Use One Big Server (2022)

>Unfortunately, since all of your services run on servers (whether you like it or not), someone in that supply chain is charging you based on their peak load.

This seems fundamentally incorrect to me? If I need 100 units of peak compute during 8 hours of work hours, I get that from Big Cloud, and they have two other clients needing same in offset timezones then in theory the aggregate cost of that is 1/3rd of everyone buying their own peak needs.

Whether big cloud passes on that saving is another matter, but it's there.

i.e. big cloud throws enough small customers together so that they don't have "peak" per se just a pretty noisy average load that is in aggregate mostly stable

But they generally don't. Most people don't have large enough daily fluctuations for these demand curves to flatten out enough. And the providers also need enough capacity to handle unforeseen spikes. Which is also why none of them will let you scale however far you want - they still impose limits so they can plan the excess they need.

> And the providers also need enough capacity to handle unforeseen spikes.

Indeed but the headroom the cloud needs overall is less than every customers individual worst case scenarios added up. They’d take a percentage of that total because statistically a situation where 100% of customers are at 100% of their peak at 100% same point in time is improbable

Must admit little surprised this logic isn’t self evident

In which cloud can I book a machine with a guaranteed (up to general uptime SLA) end/termination time that's fixed for both?

The logic isn't self evident because it is irrelevant that their total demand doesn't add up this way, because the unit cost of the capacity is higher, and so the cost still ends up being far higher.

The unit cost is higher for many reasons, but the two basic ones are margins (exorbitant ones; this is not an efficient market) and that the providers also need to charge for the unused capacity to meet demand from customers behaving in ways they don't on fixed capacity systems, where spreading workloads over time wherever possible tends to become the norm.

The demand curves when you're charged for total demand over time rather than peak demand are fundamentally different, and so while you're right the peaks rarely add up in the worst possible way, empirically the peaks end up high enough that cloud compute is expensive even before the exorbitant margins the large cloud providers charge.