We Built AI Inference That's Faster and Uses 100x Less Power

Added: 2026-05-21

[00:00:00]5 years from now, GPUs and electrical wires won't be the only way we compute.

[00:00:04]Right now, at this very moment, computing is moving [music] off transistors and onto something completely different. Computing using the physics of light, >> [music] >> and that should change this everything.

[00:00:16]Most people know AI demands computing power that is causing an energy problem.

[00:00:21]Data centers are consuming more power than entire cities, and that's only accelerating. And the physics of silicon has kicked a wall. It required frequent data movement and the old playbook of just shrinking transistors doesn't work anymore. But, what if we could rewrite the constraints entirely? Not by building more power plants, but by changing the way computation itself works. It's light is getting closer and closer to the processor. At OptiCore, we are making it the processor. We have built something that's never existed before. Rather than using electronic transistors for computation, we harness the photo electric fast through the use of photo detectors. Our chip shifts data at 56 GHz. For perspective, transferring 1 GB at 120 Hz would take 36 years. At our speed, that would take only 15 seconds.

[00:01:14]How? Most photonic approaches encode data [music] across a trillion individual devices. We can do it across a thousand. And the physics behind are new. It is the same principle Einstein described decades ago. It's the same science that powers the camera in your phone right now. Here it is. Our photonic chip. Originally, we started at 8 by 8. Today, we built a 256 by 256 photonic matrix, the largest photonic circuit ever constructed. And we did using standard fabrication processes.

[00:01:46]This isn't some science project. We are a real team based out of Berkeley, California. Our team came from the lab that started the entire field of optical computing using integrated photonics with PhDs and researchers from MIT and Berkeley. We were working on photonic computing even before ChatGPT existed.

[00:02:07]We didn't chase a trend or make a pivot halfway through. Our company was [music] built on this. The future of AI computing runs on light and our chips are 100 times more energy efficient than traditional processors. If you're interested in [music] what this technology can do for your company, visit optical.ai and book a call with our team.