How do the gpu training clusters rank against the traditional “largest supercomputer” metric? Or is that comparing apples to oranges? I think by some measures

How do the gpu training clusters rank against the traditional “largest supercomputer” metric? Or is that comparing apples to oranges? I think by some measures XAi already has the largest (200k H100 equivalent) and already constructing the 1 million gpu cluster.