Hosted by Lester Nare and Krishna Choudhary, this story dives into Microsoft’s analog optical AI computer — a breakthrough published in Nature that runs on light instead of transistors. The system cracks two fundamental AI bottlenecks, achieving 100× energy efficiency compared to GPUs. Beyond cutting costs and power demand for data centers, the same architecture could reduce MRI scans from an hour to just five minutes, showing how fundamental physics can transform both tech and healthcare.

Summary

• Microsoft unveils an analog optical AI computer, published in Nature

• Why Moore’s Law slowdown and the von Neumann bottleneck strain AI computing

• AI’s energy demand projected to hit 20% of global electricity by 2030

• Analog design uses LEDs, filters, and CCDs instead of GPUs

• Solves matrix multiplication and nonlinearities directly in hardware

• Delivers 100× energy efficiency over GPUs

• Works for inference, not training — but that’s most AI usage

• Bonus: excels at combinatorial optimization like MRI reconstruction

• Medical impact: MRI scans reduced from 30–60 minutes to ~5 minutes

Show Notes

Nature — Analog AI ResearchMicrosoft Research Blog

Id. d’épisode : 1000725986622

GUID : 841f7e04-2fe8-47f0-befc-7c3fcb39472e

Date de publication : 11/9/2025 à 03:00:00