VAST Data Platform

Separates the machines that serve data from the machines that store it, so capacity and throughput scale independently and every server can reach every drive. The pitch is that cheap high-density QLC flash becomes viable if a storage-class memory buffer absorbs the writes that would otherwise wear it out. It appears in commercial AI clusters rather than national laboratories, which is the pattern across this whole category now: the new storage architectures are being proven by companies training models, not by physics.