Artificial intelligence is placing unprecedented pressure on enterprise storage, creating a volatile landscape that defies familiar innovation cycles. For HyperFRAME analyst Don Gentile, the challenge is clear: storage must evolve to feed hungry AI pipelines, but severe supply constraints are forcing organizations to rethink their infrastructure strategies.
Storage has become a critical pressure point in the enterprise AI pipeline. The demands bearing down on it are arriving faster than most organizations ever expected.
"The evolution has been about how we feed those AI pipelines and how storage becomes an active participant in the AI process," Gentile told The Forecast at Nutanix's recent NEXT 2026 event in Chicago.
That shift is reshaping the role of storage in the data center.
Starving The Machine
For years, the storage conversation was mostly about capacity and cost. GPU-driven AI workloads changed the math, Gentile said. Those processors execute the parallel computations behind AI model training and inference. They require a continuous stream of data. If storage cannot deliver it at the required pace, the GPU sits idle and the AI pipeline stalls.
According to a VentureBeat survey of enterprise AI teams, average GPU utilization, including storage, was stuck at just 5% in early 2026.
Enterprises responded with purpose-built strategies, Gentile said. They built tiered storage environments calibrated to workload heat. Hot-data flash handled active inference. Cold data archives remained air-gapped for data protection. Object storage became a common platform layer across environments, with S3 establishing itself as a universal standard. Gentile saw vendors snapping their platforms directly into NVIDIA architectures and positioning storage as a first-class citizen in the AI data center.
No Magic Store
Even as vendors and enterprises adjusted to AI-driven demand, related supply chain challenges arose to slow their efforts.
Demand for NVIDIA GPUs surged faster than manufacturing capacity could scale. Advanced memory components, critical to AI inference and training, are produced by a small number of manufacturers with lead times sometimes stretching beyond 40 weeks.
The result is that the memory and GPU markets are running 18 to 24 months deep on constraints. Prices keep climbing. GPU compute effectively rationed to the largest players with long-term contracts secured well in advance, Gentile said. Recent reports show some enterprise organizations are stuck with “almost zero ability to get a GPU right now.”
For many organizations, the most immediate response is to squeeze more performance from the infrastructure they already own. There's simply nothing else available to buy.
"I have to get more value out of my existing resources," said Gentile. "There's no magic store I can go to to purchase more capacity like that."
Beyond that, organizations are weighing a short list of difficult options. They can postpone AI projects, use software solutions to extend the useful life of current capacity, or turn to neoclouds. These specialized cloud providers pre-secured GPU capacity and now offer it as an on-demand service. None of those options is painless.
The Two-World Problem
Those constraints are arriving just as the definition of storage itself is changing. What was once a device sitting in the data center has become a distributed layer that coordinates and governs data across on-premises environments and cloud infrastructure, Gentile explained. New storage platforms emerged to support this model, including disaggregated architectures and global namespaces that track where all of an organization's data sits.
Gentile's vision looks something like the catalog model cloud computing already offers, where systems reach for the service a workload needs and access it wherever it lives. Whether an organization wants to move data or keep it in place for regulatory or privacy reasons, the infrastructure should accommodate either decision.
"The goal would be a common operating model," Gentile said. "Being able to manage that with a consistent model is all about the efficiency of the team, right? They shouldn't have to have these two worlds."
Private cloud is one answer to that problem. Gentile described it as a rational response to uncertainty. Organizations can't always predict where technology is heading, so infrastructure has to be flexible enough to absorb new capabilities and add scale as they emerge.
"So, it's more future proofing," he said.
Sovereignty and compliance requirements reinforce the logic, Gentile said. Regulated industries can't migrate everything to a public cloud. Sensitive data must remain on site and under direct control. But with a private cloud, the same management model that runs in a hyperscale environment can run on-premises.
The Long Fuse
Security adds yet another dimension to storage's expanding responsibilities. Since the pandemic elevated cybersecurity to the top of CIO priority lists, the threat landscape has grown. Cybersecurity and risk management were the top priorities for CIOs for a fourth consecutive year in 2025, according to a Gartner survey of more than 2,000 CIOs.
A hidden danger in that landscape lies in the storage layer itself. Bad actors are already executing harvest-now-decrypt-later strategies, Gentile said. They collect encrypted enterprise data today with the intention of breaking it once quantum computing renders current encryption obsolete. CISA, the NSA, and NIST warned of the issue in 2023.
For organizations with sensitive data in archives or long-term backups, that's a storage problem with a long fuse. Defending against it starts at the infrastructure level, Gentile said. That means immutable backups and verified recovery points, tested regularly to confirm the integrity of the restored data.
"The ultimate question, I think, for companies is going to be, do you bring the compute to the data, or do you bring the data to the compute?" Gentile said. "And so that also will influence your storage decisions."
Related stories:
Editor’s note: Learn more about Unified Data Services Platform for Core, Edge and Cloud, and Governance and Cost Control for Agentic AI with Nutanix Agent Gateway
David Rand is a business and technology reporter whose work has appeared in major publications around the world. He specializes in spotting and digging into what’s coming next–and helping executives in organizations of all sizes know what to do about it.
Ken Kaplan contributed to this story. He is Editor in Chief for The Forecast by Nutanix. Find him on X @kenekaplan and LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
