Overview of Model deployment options on AWS, which also takes into account cost implications: LinkedIn.
It’s important to note that the comparison may not be entirely straightforward, as the slides alone do not provide a full picture. However, they do reveal some significant cost variances that may need to be considered. In particular, AWS Lambda stands out as a substantial expense, with the potential to exceed the costs of even the priciest foundation model inference. To clarify the $50 figure mentioned in screenshot #1, the author kindly provided further insight in a follow-up comment on LinkedIn:

Roughly lambda pricing is:
($0.0000166667\/Gb*s)*10Gb*0.3s * 1M = 50$ per 1M tokens
given 1 token is generated in 300ms