Survey Finds Most HPC-AI Users Keep Data On-Premises for Sovereignty

An Intersect360 Research survey found that 68.2% of respondents use no public cloud storage for active HPC-AI workloads, and 56.3% use no cloud compute for those systems. Among respondents, 83.5% said data and compute must stay physically proximate, with none disagreeing. The article also warns that many-small-file AI access patterns strain legacy parallel filesystems, creating a performance problem for on-premises deployments.

Why it matters: IT leaders planning AI infrastructure need to weigh data residency requirements and storage performance together, since on-premises AI is becoming the default for regulated organizations.