Stack Picker
This is a framework that teaches the decision, not a shopping list — you will not find a provider name in it. Answer for your actual constraints and it walks you to a category — the API tier, rented GPUs, or your own hardware — with the mechanism behind each verdict spelled out and the calculator that proves it one click away.
Question 1
Can prompts and outputs leave your infrastructure?
This is the constraint that removes options before economics get a vote. Answer for your most sensitive workload, not your average one.
How to read it: the order of the questions is the lesson. Privacy comes first because it removes options before economics get a vote; volume comes next because utilization — paying for idle hours or not — is the whole cost structure of owning versus renting versus paying per token. Serving control and ops capacity only decide what is left. If a verdict surprises you, change one answer and watch which mechanism flips it.