
Sign up to save your podcasts
Or


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
By Krishna Choudhary and Lester Nare4.9
7474 ratings
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

32,087 Listeners

30,693 Listeners

14,390 Listeners

1,346 Listeners

549 Listeners

1,071 Listeners

934 Listeners

4,170 Listeners

1,652 Listeners

5,495 Listeners

284 Listeners

5,535 Listeners

16,139 Listeners

10,752 Listeners

1,487 Listeners