It's also an indefinite cost. It's not like Valve decides to stop hosting a game they've sold after a while. Generally speaking, they store and host it forever even if they never get revenue from sales of it ever again. Of course, I'm sure if the revenue wanes that much that downloads will too, but there's definitely a crossover point where maintenance will start being a permanent negative cashflow. Now multiply that across tens to hundreds of thousands of games and counting. Forever. You kind of have to consider that for the long term when setting your pricing for today since sales cuts are the only revenue you get.

I don't mean to be rude either, but you shouldn't assume that someone doesn't know what they are talking about or understand.

From professional experience I can speak pretty confidently on the subject that staffing opex is almost universally going to supersede cloud opex.

EDIT: I noticed that I was being a bit unclear when saying fixed. What I mean by fixed in this context is that you need to develop the whole e-commerce infrastructure regardless of if you have 1 game or 1000 games - a simplification as you do need to take care to scale well when growing, but it is good enough for the purposes of demonstration. The more games sharing the same e-commerce infrastructure, the less the e-commerce infrastructure costs on a per-transaction basis.