Is a compute crunch coming?

We estimated trends in global inference capacity and found that token demand appears to be growing much faster than supply.