Google hasn't retired any of its now ten-year-old V1 TPUs or any of the subsequent generations so it's unclear that AI infrastructure "needs" upgrades.
Of course, Google could be wrong about the economics of running old AI processors. Or maybe TPUs are special.
I do wonder how much of their inference is running on 10-year old TPUs. I’d imagine not much, if only because 10 years ago the justification for massive compute was quite different than today.
{"deleted":true,"id":50421569,"parent":50421485,"time":1794203692,"type":"comment"}
Citation needed.