Estimate cloud cost from measured billable work rather than assume a growth curve from user count. For representative learning and contribution tasks, record database reads and writes, listener updates and reconnects, storage and transfer, server compute, model tokens, search and generated media. Apply dated prices for the deployed services and region.

Compare workloads at 1, 10 and 100 times the baseline, first holding work per user fixed, then varying graph size, concurrency and AI use separately. Inspect broad queries, repeated generation and cache misses. Plan budgets, alerts and enforceable usage controls. Include these costs alongside existing engineering capacity, review, maintenance and support when costing the funded work. Compare projections with actual bills.