AI is core to national resilience, but India’s adoption gap remains

AI is fast emerging as a foundational layer for national resilience – enabling govts and enterprises to anticipate disruptions, respond faster, and operate at scale. But while India has built some of the world’s strongest digital public infrastructure (DPI), panelists at a discussion we had last week said the country’s adoption of AI remains uneven, with govt often ahead of industry and execution lagging ambition.

Arundhati Bhattacharya, president & CEO of Salesforce - South Asia, framed the paradox sharply. “The Indian govt seems to be ahead of enterprises, which normally is not the case. I have the good fortune of looking at various countries in South and Southeast Asia. In a country like Singapore, for instance, at the enterprise level, usage of AI is far more than in India,” she said.

While cloud computing is essential for AI scale, given AI’s compute requirements, Bhattacharya noted continued hesitation in parts of the public sector. “The public sector is still not completely confident that cloud is secure enough,” she said. Without cloud, the cost of deploying AI at scale becomes prohibitive. And given the pace at which AI is evolving, catching up later would become difficult.

Dwarka Srinath, group chief digital and information officer at Tata Power, also noted the govt’s proactiveness – most recently on agentic AI running on DPI. He stressed the need to focus on foundations – strong enterprise architecture, data pipelines, and infrastructure – for AI to “do wonders”. This is particularly evident in sectors like energy, where resilience is tied to physical systems. AI-driven data centres and digital infrastructure require massive and fluctuating power. The demand is “very very oscillating,” Srinath said, especially during model training.

Managing this requires sophisticated forecasting, energy portfolio optimisation (among solar, wind, pumped hydro, batteries and thermal power), and sustainability metrics. “We need to build in measurable carbon and water usage metrics,” he said, warning that “water is going to be a bigger problem with cooling.” Tata Power, he said, is working on all of these areas.

Need to be boldSreyssha George, managing director & partner at BCG, said at the enterprise level the biggest challenge is not technology but execution. Most companies, she said, have done the experimentation, but not a lot of them have then moved on to say, “what are the 10-15 things that we’ll go deep in.”

“I think we need to boldly move forward. We’ve seen how this technology gives RoI (return on investment). If you take a simple software development lifecycle, there’s no more debate on whether there’s RoI or not,” she said.

The speed of AI change, she said, is also creating pressure on talent. “The number of open roles in the market, we just don’t have the talent,” she said, highlighting the need for large-scale upskilling. And this training, she said, has to happen really fast, not like in previous tech waves.

George said India can do what’s needed, given how its technology services business has always understood business processes, understood the complexity of organisations and solved for them. “In places where there’s deep data and complex processes, AI will show you a disproportionate multiplier effect,” she said.

Trust is keyMankiran Chowhan, managing director at Salesforce India, pointed to another foundational requirement: trust. She emphasised that enterprise AI adoption relies on embedding security and trust guardrails directly into the architectural design rather than treating them as an afterthought. This is especially important in regulated sectors like banking and insurance, where data sensitivity is high. Salesforce’s products, she said, ensures that security remains consistent while providing the flexibility to integrate third-party AI models securely.

Chowhan also highlighted interoperability as a key challenge. If one agent can’t talk to another, outcomes become a challenge, she said, pointing to the need for integrated platforms rather than siloed AI deployments.

Bhattacharya noted the importance of building good local language models for agentic AI to provide in-context answers, which in turn is necessary to build trust. “The agents we are talking about – the voice-to-text or voice-to-voice – work on databases, and these databases have to be there in that particular language for the agents to mature in using that language in the proper context,” she said.

At a broader level, Chowhan argued, the real bottleneck is not technology but imagination. “What’s happening today is more imagination deficit, not a technology deficit,” she said. Enterprises are spending too much time choosing tools and too little rethinking how their businesses could operate in an AIfirst world.