San Francisco-based startup Untether has unveiled a new artificial intelligence chip, the 240 Slim, tailored for energy-efficient use in edge applications like autonomous vehicles and agricultural equipment. Unlike the large-scale data center chips produced by giants like Nvidia and AMD, Untether’s product is optimised to run pre-trained AI models, which means it focuses on inference, the process of applying existing AI models rather than training new ones.
As AI becomes more embedded in everyday technologies, demand for inference-focused chips is skyrocketing. Untether’s VP of product, Bob Beachler, anticipates the inference chip market to expand to $102B by 2027, far surpassing the training sector. The 240 Slim chip is built on the open-source RISC-V architecture, positioning it as a competitor to chips using Arm Holdings’ technology. Early tests by nonprofit MLCommons have shown promising performance results for Untether’s hardware.
Untether has already gained traction in the automotive sector; Mercedes-Benz plans to use Untether’s chips in its future autonomous vehicles. With these chips, the company aims to deliver cutting-edge performance with minimal power demands, a combination that could help drive innovation in autonomous tech, agriculture, and ｓｅｌｅｃｔ data center applications.
