- Newest
- Most votes
- Most comments
While serverless computing and consumption-based pricing go hand-in-hand, it is important to recognize the difference between compute, storage, and database service offerings. If you look at serverless compute with AWS Lambda, it does not charge unless you invoke the function. But the same goes for EC2 which is literally you having full control of a server (i.e. not serverless). So the pricing model has more to do with the type of service than the serverless-ness of it. Amazon S3 is technically a serverless service because you don't have to think about or know what type or quantity of servers your data is stored on. However, you get charged every month based on the amount of storage you use. OpenSearch Serverless is simliar. Because you have a large vectorized database, storage costs will be incurred because the data has to be continuously stored somewhere and that is what you are paying for.
Amazon Bedrock on-demand FM invocations are compute services, and thus you won't get charged unless you use them. When you do, it is by the 1,000 input tokens, model, and other factors.
So while serverless and pay-per-invocation are a familiar relationship for services like Lambda, it depends more on whether your data is persistent that determines ongoing charges. For more information see: https://aws.amazon.com/bedrock/pricing/
answered 2 years ago
Relevant content
asked 4 years ago
asked 2 years ago
asked 9 years ago
- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated 4 years ago

"Because you have a large vectorized database, storage costs will be incurred because the data has to be continuously stored somewhere and that is what you are paying for."
I assume there are two components when you come up with units such as OCU: storage and compute. While you definitely should charge for storage all the time, I think you should charge for compute only when we use and not a flat charge per hour all the time.