As enterprises race to operationalise AI, the real battleground is shifting beneath the surface, into the data layer. While much of the spotlight remains on large language models and AI agents, the next wave of competitive advantage may depend less on model sophistication and more on how intelligently enterprise data platforms evolve.
In conversation with iTNews Asia, Tirthankar Lahiri, senior vice president, Mission-Critical Data and AI Engines, Oracle, shared insights on how AI-driven data management is reshaping enterprise platforms and why 2026 could be a tipping point for intelligent data infrastructure.
One of the most significant AI-driven data management trends Lahiri expects to accelerate is the large-scale vectorisation of enterprise data.
“Vectors are incredibly powerful constructs that capture the semantics of data. We are at the very early stage of understanding their power,” he added.
While vectors are widely used in natural language processing and image recognition, he believes the next frontier lies in applying them to enterprise datasets, from banking systems to billing platforms.
Rather than moving data to standalone vector databases, he predicts native vector capabilities will become embedded within enterprise platforms.
“We don’t think vector-only databases will exist too much longer. Enterprise platforms will have native vector support, enabling any application that runs today to be AI-enabled,” he added.
From vectorising business data to embedding trust directly into database architecture, Lahiri outlined a future where enterprises must rethink not just AI models, but the very foundations of their data systems.
From systems of record to systems of intelligence
As enterprise platforms evolve from systems of record into systems of intelligence, Lahiri identifies two foundational capabilities required to make the shift real - native vector support and data explanation for AI.
Enterprise schemas today are often cryptic, machine-generated structures. To address this, databases must allow organisations to annotate and explain their data in human-understandable terms.
If a human being cannot understand your data, neither can AI.
- Tirthankar Lahiri, senior vice president, Mission-Critical Data and AI Engines, Oracle
The goal is ambitious - transforming a passive database into a knowledge base. He explained that it means embedding semantics, enabling AI-friendly explanations, and ensuring models can interpret business meaning, not just raw structure.
Operationalising AI: A survival test
Drawing parallels to past technology inflection points, Lahiri compared today’s AI shift to the rise of the internet and cloud computing. “Companies that experimented with the internet are gone. The ones that embraced it thrived.”
The same pattern, he believes, will repeat. “In this AI shift, it’s a survival test,” he said.
Organisations that embed AI into productivity, efficiency, and customer experience workflows will outpace those lingering in pilot mode.
On the future of enterprise platforms, Lahiri advocates for converged architectures and not fragmented stacks stitched together by orchestration layers.
“If you multiply data silos, you multiply operational costs and weaken security. Your security is only as strong as the weakest link,” he warned.
Instead, he recommends a converged architecture powered by object storage and open table formats, allowing multiple repositories but governed through uniform policies.
“The future is multi-repository. But you need a single pane of glass to manage it,” he added.
The hidden fragility
One of the biggest underestimated risks in enterprise architecture today, according to Lahiri, is the persistence of hard data silos.
“Many large enterprises do not know what data they have. There is no global view,” he said.
In industries such as fintech, mortgage, banking, and credit card systems often operate in isolation, with separate governance and workflows. “That fragmentation is a big problem. AI will accelerate the need to solve it.”
While multiple repositories are inevitable, what must disappear are the hard walls that prevent coordinated access and governance.
“You need a global view of data. Not everyone should access everything, but when required, access must be granted through the right workflow,” he explained.
As AI agents operate directly on enterprise data, governance becomes critical. Lahiri is wary of soft controls. “Guardrails are insufficient. You can’t stop AI from going rogue with instructions alone,” he said.
Security policies, he argues, must be embedded in the database, not just the application layer, so agents cannot bypass controls.
The road ahead
Looking ahead, Lahiri sees a parallel evolution between model innovation and data innovation. In the near future, business users may no longer write SQL queries. “They’ll ask natural language questions,” he said.
But natural language is ambiguous. Translating intent into precise database queries will require curated knowledge layers, giving rise to a new discipline.
“Just like we have data engineering today, we’ll have knowledge engineering.”
“We are still early,” he reflects. “AI is powerful for first drafts, blueprints, acceleration but trust must be engineered into the system.”
For enterprises, the message is clear: vectorisation is not a niche innovation. It is the foundation for turning enterprise data into semantic, searchable, and intelligent assets.
And those who embed that intelligence deeply into their data platforms will not just innovate, they will survive.
