What the next month could cost
NeMo replaces GPU ASR. Other costs follow the August–September average.
How this forecast is calculated and its limits
Start with the existing 60-day billing model, normalized to October’s 31 days. Remove retired GPU compute, NeMo development compute and the historical NeMo production allocation. Add NeMo production at the selected average task count.
Every NeMo task requests 2 vCPU and 8 GiB RAM. Observed billed rates are $0.05056 per vCPU-hour and $0.00553 per GiB-hour, or $0.14536 per task-hour. Production currently runs 2 tasks and can scale to 6.
Cloud costs scale by the number of days. The LINE monthly allowance stays at $53.25. ASR savings compare a full period of GPU compute with a full period of NeMo compute. The historical average already includes NeMo usage; it is used only to reconcile the total forecast.
Assumes the retired GPU remains off and NeMo development remains at zero tasks. Storage, network, logs and all other services retain their historical averages. No further cleanup savings are assumed.
AWS was refreshed on 1 October. The complete-day baseline is still September 1–29 because September 30 billing is incomplete. GCP retains its September 1–29 export. The full September projection uses that 29-day cloud spending rate, with a fixed monthly LINE plan.
September ownership weights were applied to August charges in the original model. Product allocation and individual adjustments are estimates. This is a migration scenario, not a complete forecast of every resource.
Includes the same products and providers as the historical report. Unselected GCP products, unpriced Azure OpenAI and unverified app-store fees remain outside scope. Existing tax and credit treatment is unchanged.