SambaNova

Founded by Rodrigo Liang in 2017

Run large AI models faster using custom hardware for enterprise inference

Cost

Free Tier

Rating

★ People love it

Time to value

Quick Setup (< 1 hour)

You can use SambaNova to run large AI models like DeepSeek, Llama, and MiniMax at very high speeds using their custom Reconfigurable Dataflow Units (RDUs). It provides cloud-based inference APIs, on-premise rack systems for data centers, and orchestration tools to manage AI workloads. You can deploy open-source models, build agentic AI workflows, and fine-tune models while keeping data within national borders through their sovereign AI partner network.

What SambaNova does

Call AI inference APIs compatible with OpenAI SDK formatDeploy SambaRack hardware in a data center for on-premise AIMonitor and manage multiple model deployments across data centersAuto-scale AI inference capacity based on user demandSwitch between multiple large language models within a single agentic workflowBenchmark inference speed in tokens per second for different modelsConfigure sovereign AI deployments with regional data residencyBring custom model checkpoints to run on RDU hardwareCustom RDU chips with three-tier memory architecture for fast inferenceRuns DeepSeek-V3.1 at up to 200 tokens per secondRuns gpt-oss-120b at over 600 tokens per secondSambaRack SN50 supports multiple frontier models on a single node for agentic workflowsOpenAI-compatible APIs for easy application migrationSambaOrchestrator handles auto-scaling, load balancing, and model managementSovereign AI deployments in Australia, Europe, and the UKHigher energy efficiency compared to GPU-based inference systems