This feels like it's playing games with words a lot. What they've built here is a service which is perfectly cacheable: the same content is served to every user, and can be precomputed in advance. This allows them to run their page generation code in a batch generator instead of during processing of the http query.

It's certainly a very good idea to serve as much of your content from cache as is possible, and I don't think anybody would be surprised by that statement. However, they still have the same amount of page generation code, they're just running it at a different time. I'm not sure I'd call that "no back end". I'd call it an interesting choice of tradeoffs to optimise query latency.