Siloed and Underutilized What We Are Hearing From Customers Deploying large GPU computing environments (DGX, HGX, etc.) Time-to-value critical, as is infrastructure cost and constraints External storage systems add cost, consume power, take time *8 or 16 SSDs per Server 16 x 30.72 TB = 491.52 TB GPU Server Local NVMe Storage Largely Goes Unused • Not being used for data because it is siloed • Today: GPU server may have 60 – 500 TB* of storage • 2025: 2 PBs of storage per server will be possible