Ideally, the revenue from the new customer would be enough to cover the upgrades, so long as the new customer makes an up-front committment (from which loans can be written) that makes their risk (of having to pay for the upgrades even if they shut down much sooner than expected) about equal to if they build out their own off-grid system. And then they could sell to existing customers for slightly less than before, due to scale and an overall reduction of peak-to-baseline ratio.

But I guess this isn't how the world works.

A lot of the increase in bills people are seeing come from necessary upgrades to the distribution infrastructure. Something that was going to be happening anyway.

As you say, it's because the connection between the increased load and the factors requiring additional spending are at enough of a distance that they're hard to account for. If the datacenter operator argues (often with support from the power company, who has to convince government officials their rate increase is OK) that most of the grid upgrades were going to happen regardless and they've already paid for the increase fairly attributed to their operations, how can you really know whether that's true?

Power doesn’t just apperate out of thin air. It has to be generated and that has costs. If suddenly the grid draws more power then more costly sources have to feed it. Everyone pays for the same power.

The big consumer also buys in bulk and negotiates better rates etc.

There's also the supply/demand aspect of it. Some electricity is cheaper to provide than others - the cheapest is the renewable or nuclear that's already built in the area, but when demand is high, the grid provider will source electricity from more expensive sources - coal, natural gas, or importing it from neighboring utilities. So, using some made-up numbers, if your existing cost for 100MW is $0.10/Wh, getting the next 100MW might cost $0.50/Wh, bumping the cost for everyone up to $0.30/Wh.