Not sure, but biggest one so far is probably Deepmind, from the UK.

Magic Pony (machine learning for reverse-compressing video data, re-adding data that was removed during the compression process) is pretty cool.

I'm thinking from the perspective of "I'm sick of working at your standard SaaS. I'd like to see what else is out there for software developers."

Isn't that just a buzzword-heavy way of saying "better video compression"?

No. AFAICT, it's perceptual reverse-compression: ie, you look at a bad mp4 of a brick wall, and even though it's shitty, you mentally figure out the mass of pixels is a brick wall. Imagine if the computer actually re-added a crisp brick wall there.

Video compression algorithms already optimize for perception rather than objective measures like PSNR, so I still don't see the fundamental difference.