That’s two Petaflops per C480ML which is, lets be honest, the way better server than the Nvidia DGX.
One can argue over Arista versus Cisco Nexus but the latest Nexus ASICs are pretty brutal and a fine piece of engineering all around.
So the scale is really only limited by how many flashblades and 480ML one can plug into the Nexus layer. And since Flashstack supports Cisco ACI spine-leaf architectures this is for all intents and purposes unlimited.
Solid work Cisco and Pure. Nothing mini about this, though.