There is an article comparing python based backend framework FastAPI vs Flask for ML project. Here is a comment in the article about FastAPI:

The major disadvantage of the FastAPI framework is that it is expensive. This does vary according to the country you use it in and how many API calls you make each month. Overall, though, the cost is high.

What I don’t quite understand is that if FastAPI is good at creating API but it is expensive to call the API developed, then why it is still gaining popularity in ML? I know there are some paid libraries integrated into the FastAPI. Is it for sure those paid libraries will be called in API calls for a ML project developed with FastAPI? How to estimate how much it costs to use a ML project developed with FastAPI?