Question

A retail company runs a Bedrock-based product-description generator that processes catalog updates only during a predictable 4-hour window each night. During that window, throughput demands are high and consistent, requiring guaranteed capacity to finish before the store opens. For the remaining 20 hours per day, the model is completely idle. The team currently uses a no-commitment Provisioned Throughput allocation that runs 24/7, and finance has flagged the cost as excessive. Which approach minimizes cost while still meeting the nightly deadline?

Accepted Answer

Programmatically create a no-commitment Provisioned Throughput model unit before the nightly job and delete it when the job completes, running on-demand the rest of the time.

Answer

Purchase a 1-month commitment Provisioned Throughput plan to reduce the hourly rate and keep it running continuously.

Answer

Switch entirely to on-demand InvokeModel calls for the nightly batch and rely on automatic scaling to meet the deadline.

Answer

Submit the nightly catalog as a Bedrock batch inference job to guarantee completion before the store opens.