New Delhi, Feb. 19 -- Enterprises worldwide are struggling to move AI projects from controlled pilots into production environments, a gap that has become one of the defining challenges in enterprise technology over the past year. At a recent industry event, TechCircle spoke with Chris Chelliah, Senior Vice President for Technology at Oracle Japan and Asia Pacific, about what has shifted in how businesses approach cloud and AI, why trust remains the central obstacle, and how Oracle's cloud architecture positions it differently in a crowded market.
Edited Excerpts:
What is the one enterprise assumption about cloud and AI that has flipped in the last 12 to 18 months?
The clearest shift is the realisation that AI needs to happen wherever customers need it to happen. You cannot send your data off to a remote cloud outside your country - or even outside your company - to get AI outcomes. AI and cloud are very related, but one of the critical success factors is that the cloud running your AI must be available wherever you need the AI to be. The expression "AI changes everything" is accurate, but it also changes everything everywhere. No matter where you are in your journey, that realisation is hitting customers.
Going back 18 months, people didn't know where to start. Everyone had used AI in a consumer way - running a query on one of those large language model interfaces - but they struggled to bring that into an enterprise context. The instant fear was: that's corporate data. How do I move data from my enterprise into these models?
Our approach flips that. Don't take your corporate data and move it to the AI. Do the reverse. Leave your corporate data where it is and bring the AI and the cloud to you. That helps us differentiate strongly in the market, because your data doesn't leave, which gives us a strong security story. We're also able to turn on AI across every Oracle touchpoint a customer has.
If you're not a large development shop and don't have a big IT department, you can switch on AI across our applications out of the box - it's already embedded. We have a broad suite of industry-vertical and horizontal applications covering ERP (enterprise resource planning) and supply chain management, among others. In the last 12 months or so, we've also introduced agents into those applications. We now have well over 600 agents available out of the box. You don't have to configure anything to activate an expense claim agent or an invoice reconciliation agent - they simply appear, ready to use.
Indian enterprises seem to be stuck at the pilot stage and struggling to move AI into production. Have you observed the same trend, and where do you think they're falling short?
It's not just an Indian phenomenon - that maturity is evolving globally over the last 12 months. Everyone touched AI from a consumer perspective about two and a half to three years ago. The last two years, everyone ran some small pilot. And the last 12 months have been about asking: how do we make that production-ready?
When you ask what's holding companies back, it usually comes down to trust. How can I trust the output at scale? What if the system produced a hallucination? That could cause a frustrated customer at best, or legal liability at worst. That concern compounds in an agentic world - where the system doesn't just generate an answer but actually takes action based on it. It's not telling you what to do; it's going to go and do it. That's been the big hesitation.
What Oracle brings into the equation is trust, and trust is built on data. If you have only a partial view of your data, your agent doesn't have full visibility. Our ability to automate the reach across all of a customer's data means the agent has a 360-degree view. We also enable what I'd call a human in the middle. Whether through our AI factory or our data platform, we give customers the ability to visualise the business process, use our agents within their own agents, and embed a human checkpoint at any point in the workflow.
What we're seeing now is customers approaching this in two ways. Some are rewiring and optimising - looking at an existing process and identifying which middle steps can be compressed. Others are completely reimagining the process from scratch, using data as the foundation. Customers have different maturity levels, and we have to help each of them take their vision to fruition.
There's been a recurring conversation in the industry about data scarcity - the idea that enterprises, particularly in India, don't have enough data for AI. Where do you think that conversation stands today?
That's a large opportunity, and it plays to Oracle's strength. We have 50 years of data management experience. The issue was never that enterprises lack data - it's the ability to get access to the data they already have.
Take a manufacturing shop floor. You might have IoT (Internet of Things) sensors from one manufacturer, cameras monitoring an assembly line from a different manufacturer, and a homegrown application sitting alongside both. Each system holds data, and each is separate. If I told a customer they had to replace all of that with Oracle systems, that wouldn't happen. But if I say: leave your IoT system running, we'll simply read the data flowing through it - we access it without asking you to move anything, without disrupting your data flow - that's a different proposition.
