I didn't explain that well--initially Twitter was on RoR and sharded MySQL, which wasn't scalable. Obviously they have done a lot of scaling work and can handle their load now.
If I were implementing Twitter today, I'd probably do Elixir and Cassandra. I don't generally leap to go NoSQL but Twitter's structure is particularly suited to Cassandra. Elixir gives the ease of development I associate with Ruby with the scalable, decentralizable BEAM. I'd also consider going with Erlang straight up--it may not be as pretty as Elixir, but there's a lot to be said for maturity in a language.
But I'm saying this because I have the benefit of newer technologies and having seen Twitter's mistakes. From a technical perspective, I think Twitter is pretty well done.
Curious, what would be more scalable than Twitter's existing JVM based architecture? 3 years ago[1] they were able to handle upwards of 140 (edit) thousand tweets per second without any latency -- that's pretty ridiculous.
[1] https://blog.twitter.com/2013/new-tweets-per-second-record-a...