Will AI expose the rickety infrastructure of the internet?
The phrase "held together by glue and dreams" has been doing a lot of heavy lifting in tech circles, and it's more accurate than most people want to admit. The internet's backbone was engineered for a world of static pages and bursty browsing, not for millions of agents making synchronous, low-latency calls to inference endpoints around the clock. AI doesn't just add traffic, it changes the shape of it, and that's where the cracks show.
What I find interesting is how quickly the bottlenecks are moving up the stack. We spent years worrying about data center power and chip supply, but the conversation is now drifting toward peering agreements, last-mile fiber, and the fact that undersea cables have finite capacity. Training runs can be scheduled and batched, but inference is real-time. You can't queue a user's question.
I don't think the internet collapses, but I do think we'll see a lot of expensive, ugly inefficiency first. Latency spikes, regional congestion, and a fresh wave of consolidation as a few big players buy up the pipes. The question I'd want answered is whether the economics of AI inference will force a redesign of how data moves, or whether we'll just keep patching the glue.