💡

We need to stop treating light as something that only connects electronic processors and start seriously rebuilding the processor around light itself.

Optical interconnects are important, but they leave the basic architecture intact: electronic compute generating enormous heat while photons clean up the traffic between chips.

The harder bet is photonic compute as a first-class architecture.

That means funding the ugly, long-horizon problems too: optical memory, nonlinear operations, precision, fabrication, integration and a complete compute stack.

It may take longer to monetize than shipping next year’s faster GPU. That’s exactly why the market will chronically underinvest in it.

If energy and heat are becoming fundamental constraints on AI scaling, we shouldn’t spend all our capital optimizing the architecture creating the constraint. We should be funding serious attempts to replace it at the root.