AI chip startup Untether launches energy-efficient AI chip for autonomous vehicles and edge tech; eyes $102B market by 2027

Three years have passed since we last featured Untether AI, a Toronto startup advancing ultra-efficient, high-performance AI chips to unlock new possibilities in AI applications. Since then, the Untether team has spent years refining chips that bring AI capabilities to a diverse range of devices, from cars to agricultural machinery, with efficiency at the forefront.

Today, Untether unveiled its latest AI chip, designed to power applications at the edge, where devices rely on inference—the task of running AI models rather than building or training them. Considering that energy use is now at the front and center of challenges facing widespread AI adoption, Untether’s new AI chip could be a breakthrough solution.

Traditional data center hardware, such as Nvidia’s flagship chips, demands substantial power to handle complex AI models, leading to high costs and sustainability challenges. In contrast, Untether’s chips are designed not for model training in massive data centers but for running AI applications with significantly lower energy requirements.

In an interview, Untether’s VP of Product Bob Beachler pointed to a significant shift in the “inference” chips—those handling already-built models. He projected that the inference chip market to surpass training in market size by 2027. He anticipates demand to hit $102 billion as AI moves into broader applications outside traditional data centers, Reuters reported.

“Where that inference is going to be deployed is everywhere,” Beachler said.

Untether’s newest release, the 240 Slim chip, stands out for its blend of performance and energy efficiency, making it ideal for autonomous vehicles and precision agricultural machinery. Mercedes-Benz has even chosen Untether to power its next-gen autonomous vehicle technology.

A crucial factor in Untether’s architecture is the open-source RISC-V technology, positioning the company as a competitor to Arm Holdings in the AI chip landscape. The hardware has already demonstrated strong results in peer-reviewed MLCommons benchmarks, underscoring its edge in energy-efficient AI processing.

Founded in 2018 by Darrick Wiebe, Martin Snelgrove, and Raymond Chik, Untether AI targets the sweet spot of low power consumption without sacrificing performance. Its innovative “at-memory” compute design reduces the need for data movement, overcoming a longstanding barrier in traditional chip architecture.

The company’s tsunAImi accelerator cards, driven by its runAI devices, showcase record-setting energy efficiency for inference acceleration. With this architecture, Untether has stepped into a growing field of specialized AI processors, aiming to redefine expectations for efficiency and performance in the AI hardware space.

Untether AI