zlacker

Google kind of has done this with Coral: https://coral.ai/about-coral/

These TPUs obviously aren't the ones deployed in Google's datacenters. That being said, I'm not sure how practical it would be to deploy TPUs elsewhere.

Also, Amazon's Infinera (sp?) gets a fair bit of usage in industrial settings. It's just that these nvidia GPUs offer an amazing breeding ground for research and cutting edge work.

>>sargun+(OP)
The Coral line of products targets embedded/IoT inference applications, and are on the low end as far as processing power goes. AFAICT the recent surge of demand, mostly fueled by LLMs, is for GPUs on the opposite end of processing power and RAM size.