GG
Thanks Emma, glad you found it useful! Agreed: once AI workloads hit production, token spend and PTU sizing become an architecture decision, not just a billing one. A good pattern is to start on pay-as-you-go, measure steady-state throughput for a few weeks, then move predictable baseline load to PTU and keep bursts on PAYG.