AI may be on the brink of a major leap — powered not by silicon, but by light. Researchers have demonstrated a breakthrough called single-shot tensor computing, where complex AI calculations happen in the time it takes light to pass through an optical system.

Tensor operations — the multi-dimensional math behind image recognition, language models, and deep learning — usually demand massive GPU power. But as data grows, even advanced chips struggle with speed, energy use, and scalability.

A team led by Dr. Yufeng Zhang of Aalto University has introduced a radically different solution. Instead of running calculations electronically, they encode information into the amplitude and phase of light waves. As these waves interact, they naturally perform matrix and tensor operations — the core of AI — instantly and in parallel.

By using multiple wavelengths, the researchers also unlocked support for high-order tensor tasks, pushing the boundaries of optical computing.

Dr. Zhang explains it simply: “Our method performs operations like convolutions and attention layers — but at the speed of light.”

This innovation could usher in a new era of ultra-fast, energy-efficient AI, where a single beam of light carries the computing power of today’s supercomputers.