Nvidia Just Bet $4 Billion That Copper Is Dead. Photonics Is Next for Data.

Nvidia just committed $4 billion to two photonics manufacturers most people have never heard of. Not chip companies. Companies that build infrastructure out of light. The bottleneck in AI is not the chips. It is the wires between them. Copper interconnects hit physics they cannot negotiate - heat, signal loss, energy per bit. At the scale of 100,000 GPU clusters, these problems multiply across every dimension. What actually happens inside a modern AI training cluster: tens of thousands of GPUs in constant conversation, passing gradient updates, synchronizing parameters, moving activations billions of times per second. That constant communication rides on copper - and copper at this scale hits three physics problems simultaneously. Resistance converts signal to heat. Signals degrade over distance, faster or farther but not both. And energy per bit is small but real; at petabyte-per-second scale it shows up on the operating budget. These are not engineering challenges. They are properties of the medium itself. No architecture decision removes friction from copper. Photons work differently. No charge. No resistive heat. No degradation over distance. Through wavelength division…