I have fallen back to identical jetty machines instead of lambdas for real time applications. Much more predictable for me (also my ops are remote io related, so a threading jetty is more cost effective).
And the provisioned concurrency does not win money wise either. You pay the fair ec2 price for the wait time.
For unpredictable loads where load time doesn't matter lambdas are great though. Especially when you can launch hundreds at a time.