Have you looked at how many billions of dollars ISPs spend for wireless spectrum or deploying cell towers? Buying more is neither fast nor simple, so there's a real benefit to figuring out how to allocate a naturally limited resource fairly. The approach of charging for total bandwidth consumption doesn't do a good job of this since it's only a crude proxy for the actual problem.
> Further, the rest of the internet is going to ignore your high vs low latency tag
QoS has had various standardized forms for a long time so “the rest of the internet” hasn't been a factual statement since at least the 90s, although it's certainly far from a universally-deployed feature.
More to the point, however, we were talking about the last-mile network which tends to be both the most congested and the only point where pricing is surfaced to the user. It doesn't matter whether, say, a backbone provider does anything other than deliver the packet if your wireless carrier does use that flag to throttle packets at different rates when the network is congested. Many ISPs have already deployed equipment capable of doing this but the industry billing models do not reflect that, which is something I'd like to see change.
As to upgrading networks, the wireless industry is pulls in $187.8B per year and spends $32.1B on capital investments. Yes, the network is expensive, but it's less important to there bottom line than you might think.