Too many customers are currently building what I call scaffolding: constructing ladders outside the building just to try to reach their own data, without actually solving the underlying problem. They're spending time and resources collecting data in ways that add no direct business value. We're saying don't collect the data - leave it where it is, we'll reach into it, and we'll build systems on top of it.
AI has also shortened the time it takes to show customers real value. You don't have to commit to a six-month engagement before seeing results. We can spend two to three weeks with a customer - because we're not touching their systems, only reading from them - and show them what's possible without disrupting the business.
The key point on data is this: the large language models that power AI today are trained on public information. To make AI useful in an enterprise, you need to marry that public information with your internal information. That's the gap Oracle helps customers close.
Multi-cloud promised run-anywhere deployment with unified, governed data movement. Which part of that promise has been structurally difficult to deliver, even when hyperscaler partnerships are in place?
People are often a product of where they start from. Oracle began its cloud journey in the multi-cloud era. When we launched OCI (Oracle Cloud Infrastructure) - our first region was in 2016, and we scaled meaningfully from 2019 onwards - every customer I approached was already with another cloud provider, sometimes two. We were entering as the second, third, or fourth cloud provider. That meant multi-cloud coexistence was in our DNA from the beginning.
We leapfrogged by shrinking our cloud footprint. When a large hyperscaler announces a new cloud region, there's typically a lag of two to three years from announcement to going live, because the infrastructure involved is enormous. We designed OCI differently - small, agile, and quick to deploy - starting the build in 2014. We wanted to deliver a cloud that could start as small as a customer needs it, then grow as large as required.
That decision produced three distinct benefits. First, it lets us expand coverage quickly. We now have well over 80 public cloud regions, and we're adding more frequently - we opened another region in Southeast Asia just last week. Second, we were able to take that same cloud and put it inside a customer's own environment as a dedicated region - same services as the public cloud, but where only that customer holds the keys. My competitors scale down their large cloud infrastructure to two or three services for a dedicated deployment. We offer full service parity. Third, because our cloud was already small, we were able to embed it natively inside Amazon, Azure, and Google. None of the other hyperscalers can do that.
Where do you draw the line commercially when it comes to hyperscaler partnerships - and how do customers avoid ending up in a shared accountability problem?
The customer draws the line, not us. If a customer has an existing commitment with a hyperscaler - any of the three major ones - they don't even need a separate agreement with Oracle. We'll draw down from whatever contracts and agreements they already have. I operate as a native service within that hyperscaler; as far as the customer is concerned, they're dealing with the hyperscaler, not us. We give our SLAs to the hyperscaler, and the hyperscaler passes them on to the customer.
Alternatively, a customer can choose to have a direct relationship with Oracle and ask us to deploy across our cloud and others. In that case, we take on the commercial relationship and the SLA, and we run the workload across both environments. The customer might also say: I have a three-year commitment with a particular hyperscaler and I'm not going to walk away from it. We won't ask them to. We'll fit into that.
The same flexibility applies at the application and database layer. A customer can split the application tier and the database tier across different environments, and we'll work with any combination. The choice is entirely the customer's.
Looking at the next 12 to 18 months, where do you see enterprise AI heading - and what does mainstream adoption actually look like in practice?
I think the statement "AI is going to change everything" is going to come to life visibly. You're going to see many more customers move from pilots to production. Whether that's e-commerce, retail, manufacturing, banking, fintech, stock investments, insurance claims, or healthcare, AI agents are going to be embedded into those value chains.
Loan approvals are a good example. The loan approval process is essentially about gathering data from multiple sources, processing it against defined criteria, and routing cases that exceed certain thresholds to a human. A large portion of that can be handled by agents. The process becomes simpler for the consumer; approvals come faster.
The real measure of success will be when you don't know it's an agent. When a company is confident enough in its agents - confident that they're making decisions based on access to all relevant data, with appropriate human checkpoints in the workflow - consumers won't experience it as technology at all. They'll simply notice that a company is very good at processing loan applications or insurance claims. AI is going to be embedded in nearly everything.
Published by HT Digital Content Services with permission from TechCircle.