Three years ago, storage conversations were straightforward. How much can a data lake hold? How fast can it be read and written? How efficiently can it be backed up?
For AI workloads, storage was a passive repository, a system of record that held data until something needed it.
Agentic AI has changed that. These systems do not just follow instructions; they act, learn, and adapt on their own, negotiating contract terms, managing supply chains, optimising production lines, and resolving customer issues with minimal oversight.
Unlike traditional AI, which answers questions or spots patterns, agentic AI reasons and decides using context, history, and an organisation’s accumulated knowledge and the quality of those decisions depends on the infrastructure beneath them.
Executives at enterprise-grade data storage companies that Entelechy Asia spoke to explained that AI agents have transformed the way storage operations work.
“Storage is no longer just a depository for data,” said Lim Hsin Yin, Vice-President of Sales for Asean at Cohesity, adding that it is becoming the strategic platform that delivers trusted data to people, applications, and AI agents.
From repository to intelligence layer
Agentic AI cannot work from isolated records. It needs to see how information connects, how it has changed over time, and what shaped earlier decisions. A contract carries a trail of revisions and negotiations; a support ticket links back to prior conversations and what ultimately solved the issue.
“Raw data without history is just a snapshot, and a snapshot can be misleading. We index time-series versions of files, so an agent asking ‘what changed and when’ gets a real contextual answer instead of a guess,” said Lim (left).
Storage is evolving from a system of record into a system of intelligence. “We are seeing storage transform into an active participant in AI workflows, continuously supplying governed, trusted data,” said Amit Deshmukh (right), AI Data Infrastructure strategist of Hitachi Vantara for Asia Pacific.
“The bottleneck is no longer compute alone. Increasingly, it is data accessibility, quality, and trust.”
Elaine Chan (right), Director and Head of AI Sales and Go-to-market of Asia-Pacific at NetApp, puts it more directly: “AI systems are only as good as the data behind them, and without a robust data platform, organisations can’t extract full value from their AI investments.”
For Synology, the shift means rethinking how data is structured.
Synology Country Manager Emily Oh (left) highlighted: “Metadata tells you what a file is, but an ontology tells you what it means. We need a layer that turns version history into provenance, folder labels into lifecycle states, and implicit connections into explicit relationships agents can understand.”
Non-negotiable capabilities
Autonomous decision-making demands capabilities storage never needed before. Top of the list: real-time, unified access. An estimated 80 per cent to 90 per cent of enterprise data is unstructured, scattered across clouds, on-premises servers, and edge locations, a fragmentation that is the enemy of agentic AI.
“Agents cannot function effectively when data is fragmented across distributed IT environments,” Chan said. “Enterprises need infrastructure that delivers low-latency retrieval, continuous indexing, and metadata enrichment across every environment.”
Trust and control matter just as much. As AI systems act autonomously, the risk of error or compromise grows.
Cohesity’s Lim said: “AI agents should never retrieve data that a human user would not have access to. Fine-grained, identity- and role-based access controls need to apply alongside data privacy and industry compliance requirements.”
Immutability matters too to prevent tampering and giving organisations a safety net, letting them restore data to a verified point in time if an agent errs or is compromised.
“We treat an AI agent as a new class of user, a non-human actor that needs its own identity, permission boundaries, and audit trail, exactly as an employee does,” Oh stressed.
Preserving institutional knowledge
Deciding what to keep, archive, or delete used to be a cost question. Now the priority is preserving institutional memory.
Hitachi Vantara’s Deshmukh pointed out: “The real risk is not storing too much data. It is deleting institutional memory that could help AI systems make better decisions in the future. A customer interaction from five years ago may seem unimportant today, but it could be the key to understanding behaviour or resolving a complex issue tomorrow.
That does not mean keeping everything – duplicate, obsolete, or low-quality data adds noise and slows AI reasoning. Instead, enterprises are turning to intelligent lifecycle management, moving older, less-accessed data to lower-cost tiers while keeping it indexed, discoverable, and governed.
Data readiness, the biggest hurdle
The biggest obstacle is not the AI itself, it is the foundation it runs on.
“Most organisations do not have an AI problem as much as they have a data execution problem. They struggle to bring fragmented data together, apply consistent governance, and make information usable at scale,” said Deshmukh.
Chan agreed that data readiness is the real blocker: high-value data scattered across legacy systems, multiple clouds, and different governance frameworks makes it hard to move from AI pilots to production at scale.
Storage designed correctly becomes a competitive advantage since it leads to faster decision-making, lower compliance risk, and AI that scales without losing control.
Over the next five years, the line between data and intelligence will blur further. Storage will move out of separate consoles and into open standards accessed directly by the agents that rely on it.
But the principle holds: beyond terabytes and throughput, the true measure of storage is how well it preserves the knowledge that powers autonomy.
As NetApp’s Chan put it: “Enterprise storage will no longer be judged solely by how well it stores data, but by how well it helps enterprises understand, govern, activate, and defend that data.”
By Edward Lim
TOP IMAGE: Generated by Dola AI
Entelechy Asia 