Lumai's Optical Computing System Cuts AI Energy Use by 90%

A British company born from Oxford University, Lumai, has introduced an innovative optical computing system capable of running large-scale language models in real time. This breakthrough marks a major step for the tech industry, which faces mounting energy challenges due to AI expansion.

The rise of AI is reshaping digital infrastructure needs worldwide. Data centers supporting this new era could double their energy demand by 2030, threatening the sector's sustainability.

The Lumai system performs millions of operations simultaneously using light in a three-dimensional volume, consuming up to 90% less energy than traditional servers.

Their server, called Iris Nova, uses photons instead of electrons for information processing, employing an optical tensor engine for mathematical operations and digital control. This hybrid architecture overcomes physical limits of silicon chips, enabling greater parallelism and efficiency.

Tests with large models like Llama 8B and 70B showed the platform can perform real-time inferences, demonstrating its potential for integrating advanced AI into productive environments without high energy costs. The technology is now available for evaluation by cloud providers, research centers, and companies.

Additionally, Lumai offers other servers, Aura and Tetra, which further increase computing capacity and energy efficiency compared to GPU-based systems. This innovation has garnered interest from UK innovation agencies, viewing it as a solution to the sector's energy challenges.