Overview

Untether AI focuses on speeding up AI inference by using at-memory computing, which places compute close to memory cells to boost density and efficiency. The company ships runAI200 devices and tsunAImi accelerator cards, designed to deliver high performance within a PCI-Express form factor for tasks like vision, natural language processing, and recommendation engines. Unlike traditional accelerators that keep compute separate from memory, Untether AI’s architecture reduces data movement and increases throughput for inference workloads. The main difference from competitors is the hardware design that tightly integrates computation and memory to achieve very high ops-per-card while fitting standard bus interfaces, enabling businesses to run demanding AI tasks more cost-effectively. The goal is to help organizations optimize AI inference workloads by providing powerful, energy- and cost-efficient hardware that accelerates neural networks across various applications.