Almost everything in computing got faster over the last fifteen years. Chips, networks, databases, model inference: all of it accelerated by orders of magnitude. Real-time bidding is a strange exception. The auction that decides which ad you see runs on roughly the same latency budget it always has. Ask a simple question: has RTB actually gotten faster? Mostly, no.

The reason is a fixed latency budget. The whole auction has to resolve in the sliver of time before a page paints (call it a hundred-odd milliseconds). That ceiling has not moved much, because the human patience it is measured against has not moved. So instead of getting faster, the system spent every efficiency gain on cramming more intermediaries into the same fixed window.

• Faster hardware did not shrink the auction. It just let more middlemen fit inside the same time budget.
• Legacy plumbing persists because every hop bills for its participation, so no one is paid to remove hops.
• The constraint is a business constraint wearing a technical costume: latency is fixed by patience, not by physics.

When a system stops getting faster while everything around it speeds up, stop looking at the engineering. Start looking at who gets paid for the slack.

That is the tell. Stagnation in an environment of universal acceleration is almost never a technical limit. It is an incentive structure. RTB did not stay slow because it could not go fast. It stayed the same speed because the extra capacity was more valuable as room for another intermediary to take a cut than as a faster result for the buyer or the human.