Your Data Is King Now. Your Storage Still Works for the App.

Matthew Oostveen was in Japan recently, sitting across from bankers, when he told them what his company could now do with a mainframe. They made him say it twice.

“They made me repeat it, just really honing in,” recalls Oostveen, chief technology officer and vice president for Asia Pacific and Japan at Everpure (formerly Pure Storage). “Is that true? Is that accurate? And yes, it is.” For institutions that have poured decades of investment into systems most data platforms can’t even read, the claim was big enough to warrant the double-check.

That small moment of disbelief is the whole story of Everpure’s Accelerate 2026 announcements, compressed. Everpure is wagering that the hardest problem in enterprise AI is not the model, or even the GPUs sitting idle while they wait. Rather, it is that most organizations cannot say where their own sensitive data lives.

Oostveen explains: The enterprise CIO genuinely cannot answer the simplest question about their own shop: what sensitive data do we have, and where is it? “If you don’t even know how many systems you’ve got inside your organization from an accounting perspective, you’re probably not going to know where all the data is.”

The inversion: why data, not the app, is now the system of record

For half a century, one assumption held: data served the application. You built an app, gave it a database, and did that a few hundred more times until every system hoarded its own copy and its own private definition of “customer.” App sprawl bred data sprawl, and every handoff between systems demanded another pipeline, another reconciliation, and another meeting about whose number was correct.

Everpure’s counter-argument has a name: data primacy. Data primacy is an architecture in which data, not the application, becomes the system of record — self-describing, governed once, and read by every app and AI agent rather than copied into each one. The logic is simple: AI needs the cross-system picture in real time, not one copy at a time.

“Data is now the new king,” Oostveen says. “And you go visit the king. The king doesn’t go visit you.”

In the press release accompanying the announcement, Chairman and CEO Charles Giancarlo names the stakes plainly. “AI completely upends the traditional IT hierarchy; enterprises that do not shift from app-centricity to data primacy will fall behind,” says chairman and chief executive officer Charles Giancarlo. “Enterprises need to consolidate their fragmented enterprise data footprint into a real-time corpus of trusted intelligence.”

The anxiety is measurable. In a separate press release, Everpure cited a commissioned IDC survey showing 94% of IT leaders naming data quality as the “absolute determining factor” in whether their AI succeeds. It makes data readiness, not model selection, the variable most CDOs are actually fighting.

The 70% tax: where AI teams actually lose their time

The reason a storage engineer should care about a philosophy of data comes down to where the hours go. “Research is saying that your teams are probably spending something like 70% of their time on getting this data story right,” says Oostveen. It includes finding the data, cleaning it, cataloging it, and nursing the pipelines that move it.

Everpure’s answer is Data Intelligence, built on 1touch.io technology that they recently acquired. Data Intelligence auto-discovers enterprise data, classifies it, and wraps a semantic knowledge graph around it, mapping raw fields to what the business actually means by them — across cloud, on-premises, SaaS, and, unusually, the mainframe. That last environment is the differentiator that made the Japanese bankers lean in.

What it surfaces is the stuff nobody remembers owning. “It finds shadow AI, it finds the dark data, it finds the sensitive data that’s sitting inside an archive in cold storage, a dev environment,” Oostveen says, “things that even developers might have forgotten about.”

The compliance angle: a regulator with teeth

The governance pressure reads as almost regional. When Oostveen asked banking CIOs in Singapore how often they get briefed on regulatory changes, the answer floored him. “Every day is the answer. I was kind of shocked at that.” Across GDPR, CCPA, HIPAA and a thickening stack of local privacy law, a line drawn in the regulatory sand last year can leave you in breach this year through no fault of your own.

Data Intelligence scans continuously rather than once at deployment, so the governance picture reflects today’s estate, not last quarter’s. It also adds sovereignty controls that can block PII or proprietary data before it ever reaches an external model, with support for air-gapped and restricted-region deployments. For any CDO under data-sovereignty rules, that is the line between an audit you can defend and one you cannot.

The plumbing underneath: Data Stream, Purity Turbo, and Overdrive

None of this matters if the storage buckles, so the hardware got its own upgrades. Everpure Data Stream, now available, vectorizes enterprise data into AI-ready pipelines in days rather than the usual weeks or months, extending NVIDIA’s AI Data Platform reference design while keeping data on-premises with no copy-out to an external vector store. FlashArray//XL190 with Purity Turbo lifts backup throughput from 30 to 40 GBps under concurrent transactional workloads, delivered as a software update. And Evergreen//One Overdrive adds the dimension that storage-as-a-service always lacked, letting customers burst 25% above their SLA and pay only for the performance they use.

That last feature is the most candid thing in the lineup, because it concedes that nobody can forecast AI demand yet. As Oostveen puts it on behalf of every buyer: “I don’t know how much storage I need, and then I don’t know also how much performance I need.”

Asked directly in a written follow-up whether Everpure can quantify performance advantages over commodity object storage and local NVMe for AI workloads, Oostveen frames the case architecturally. “AI training and inference have different usage patterns,” he adds. Overdrive lets customers reserve capacity for sustained workloads and burst for peaks, while the Enterprise Data Cloud (EDC) framework provides a scaling path from FlashBlade//S for early AI deployments to FlashBlade//EXA for GPU-scale throughput and ultra-low-latency inference, without rebuilding the operating model or disrupting the environment.

Tail latency or cost per inference token versus open-source alternatives? That comparison remains the buyer’s homework.

The part worth focusing on

The skepticism writes itself, and Everpure mostly doesn’t dodge it. Oostveen admits the rollout is “a slightly different selling motion,” heavier on discovery, pre-sales and professional services “just by the very nature of the fact that you touch more elements within the architecture.” You are re-plumbing the building while people are still working in it.

There’s also the bet on MCP, the Model Context Protocol, which Everpure is leaning into just as some teams drift back toward direct REST APIs over performance concerns. Oostveen refuses the binary: “We’ve got a horse in both races.”

The sharper concern for storage engineers running heterogeneous estates — where Everpure sits alongside other vendor or legacy systems it doesn’t control — is what happens when agentic workflows start writing and executing infrastructure changes across the whole footprint.

In a written follow-up, Oostveen says the workflows are built to be cross-vendor, with governance policies enforcing strict guardrails and full audit trails feeding into customers’ existing service change management systems. The most specific assurance: the workflows default to human approvals for sensitive operations. For anyone worried about an AI agent making irreversible changes at 3 am, that last line is the one to hold onto — and to press on.

Overall, Everpure’s data-platform story is similar to what every storage vendor of consequence is telling. But what sets it apart is how the discovery layer actually reaches the silos that everyone else skips. Here, the mainframe claim that prompted the Japanese bankers to ask Oostveen to repeat himself is of great importance.

Everpure’s announcement is a sobering acknowledgment that the center of gravity in enterprise computing has moved, and most architectures were built on the opposite assumption: the application would always be sovereign, and that data would travel to meet it. The companies that adjust fastest get an AI estate that works; the ones that don't are still waiting for the king to come to them.

Image credit: iStockphoto/Apichet