[jimmychangas]

Yeah, absolutely, if your engineers decide to adopt serverless due to hype or just to improve their own curriculum, you are going to spend a lot more on infrastructure than you would by provisioning VMs or running containers. By being selective about which workloads are eligible to become FaaS and doing a little of optimization, however, you can cut some costs and avoiding overprovisioning, with automatic and efficient scalability.

I believe that, in most cases, it is better to control the exit costs of your architectural decisions than to avoid lock-in at all costs.

[Coredalae]

This is why the serverless framework (trying to give the possibility to deploy to any cloud provider) is so important. Some repetitive simple tasks are extremely well suited for lambda/functions /whatever name and some tasks are suited for big machines with gazillion terraflops and terabytes of ram. The job of the engineer is to know what his software needs, and what the most optimal path for this is.

Business is ever changing. This is just another step