AWS is kind of like a bank, making money on the spread between the cost of compute at different time scales. They can get cheap compute (by committing to owning datacenters and filling them with silicon) and sell expensive compute by letting customers buy it by the second. This is a great business if the demand for per-second compute is always rising, but it isn't if all your customers suddenly take you up on it all at once and start spending drastically less but you huge datacenter investment is illiquid.
A projected revenue drop is not the same as a loss though. You already bought that computer.
Less so if it's your own company who was using their stuff.
Worst case scenario: they have stuff idling and more headroom to accommodate demand.
Heck, why waste more money building yet another data center if the ones already in place have already plenty of cycles rand bandwidth to spare?
It's all an accounting trick. You don't waste more power. You don't have to but more hardware. Turning off your lambda saves nothing.
I'm not sure what the "accounting trick" is though, except in the sense that anything involving money or finance is an accounting trick. Can you elaborate?
The interesting thing about cloud costs for companies like Amazon or AWS is that in the short term it's only an accounting trick. The company is not spending money if there's a lambda deployed somewhere, or a EC3 instance idling. It's nice to erode revenue in the books from these expenses, but it's not like Besos needs to reach out of his wallet to cover the cost of a container launched by some random team at Amazon. At most he moves the cash from one pocket to the other and there's that.