Cassia.ai Technology

Impossibly Tiny

Cassia.ai’s patented approach to fundamental computer arithmetic reduces the size of multiply-accumulate (MAC) and exponentiation circuits by more than 10x.

No Accuracy Loss

Many ML models accelerated with Cassia.ai technology suffer little to no loss in accuracy.

Higher Performance, Lower Power

Our technology utilizes a fundamental change in architecture to realize lower power with higher performance. A true game changer from the edge to the cloud.

About Cassia.ai

In 2021, the founders of Cassia.ai, realized that power-hungry GPUs would become a barrier to more advanced neural networks. Our company was launched to bring our research in low-power advanced arithmetic circuits to bear on artificial intelligence. Today, we believe that minimization of power consumption while maximizing performance is the key to continuing the improvement of machine learning models as the technology moves to the mainstream.

Training a self-driving car model may require 70 thousand GPU hours to train but the energy from a single training can power more than 100 houses for a month. Large Language Models (LLMs) are an order of magnitude larger! The number and size of these models is growing.

At Cassia.ai we not only increase the efficiency of intelligent systems, we make artificial intelligence accessible to everybody.