New Microsoft optical computer uses light to cut AI energy costs by 100x
Microsoft has unveiled a new analog optical computer that runs on light, not electricity. The company claims it is 100 times faster and 100 times more energy efficient than today's GPUs, with applications in AI, banking, and healthcare.
New Delhi: Microsoft has unveiled a computer that runs on light instead of electricity, a step the company claims could change the future of technology. The device, known as the analog optical computer (AOC), has been in the works for four years and is said to be more than 100 times faster and 100 times more energy efficient than today’s best processors. The study was published on September 3, 2025, in the journal Nature.
On X (formerly Twitter), Microsoft CEO Satya Nadella called it a major step forward in solving real-world problems, adding that the research shows how analog optical computing could redefine efficiency in AI and optimization workloads.
Built with everyday components
Unlike digital machines that process data in binary, the AOC works with beams of light. Numbers are represented through light brightness, and when passed through sensors and lenses, the machine performs calculations like addition and multiplication almost instantly.
The prototype was built with everyday components such as micro-LEDs, smartphone camera sensors, and optical lenses. Microsoft says this design choice keeps the system practical, affordable, and easier to scale. Researchers estimate the machine can perform 500 trillion operations per second per watt, a massive leap compared to existing GPU systems.
Francesca Parmigiani, principal research manager at Microsoft, said the AOC is not a general-purpose machine but is well suited for optimization-heavy problems across industries. To help accelerate adoption, Microsoft is releasing its optimization solver algorithm and a digital twin of the AOC, which allows organizations to test the technology virtually.
To show real-world use, Microsoft worked with Barclays Bank to test the AOC on a transaction settlement problem involving thousands of trades. Researchers say the system resolved complex financial processes quicker and at lower cost compared to traditional methods.
Healthcare trials looked at how the AOC could reduce MRI scan times. Current scans often take 30 minutes, but Microsoft demonstrated the potential to cut that to just five minutes by reconstructing images faster. Michael Hansen, senior director of biomedical signal processing at Microsoft Health Futures, said while the prototype is not yet ready for clinical use, the early results show a spark of what full-scale systems might achieve.
Beyond finance and healthcare, Microsoft researchers mapped the AOC to basic machine learning tasks. The findings suggest it could eventually support large AI models with far less energy than current GPUs. Jannes Gladrow, a principal researcher, explained that the design allows for reasoning tasks that GPUs struggle with, such as state tracking, at a fraction of the energy cost.
The current version can handle 256 parameters in a single pass, four times greater than the previous prototype. Researchers believe future versions could scale to millions of parameters, making them capable of running very large language models.
Hitesh Ballani, who leads research on future AI infrastructure at Microsoft’s Cambridge lab, said the breakthrough shows real applications for analog optical computing. He added that while it is still early days, the long-term vision is for this technology to play a significant role in the future of sustainable computing.
For industries, the benefits are clear. Banks could settle global transactions faster and cheaper, hospitals could see more patients with quicker scans, and AI research could grow without ballooning energy costs.
The announcement comes even as Microsoft’s stock dipped 5.65% over the past month. Yet the company is still up more than 20% year-to-date, and with innovations like the AOC, it is positioning itself as a leader in next-generation computing.