That's crazy. Am I wrong in thinking that this is a scary bubble?

I don't know if you're wrong, but you're certainly not alone. If you read history, or you're old enough, you'll know that we've already had at least two AI winters (mid 70s to early 80s and early 90s to early 00s), depending on how you count them, and the hype around machine learning is almost word-for-word the same as the hype around expert systems was last time around. This bubble is ripe for bursting.

If I remember correctly, this salary was offered to Andrej Karpathy, who's easily the top 10 in the world in DL field. So, no, it's not a bubble. Top experts have always been paid well.

DL is over hyped. This idea that people from a few AI labs are producing everything is basically SF VCs sniffing the farts of AI departments near SF telling them that only their labs and the labs of a few friends are magically producing pixie dust. It must be nice to be so well connected. When the bubble pops very soon, 1 year on a 500k salary will have been enough for them anyway.

I think that there are 3 or 4 companies who would have happily paid that for say Alex Krizhevsky ("Alex Net") a few years ago, and it's pretty clear that would have paid off (it certainly has for Google).

Google makes $1m per average employee; it's not hard to believe they can make many times that on an AI researcher.