I figured that the truly innovative aspect here was the pricing model. Other similar services expect your "containers" to run 24/7. Lambda runs your container for as long as the request takes, and only charges you for that. It's like what Heroku was originally trying to do with its Aspen Ruby stack, plugging everyone's apps as handlers into Passenger and only (quickly) spinning them up when they get requests.

You've been able to do the same thing forever by just writing everything as (non-F/non-WS)CGI scripts, or as isolated PHP servlets. But nobody ever charged for those by tracking CPU time, rather than just imposing a monthly fee + caps.