By Max A. Cherney
SAN FRANCISCO (Reuters) – Startup Untether launched an artificial intelligence chip on Monday that aims to power AI applications in cars, agricultural devices and other edge-case scenarios, the company said.
The flagship AI chips made by Nvidia and Advanced Micro Devices are designed to run inside of data centers tied together with thousands or tens of thousands of others. Such installations require large amounts of energy and additional equipment and are often geared toward building, or training, large language models that power applications like ChatGPT.
Toronto-based Untether said its chips are designed to run the AI applications once a model is already built and can do so at a greater level of efficiency.
The market for chips that only run AI applications, also called “inference,” will dwarf the market for training by 2027, Untether’s vice president of product, Bob Beachler, told Reuters in an interview. The company expects the inference market to grow to $102 billion by 2027.
“Where that inference is going to be deployed is everywhere,” Beachler said.
Untether’s chip, called the 240 Slim, offers substantial performance but with significantly less energy, and is suited for use inside autonomous vehicles, autonomous agricultural equipment and in some data centers.
Mercedes-Benz has said it is working with Untether on its next-generation autonomous vehicles.
The chips made by Untether were developed around the open-source RISC-V tech. RISC-V technology competes with designs made by Arm Holdings.
In a recent peer-reviewed benchmark from nonprofit MLCommons, Untether’s chips performed well in tests for the applications the hardware is designed for.
(Reporting by Max A. Cherney in San Francisco; Editing by Leslie Adler)
