Unlike the dot-com era's unused fiber infrastructure, today's AI boom faces a different constraint: GPUs are already saturated. The bottleneck is hardware capacity itself, not idle resources, which means demand for AI compute will keep growing and drive new applications we haven't imagined yet.
Everyone's doing this AI bottleneck trade right now. Anything that's where it's constrained right now in the supply chain could be memory right now, could be cooling, for example, could be other things. During the dot-com before the bust, everyone was talking about you got to build out more fiber, more fiber for the internet. You know why they call it dark fiber? It's because 97% of that buildout was unused. But today you don't have any dark GPUs. All the GPUs are being used, and more and more and more are being used. When you and I got broadband, what happened? It wasn't that we just got to use the internet faster. We got streaming. We got to do a lot of other things. We got voice communication, all these other things. So I think it's going to spawn a lot more. I don't think the need for intelligence is ever going to come down.
