A hardware platform exposed as a service
SambaNova Systems combines specialized AI hardware with model-serving software. Its SN40L Reconfigurable Dataflow Unit uses streaming dataflow and a three-tier memory system spanning SRAM, HBM and DDR memory. A hosted API lets developers use that infrastructure without installing a complete local system. [2][1]
Public cloud access with staged tiers
The SambaNova Cloud launch made free API access and an enterprise tier available on September 10, 2024. The developer tier was still planned for later. Later tiers and feature releases have their own access dates. [1]
Serving models is a distinct contribution
The launch served Meta’s Llama models. That makes SambaNova an inference provider, rather than the author of those model weights. Its advertised speed depends on the model, precision and serving configuration; API availability does not turn every comparative claim into a general hardware result. [1]
Sources
- SambaNova Launches The World’s Fastest AI Platform
SambaNova · Sep 10, 2024
- SambaNova SN40L: Scaling the AI Memory Wall with Dataflow and Composition of Experts
SambaNova researchers · Nov 5, 2024