Instagram were heavily using Python (and Django) when acquired by Facebook, whereas Facebook proper were a PHP shop.
The most interesting aspect really is that both companies have run in to the limits of performance with dynamically typed languages and had to implement their own language runtimes, AOT compilers, and JIT's.
Do we have proof that they had performance problems? At least in Python you can write a C extension for a problematic code path which would probably take much less time.
I think what they are looking for is to reduce the amount of machines they need. Which in the end is more money in their pockets. Making the website faster for the end user is a welcome side effect.
There are articles like https://instagram-engineering.com/dismissing-python-garbage-... which seem to confirm some problems. One of the people involved with this Cinder project was an original author of https://github.com/microsoft/Pyjion (it's now being revamped by a new dev.) Dino had deep .NET/CLR implementation experience and had been the lead for IronPython which I think spawned the interest in perf improvements through Cinder. I hope this effort is more successful for him.
The most interesting aspect really is that both companies have run in to the limits of performance with dynamically typed languages and had to implement their own language runtimes, AOT compilers, and JIT's.
Static typing is the best way forward at scale.