Positron AI
Positron AI builds AI inference accelerators, including Atlas and Titan, for cost- and energy-efficient LLM inference at scale.
Positron AI builds AI inference accelerators designed to deliver leading performance per dollar and per watt, with the stated aim of making advanced machine learning cheaper to deploy. The company works across silicon design, systems engineering, cloud infrastructure and LLM inference, and targets workloads running to billions of daily model inference queries. It is based in the United States, with manufacturing described as proudly American.
Two products define the current roadmap. Atlas, described as the first accelerator built specifically for LLM inference, went from conception to completion in 18 months. Titan, the second-generation system, is designed for near-limitless context and offers terabytes of per-accelerator memory.
The team reports a combined experience of more than 400 years across AI, systems, silicon and cloud engineering. The company's stated mission is to stop GPUs from bankrupting organisations deploying AI, with an emphasis on affordability and energy efficiency, and an internal communication style that is direct and informal.