Most caching is reactive. Someone requests a thing, you miss, you fetch, you keep it for next time. The first user always pays.
There's a stronger move: pre-position. Predict what will be wanted, and put it in place before anyone asks. Nobody pays the miss.
It only works when three things are true:
→ demand is predictable
→ the content doesn't change per user
→ off-peak capacity is cheap or free
That last one is the quiet trick. You're not avoiding the transfer cost. You're moving it to an hour when nobody's competing for the pipe.