Three key insights you may have missed from theCUBE’s coverage of the Supermicro Open Storage Summit interview series
SiliconANGLE Victoria Gayton
AI storage is becoming the real AI bottleneck, not the model. The big shift is from experiments to production, where data access, governance and inference costs decide the outcome.
Based on reporting by SiliconANGLE, Victoria Gayton — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The Supermicro Open Storage Summit coverage keeps circling back to the same point: AI is now an operations problem. The flashy part may still be the model, but the hard part is making the storage stack fast, shared and dependable enough to support real business use. Rob Strechay of theCUBE Research put it plainly: the companies making progress are less focused on models and more on operationalization, with data teams, infrastructure teams and business stakeholders lined up around measurable outcomes.
That shows up first in inference economics. Greg DiFraia of Scality said the job is to keep GPUs busy while handling much larger pools of data that do not belong in flash. Anat Heilper of Vast Data added another wrinkle: expanding agent contexts are driving demand for key-value cache storage, which can act as a substitute for compute when hit rates are high. In that setup, storage is not just where data sits. It becomes part of the performance trick.
The second insight is that production AI cannot live on generic infrastructure. Moiz Kohari of DataDirect Networks tied data movement speed to financial risk and capital availability, using firms like BlackRock and State Street as examples. Vince Chen of Supermicro described a partner-led approach built around vertically integrated, pre-validated configurations in different sizes. And Ruhi Sehgal of Nutanix pointed out that once AI moves beyond pilots, the familiar chores return: multi-tenancy, security, resilience and performance tuning for a much larger user base.
The third thread is data access and control. Greg DeMichillie of MinIO argued that open table formats like Iceberg have become normal enough to support storage-tier integration, not just database design. Around that, the interview series paints a wider stack: Supermicro systems, Hammerspace orchestration, Cloudian object storage and Seagate retention for unstructured data that has to stay under control while moving through its life cycle. Peter Sjoberg of Cloudian said that governance is the point — keep the data protected, safe and secure, because AI will keep asking for more of it.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that gets ignored until the budget hurts: storage, control and operational plumbing. The market loves to talk about models as if they float above infrastructure; they don’t, and the invoice always arrives with the data layer attached. Open formats and pre-validated systems are the sensible answer here, which is exactly why they’re less glamorous than the pitch decks pretend to be.
Read more about this at: SiliconANGLE