Maybe Chris Lattner leaving and creating Mojo also didn’t help in that regard.
Swift for TensorFlow was a cool idea in that time …
Lattner probably left because Apple didn't give the team any breathing room to properly implement the language. It was "we must have this feature yesterday". A lot of Swift is the equivalent of Javascrip's "we have 10 days to implement and ship it":
https://youtu.be/ovYbgbrQ-v8?si=tAko6n88PmpWrzvO&t=1400
--- start quote ---
Swift has turned into a gigantic super complicated bag of special cases, special syntax, special stuff...
We had a ton of users, it had a ton of iternal technical debt... the whole team was behind, and instead of fixing the core, what the team did is they started adding all these special cases.
--- end quote ---
It remains to be seen how much Mojo has learnt from that effort.
NVidia, AMD and Intel now have doubled now into giving Python GPU JITs, and Julia, the same capabilities as their CUDA, ROCm, and SYSCL offerings with C++.
With Julia and Python having their 1.0 long behind them.