation SambaNova SN50 RDU chips in one rack, and is designed so that inference workloads can be scaled out across a dozen racks - a deployment shape aimed squarely at cloud-scale inference rather than single-node experiments.
SambaRack sits at the hardware layer of SambaNova's stack, below the SambaStack full-stack enterprise AI platform and alongside SambaCloud. The architectural proposition is the RDU's dataflow execution model, which SambaNova positions as delivering roughly four times the energy efficiency of comparable GPU-based inference infrastructure, with the practical benefit for operators being air cooling rather than a liquid loop.
For organisations building dedicated AI factories - neoclouds, sovereign AI programmes, government and public sector deployments - SambaRack offers a factory-integrated building block that can be deployed on-premises or in cloud environments with the same software stack.
Key Benefits
Air-cooled deployment. No liquid cooling loop simplifies facility requirements in retrofit data centres. Rack-scale agentic inference. Sixteen SN50 RDUs per rack with clean scale-out across a dozen racks. Full-stack software. SambaStack provides the software layer above the hardware, with on-premises or cloud deployment options. Energy efficiency focus. Positioned for substantially lower energy per token than GPU-based inference infrastructure.
Applications
Agentic AI inference services, neocloud deployments, sovereign AI programmes, government and public sector AI factories, enterprise private inference, and rack-scale model serving.
Request a Quote — SAMBANOVA SAMBARACK SN50 — 16-CHIP AIR-COOLED RACK FOR AGENTIC AI INFERENCE
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →