Graphics processing units (GPUs) , the chips that power most AI models, consume extremely high levels of energy.
Their growing use in data centers is projected to increase electricity demand by 160% by 2030, according to Goldman Sachs. Vishal Sarin, an analog and memory circuit designer, believes this trend is unsustainable.
Read more: Nvidia dominates AI PCs with new generation of chips
After over a decade in the chip industry, he founded Sagence AI (formerly known as Analog Inference) to create energy-efficient alternatives to GPUs.
Sagence develops chips and systems for running AI models, as well as the software that goes with them. While many companies are building specialized hardware for AI, Sagence stands out for its analog chips, rather than conventional digital ones. While most chips, including GPUs, store information digitally, analog chips can represent data with a variety of values.
Analog units are not a new idea; they dominated from 1935 to 1980, contributing to the modeling of the North American electrical grid and other engineering achievements. However, the drawbacks of digital chips make analog units attractive again.
See also: Apple M4 chip MacBook Air 2025 – What to expect?
For example, digital chips require hundreds of components to perform calculations that analog units can perform with a few components. In addition, digital chips often have to transfer data between memory and processors, creating bottlenecks.
Sagence seeks to overcome limitations in performance and economy with environmental responsibility, offering innovative solutions in AI processing.
Source: techcrunch
