Microsoft’s New Light-Based Computer Could Revolutionize AI Efficiency Using Decades-Old Technology
September 9, 2025 — Microsoft has unveiled a groundbreaking computing system that uses light, rather than traditional digital switches, to perform calculations — a development that could make artificial intelligence (AI) up to 100 times more energy efficient. Drawing inspiration from technology dating back 80 years, this new analog optical computer (AOC) represents a potentially transformative computing paradigm aimed at overcoming the limits of current digital hardware.
A New Computing Paradigm
Unlike conventional digital computers that rely on billions of tiny electronic switches to perform calculations, Microsoft’s prototype uses micro-LEDs and camera sensors to manipulate light signals. By employing light and varying voltages to add and multiply numbers in continuous feedback loops, the AOC processes data in an analog fashion. This allows the system to iteratively refine its computations until it reaches a "steady state" — essentially the final solution to the problem at hand.
Microsoft researchers emphasize that this innovative approach drastically reduces energy consumption since it bypasses the need to convert analog signals back into digital during calculations, which saves both energy and time.
“The most important aspect the AOC delivers is that we estimate around a hundred times improvement in energy efficiency. That alone is unheard of in hardware,” said Jannes Gladrow, a Microsoft AI researcher and co-author of the study.
Performance and Applications
The research team demonstrated their optical computer’s capabilities through a variety of tasks currently handled by AI and optimization algorithms. These included:
- Image classification: The physical AOC performed comparably to a digital computer on basic machine learning tasks.
- Medical imaging reconstruction: Using a digital twin (a computer model replicating the AOC’s behavior), researchers reconstructed a 320-by-320-pixel brain scan using just 62.5% of the original data. This holds promise for reducing the time required for MRI procedures.
- Financial optimization: The system solved complex financial problems involving risk minimization and fund exchanges more efficiently than some existing quantum computers.
Aydogan Ozcan, an optical computing expert at UCLA not involved in the project, noted that the AOC is a specialized “steady-state finder” designed for particular AI and optimization problems rather than a general-purpose computer. Still, for the applications it targets, the AOC could deliver significant advances over traditional digital systems.
Bridging Physics and Computing
The inspiration for this light-based system harks back to analog computers developed roughly 80 years ago but enhanced today with modern micro-LED and sensor technology. The team also created a digital twin of the prototype, allowing researchers to simulate larger-scale problems beyond the physical device’s current capacity. This digital twin model can scale up to millions or billions of variables, paving the way for future iterations that could vastly outperform current computers in speed and energy consumption.
Michael Hansen, senior director of biomedical signal processing at Microsoft Health Futures, explained, “The digital twin is where we can work on larger problems than the instrument itself can tackle right now.”
Future Prospects
Although still a prototype, Microsoft’s AOC represents a bold step toward a new class of computers tailored for AI workloads and complex optimization tasks. The developers envision future versions with added micro-LEDs that can simultaneously handle massive datasets, potentially revolutionizing how AI computations are performed.
“Our goal, our long-term vision is this being a significant part of the future of computing,” said Hitesh Ballani, a researcher with Microsoft’s Cloud Systems Futures team.
As AI continues to grow rapidly in both complexity and energy demands, innovations like Microsoft’s light-based computer could offer sustainable, scalable solutions that enable smarter and greener technology worldwide.
The underlying research was published in the journal Nature on September 3, 2025.
Credit for the development and research goes to Microsoft researchers and associated collaborators.
Related Stories:
- Quantum AI algorithms outpacing supercomputers
- Room-temperature quantum computers using photonic technology
- Advances in magnetically controlled memory chips to reduce AI power usage
Author: Skyler Ware is a freelance science journalist with expertise in chemistry, biology, and computing technologies. She holds a Ph.D. in chemistry from Caltech.
