Data Infrastructure Not Ready for AI? Hitachi Vantara Thinks It Has the Fix.

At a Singapore media briefing this month, Hitachi Vantara’s chief technology officer for Asia Pacific, Matthew Hardman, pulled a journalist from the audience and asked him to recite the alphabet. Repeatedly. It was, improbably, the most clarifying moment in the room.

The exercise was a live simulation of storage downtime. Six nines of availability, which is the standard most enterprise storage vendors offer, translates to about 31 seconds of downtime per year. The journalist recited the alphabet four times. Eight nines, the availability rating Hitachi Vantara claims for its newly announced VSP One Block High End? The journalist barely got through the first few letters. That’s 314 milliseconds — the flap of a hummingbird's wing. Hardman let the silence do the rest.

The real point wasn’t the number. It was everything downstream of it: databases that unsynchronize from hosts, applications that block, and servers that cascade. “It's not the machine,” Hardman said. “It’s everything attached to it.” That’s the argument getting buried under GPU headlines across the Asia Pacific right now. And it’s the one Hitachi Vantara is betting its expanded VSP One portfolio on.

Along with the Block High End, the company announced local availability in Singapore of VSP One Object, its enterprise object storage platform. Together they form what the company calls a unified data foundation. Whether you buy that framing depends on what you think the actual problem is.

When everyone has AI, nobody has infrastructure

Hitachi Vantara's own research puts the tension in sharp relief. 96% of Singapore organizations are already running AI in some form. Only 23% rate themselves as having a strong readiness to achieve ROI from it. That's a lot of expensive infrastructure producing uncertain returns. The company argues that the gap lies not in the models or the GPUs, but in the storage layer underneath them.

“Let everybody talk about AI,” Hardman told the briefing. “I want to talk about what runs it.”

Hitachi Vantara built VSP One Block High End for environments where that argument lands hardest, including financial platforms, real-time analytics, and large-scale AI inference running on Oracle or SQL Server. At 50 million IOPS and eight nines of availability, it’s engineered for workloads where even milliseconds of downtime are measured in transactions, not inconvenience. The platform scales to 24-nines availability across a three-data-center architecture for organizations that must prove uptime to regulators and engineers.

The object storage argument

The Block High End is the product story most enterprise storage engineers will recognize. VSP One Object is the one with longer legs.

Unstructured data now accounts for more than half of what most organizations store. That includes images, logs, scan files, geospatial data, etc. None of it with a schema, all of it accumulating faster than anyone planned for. Object storage is the architecture built to handle it, and the market is crowded.

What Hitachi Vantara is claiming as differentiation is native, on-premises support for Amazon S3 Tables and Apache Iceberg, a specific claim it’s prepared to defend. “We are the first enterprise object platform to deliver S3 Tables with an added Iceberg layer natively on-prem at scale,” Hardman said in a separate interview, “rather than through cloud services and additional external metadata services.”

It’s a meaningful distinction for the CDO running a data lakehouse strategy without wanting every query routed through AWS. SQL queries run directly on object storage, AI training pipelines shed the cloud round-trip, and the economics shift considerably at the petabyte scale.

The governance story runs alongside it: embedded PII detection and storage immutability built into the platform. Joe Ong, Hitachi Vantara’s vice president and general manager for ASEAN, noted that the company's research shows 52% of Singapore respondents say the complexity of their data makes it harder to detect a security breach. It’s a number that compounds as AI adoption accelerates.

The skeptic’s question is about metadata. Petabyte-scale data lakes have a long history of creating a new metadata management bottleneck. Hardman’s answer is architectural: “Iceberg’s schema changes update metadata pointers, not the data itself, keeping the catalog lightweight even at very large scale. VSP One Object’s native implementation brings those open-standard capabilities into an enterprise-grade, on-premises platform — which is why metadata scalability is a core part of the ‘industry-first’ claim rather than an afterthought.”

The GPU elephant in the room

For organizations scaling GPU training pipelines, the architecture question goes beyond block or object storage. Disaggregated, RDMA-native storage has attracted serious attention from teams pushing high-throughput AI workloads, and some competitors have moved hard in that direction.

Hitachi Vantara’s answer is that VSP One isn't the complete picture for GPU-scale AI. Instead, the fuller architecture lives in its Hitachi iQ solution suite, which layers a high-performance parallel file system on top of VSP One Object to feed GPU clusters at speed. “VSP One Object provides the economic scale meeting exabyte demands,” Hardman explained. “Atop that sits a high-performance parallel file system that can tier the required data at high speed from the object storage tier and feeds that into the GPUs for training.”

On RDMA-native disaggregation, the company isn't chasing it — at least not yet. “We’re very, very focused on what we do with the controller,” Hardman said. “The controller is where we can really go ahead and actually tweak for maximum requirements.” It’s a deliberate bet that the Asia Pacific enterprise, whether it is a bank, government agency, telco, etc., values reliability and ecosystem consistency over raw throughput benchmarks. That bet may prove right, but may also age.

The mainframe edge

The more surprising thread in Hitachi Vantara's positioning is the mainframe. Most vendors either ignore it or treat it as a migration problem waiting to be solved. Ong noted plainly that Singapore government agencies are still running mainframes for core operations. Across ASEAN banks and large government installations, the mainframe is live production infrastructure, not a footnote, and VSP One's support for both open systems and mainframe environments matters for enterprises trying to make that data useful for AI analytics without destabilizing the systems running on top of it.

The architecture reality is unglamorous but important. EBCDIC-to-UTF-8 translation — the format mismatch that makes mainframe data awkward for modern AI tooling — happens outside the transactional path, during ingestion into analytics environments, not in the storage layer itself. “Mainframe systems remain performant while downstream AI models receive data in usable formats,” Hardman said. The IBM z17's dedicated AI coprocessor means the most modern mainframe installations can handle inference closer to the data anyway. Hitachi Vantara’s role is to ensure storage consistency across both worlds, not to own the compute layer, but to make sure it doesn't become the constraint.

One platform, one management layer

Threading through all of it is VSP 360, the unified management layer Hitachi Vantara argues is its most underappreciated differentiator. “Gone are the days of a unique set of services for one area and another for another,” Hardman said. “Your IT teams can't manage 70 to 100 different tools. You've got to have a few service tools that give consistency of experience across the entire data architecture.”

It's an operational argument aimed squarely at the stretched infrastructure teams managing the intersection of legacy debt and new AI ambition. Wendy Koh, Hitachi Vantara’s vice president and general manager for Asia Pacific, anchored it with market position: the company ranked number one in the high-end external OEM storage market share according to IDC, with top positions across Australia, Hong Kong, India, and Korea. “That speaks to the trust customers have in supporting mission-critical environments,” she said.

The strategic framing Hardman returned to throughout was “data singularity.” Unified data platforms spanning block, file, object, mainframe, and cloud under a consistent services layer replace discrete storage silos. Competitors, he acknowledged, are now arriving at the same conclusion. “We’ve led the charge in providing unified platforms. The advantage is we’ve been doing it for two years.”

Two years is a meaningful head start in enterprise storage. It’s also a short runway if the market moves faster than the roadmap. For storage engineers running AI ambitions on infrastructure never designed for them, VSP One is a serious, well-engineered answer to a real problem.

The hummingbird demo will be forgotten by next quarter; the 77% of AI investments still waiting to deliver ROI will not. And that's the number Hitachi Vantara is really selling against, whether it says so or not.

Image credit: iStockphoto/Nikolay Evsyukov