Microsoft, in collaboration with researchers from the University of Cambridge, has unveiled a groundbreaking analog optical computer (AOC) that could revolutionize energy efficiency in artificial intelligence and complex optimization tasks. Published in Nature, the study details a prototype that leverages light instead of electricity to perform computations, offering a promising alternative to traditional digital processors. The AOC uses a combination of analog electronics, microLED arrays, spatial light modulators, and photodetector arrays to execute vector-matrix multiplication—the core operation in AI inference and optimization—directly in the optical domain. By bypassing energy-intensive digital-to-analog conversions, the system significantly reduces power consumption and latency. According to the researchers, the AOC could achieve up to 100 times greater energy efficiency than today’s leading GPUs. One of the key innovations lies in its ability to handle both AI inference and combinatorial optimization on the same platform. This dual-purpose design is rare and highly valuable, as many real-world problems—such as logistics, financial modeling, and medical imaging—require solving complex optimization tasks alongside AI processing. To validate the system, the team created a digital twin—a high-fidelity software simulation of the hardware—enabling large-scale testing and model training. Case studies included image classification, nonlinear regression, MRI image reconstruction, and financial transaction settlement. Results were striking: in the MRI application, imaging time was reduced from 30 minutes to just five minutes while maintaining high image quality. The financial optimization test also demonstrated near-perfect accuracy in resolving complex settlement problems. Cross-validation between the digital twin and physical hardware showed over 99% agreement in inference performance, giving strong confidence in the system’s reliability and scalability. Despite the promising results, the current prototype is still limited in scale. It operates with 256 weights for inference and up to 4,096 weights for optimization tasks, with 64 variables in the optimization model. The researchers acknowledge that real-world deployment will require scaling to hundreds of millions or even billions of weights. However, they believe this is technically feasible as microLED technology continues to advance, enabling denser and more efficient optical components. The team emphasizes that the AOC’s co-design approach—where hardware and algorithms are developed in tandem—could spark a new wave of innovation in computing. This synergy may be essential for building a sustainable future, especially as AI systems grow in size and energy demands. While the technology is not yet ready for widespread use, the research marks a significant step toward more efficient, scalable, and environmentally responsible computing. If scaled successfully, Microsoft’s analog optical computer could become a cornerstone of next-generation AI and optimization infrastructure.