Tensordyne designs and builds high-performance, low-power systems for generative AI inference. The company works across the full stack - custom silicon, hardware and software - aimed at accelerating multimodal generative AI inference at scale. Its technical work spans generative AI inference, custom silicon design, low-power computing, logarithmic mathematics, scale-up networking architecture and AI inference acceleration software.
The firm's flagship product is Napier, a generative AI inference system that combines purpose-built silicon with logarithmic mathematics and an optimised scale-up networking architecture. Tensordyne describes Napier as a new class of inference technology designed for speed and profitability, positioning it as an alternative to general-purpose approaches to AI inference.
Tensordyne serves hyperscaler and neocloud data centres, application-layer companies, foundation model builders and enterprises. The company is headquartered in Sunnyvale, California, with an additional office in Munich, Germany. Its stated focus is on safe, sustainable and cutting-edge AI inference solutions, drawing on recent AI research.
