The simple reason is that Mercurial and git were way too slow for large repos, and FB wanted a monorepo for productivity reasons. It's cheaper to fund this project than reduce engineering productivity by even 1% at the scale of FB.

Google did exactly the same thing, but hasn't open-sourced their tools.

Projects to drive incremental productivity make not a lot of sense for small companies, but become immensely valuable at very large companies. A 1% improvement if you have an engineering staff of even just 10k is worth 100 engineers, and FB is much larger than 10k.