Not OP, but a pretty common reason is having a super-modular and hackable IDE that can be used to develop pretty much anything. Everything is JSON-configurable, all editors are webviews, so adding stuff like HTML rendering in Jupyter notebooks is almost trivial from a technical perspective. Fleet might be a step in the right direction, but still feels like a layer on top of IntelliJ, which is a beast in of itself, plus it is closed-source.

Also the approach of decoupling editors from the language support via LSP might be one of the biggest innovations in this space in recent years, IMO. Having a widely adopted and open protocol for language support effectively made Neovim, Emacs etc. a viable choice. It has spawned several high-quality LSP implementations, often directly supported by the compiler vendors, e.g. clangd or rust-analyzer.

Arguably Microsoft has been monetizing a bunch of services on top of VSCode too and they haven't always stuck to their own principles (see Pylance, a closed-source language server that only runs in official VSC builds), but the LSP itself was still a pretty big net positive.