There's storage that is attached to the GPUs. So that's driving very high performance liquid cooled storage that feeds GPUs directly. 

And there's network attack storage. It's very high capacity and space efficient, power efficient storage. We're the leader in high capacity storage for AI. 

And in Paris, Avi Shetty caught up with John Furrier again where he talks about the reality behind tokenomics. 

New metric for data centers is token per watt per dollar per token per watt.