# SambaNova

> AI infrastructure platform delivering fast inference on large open-source models with custom dataflow technology and energy-efficient RDU chips.

SambaNova provides purpose-built AI infrastructure for scalable inference, featuring custom dataflow technology and a three-tier memory architecture that delivers energy efficiency for fast inference and model bundling. The platform powers sovereign AI data centers globally and supports the largest open-source models including DeepSeek, Llama, and gpt-oss with lightning-fast inference speeds.

- **Reconfigurable Dataflow Unit (RDU)** - Custom AI chips with unique 3-tier memory architecture and dataflow processing that achieve faster inference using significantly less power than traditional GPU architectures.

- **SambaCloud** - Cloud-based inference platform with OpenAI-compatible APIs, enabling developers to build applications with the best open-source models in minutes.

- **SambaStack** - Complete chips-to-model computing stack built specifically for AI, providing a fully integrated end-to-end agentic AI platform spanning agents, models, knowledge, and data.

- **SambaRack** - State-of-the-art hardware system that delivers fast inference on the largest models with an average of just 10 kW of power consumption.

- **SambaManaged** - Managed infrastructure solution for enterprises requiring dedicated AI infrastructure with full control.

- **Sovereign AI Solutions** - Partnerships with data center providers across Australia, Europe, and the UK to deliver top-tier AI performance within national borders while maintaining data sovereignty.

- **Model Support** - Runs frontier-scale models including DeepSeek-R1 at up to 200 tokens per second and gpt-oss-120b at over 600 tokens per second, with support for Llama 4 series.

- **Energy Efficiency** - Generates maximum tokens per watt with RDU technology measuring 4X better than GPU as measured by Intelligence per Joule.

- **Model Bundling** - Infrastructure flexibility allows switching between multiple frontier-scale models, enabling complex agentic AI workflows to execute end-to-end on one node.

To get started, developers can access SambaCloud directly through the web interface, obtain API keys, and begin building with OpenAI-compatible APIs. Enterprise customers can contact SambaNova for custom SambaStack or SambaManaged deployments.

## Features
- Custom RDU (Reconfigurable Dataflow Unit) chips
- Three-tier memory architecture
- OpenAI-compatible APIs
- Lightning-fast inference
- Model bundling on single node
- Sovereign AI data center solutions
- SambaCloud inference platform
- SambaStack full-stack AI infrastructure
- SambaRack hardware systems
- SambaManaged enterprise solutions
- Support for DeepSeek, Llama, and gpt-oss models
- Auto scaling and load balancing
- Model management and monitoring
- Energy-efficient inference

## Integrations
Meta Llama, DeepSeek, OpenAI gpt-oss, Hugging Face, CrewAI, OpenRouter, Zilliz, Hume AI, aiXplain, Parasail

## Platforms
WEB, API

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://sambanova.ai
- Documentation: http://docs.sambanova.ai/
- EveryDev.ai: https://www.everydev.ai/tools/sambanova