About Us

Intentional since the start

SambaNova designs AI infrastructure that is built around our proprietary reconfigurable dataflow unit (RDU) chip and architectures to run large models fast and efficiently. Founded on the idea of designing the full AI stack from the ground up, our engineers innovate at every layer to deliver a fully integrated AI platform with the speed to run even the largest models.

But, we aren’t just thinking about the speed of our platform, because we know it’s fast. The industry has evolved to the point where economics, not the benchmarks, drive adoption. As AI moves into everyday operations, costs must be sustainable at scale.

The spark of a new dance

The genesis of SambaNova came from two Stanford professors, Kunle Olukotun and Christopher Ré, who were focused on solving the same problem from different perspectives. Kunle was building chip multiprocessor designs, while Christopher was building the machine learning systems that ran on them. The founding team was completed with Rodrigo Liang, a semiconductor executive who knew what it takes to move a chip from a whiteboard into a production data center.

The company name, SambaNova, is Portuguese for “new dance,” reflecting a new way of designing AI chips that leaves the old technologies behind.

Meet the Team ⟶

SambaNova started operations in 2017 in Palo Alto, California on a disruptive premise: The hardware the world was using to run AI was never designed for it.

We design our own RDU AI accelerators, build the racks and systems they run in, write the compiler and orchestration software that drives them, and deliver the whole stack as a platform that neocloud operators and enterprises can run in their own data centers.

Very few companies attempt all of that. It is the only way to get the result we want: A machine specifically designed around how AI models actually compute.

What makes our products faster for AI decode is the proprietary reconfigurable dataflow unit (RDU) chip. The combination of Dataflow Architecture and three-tier memory technology keep data moving continuously to run larger and larger models, eliminating the energy-hungry memory bottlenecks that slow other systems down.

We believe in making AI infrastructure more efficient, more sustainable, and more cost effective.

SambaNova's vision for accessible AI inference

Our vision is to make high-performance AI inference fast, efficient, cost effective, and sovereign for every organization building the next generation of intelligent systems. We all benefit from empowering teams with the tools to do more and do better.

Ultimately, faster token generation, higher model utilization, and better energy efficiency represent the trifecta for meeting frontier-scale AI demands. This premium inference fuels faster experiences, more users served per unit of compute, and economics built for scale. Our customers treat AI as operational infrastructure, not an experiment.

Solve it at the root

SambaNova exists because our founders refused a workaround. We prefer the harder, lower-level fix to the clever patch on top, whether it’s in the silicon, in the software, or how we answer a customer's question.

Efficiency is the product

Tokens per watt, per rack, per dollar. We believe AI inference performance should be measured with the cost of running it at scale, and staying within a production budget.

Open, and in your control

We support open models and open standards, and build for the places data actually lives, on premises, in a sovereign cloud, or behind a regulator's requirements.