The Future Of Storage For AI And HPC: The Requirements, They Are A Changing
HPCwire, Monday, October 20th, 2025
When the great Yogi Bera once quipped, 'The future ain't what it used to be,' he probably had no idea he was describing the future of data storage for AI.
Generative AI is barely five years old, but the storage requirements are changing rapidly as organizations seek to build scalable data foundations to get AI apps from model training into production.
As the initial fascination in AI model training starts to wind down and organizations turn their attention to putting AI inference and agentic AI workloads into production, they're facing some architectural and technological uncertainty.
Some of the time-tested HPC kit that came in extraordinarily handy for training large foundation models, such as speedy parallel file systems feeding data to GPUs over high-speed networks and RDMA, aren't necessarily what you need to run complex, large scale AI inference or agentic workloads.