loading
loading
For Indian enterprises adopting AI, choosing the technology is only part of the challenge. Deciding what to own, how much to spend, and where AI can deliver measurable business value is becoming just as important. A December 2025 government backgrounder, citing Nasscom’s AI Adoption Index, said 87% of enterprises in India were actively using AI solutions. As adoption expands, technology leaders face a more complex choice: buy ready-made solutions, build capabilities in-house, or access models from external providers. Frontier models have made it easier to add AI capabilities to products and workflows. But the decisions around cost, speed, data, and business outcomes are increasingly intertwined. For enterprises and startups, the challenge is to identify which parts of the AI stack can create a lasting advantage and which to source externally as the technology evolves. These questions shaped the second roundtable held on the sidelines of Inc42’s inaugural CTO Summit 2026, where technology leaders discussed how they are building AI-led systems and organisations. Moderated by Abhinav Johri, partner, technology consulting, EY India, the conversation brought together: Govind Soni, cofounder, CoinSwitch & Lemonn Harsh Sahu, cofounder and CTO, Matters.ai Kiran Reddy Singam, VP, product and tech, BigHaat Shanti Kuropati, founder, SeedlingLabs Shiv Buddharaju, VP, engineering, Oxmiq Labs Vikas Roy, director, engineering, Vahan.ai Vinay Rai, executive VP, technology, Netradyne Vivek Manoharan, CEO and CTO, Gravitas Building An Edge With Context And Workflows The conversation began with a question: as AI moves from experimentation into core business processes, does every problem require a new technology layer? Matters.ai’s Sahu argued that the starting point should be a deep understanding of the domain. Knowing where existing solutions fall short can help technology leaders build differentiated products. “I feel that founders and tech leaders need to know their specific domain really well to not only see the existing gaps but also use AI to fill in those gaps in a way that was never done before,” said Sahu. This emphasis on domain knowledge led to a distinction between the intelligence a company can source and the business context it develops itself. SeedlingLabs’ Kuropati argued that while frontier models change rapidly, proprietary data, context, and workflows can offer a more durable advantage. “If it is around data or context, you have to own it, irrespective of which frontier model you are working with, because that eventually becomes the moat for any AI-native founder, while the rest of the technology solutions remain largely the same,” said Kuropati. For CoinSwitch & Lemonn’s Soni, operational context matters most when accuracy directly affects customers’ money. “When you’re dealing with customer money, all operations become very critical, and accuracy can’t be compromised. So whenever you’re building something, that has to be built on top of that 10-year-old context, and not just with some new technology in isolation,” said Soni. What To Own, What To Rent Retaining control of context does not necessarily mean owning every technology layer around it. The discussion moved beyond the traditional build-versus-buy framework to examine what companies need to own, what they can rent, and how those choices change over time. Oxmiq Labs’ Buddharaju said these decisions span multiple dimensions, with one choice often affecting another. In AI, the workload involved in developing a product can also differ substantially from what it takes to serve customers at scale. “Technology choices are not simply about buy versus build. The real question is what you want to own, what you want to rent, and for how long. In AI, that distinction becomes even more important because the workload for building a product can be very different from the workload for serving customers,” said Buddharaju. Netradyne’s Rai made the case for renting intelligence. With frontier models evolving rapidly, he argued that building and maintaining an in-house model could leave companies playing catch-up. “Unless regulation requires us to run our own models, we are better off renting intelligence right now. The leaders keep changing, and an open-source model that is ahead today could leave you behind within weeks. For most companies, speed and flexibility matter more than owning the model itself,” said Rai. Making AI Economics A Product Decision Alongside ownership and flexibility, participants examined another test for AI deployment: its impact on business economics. BigHaat’s Singam said technology decisions must be tied directly to business metrics, with their contribution to margins assessed before deployment scales. “Whenever we deploy anything, we look at which metric it is moving and whether that change is positive or negative for our contribution margin. Whether it is fulfilment, RTOs, users or technology costs, we need to understand the economics before we scale,” said Singam. For Gravitas’ Manoharan, this assessment begins at the product development stage. A model’s ability to perform a task is only one consideration. The cost of running it also determines which features are viable. “Cost economics really matters to us, and it defines the product features we build. When we develop AI, we have to ask whether the economics will be sustainable within one or two years,” said Manoharan. He cited the example of converting US healthcare clinical charts into ICD and CPT codes. While AI models could produce useful outputs, the cost of processing each chart had to be weighed against the existing BPO-led process. This comparison makes unit economics an early filter for deciding which AI features to pursue. At Vahan.ai, Roy said AI expenditure is similarly linked to measurable productivity gains. The objective is to automate processes and increase the number of placements each telecaller could handle. “Clients care about outcomes, while we care about efficiency. For us, AI meant automating processes and asking which human decisions AI could handle. The real test was whether that could help us do more placements per telecaller,” said Roy. Across the roundtable, a common approach emerged: retain control of the context that differentiates the business, stay flexible on the models powering it, and assess the economics before scaling. For technology leaders navigating the next phase of AI adoption, these decisions will shape both the systems they build and the value they deliver.
document fetched 2026-10-14 03:05 · canonical https://inc42.com/buzz/build-buy-or-rent-how-indian-ctos-are-rethinking-the-ai-stack