“If a person can talk to a computer and get instructions, information, or an agent to work on their behalf, that’s the final concept — especially in a country with so many languages and dialects,” Nilekani said during a fireside chat with Vishal Dhupur, managing director for Nvidia in South Asia. India has a population of 1.4 billion, with hundreds of millions still using feature phones and more than 100 languages in active use.
Nilekani said voice AI is key to digital equality. “Just as the Unified Payments Interface (UPI) made digital payments effortless for everyone, voice-driven interfaces can remove barriers to opportunities in sectors such as agriculture and education. It’s the final frontier,” he added.
He emphasised the need to build population-scale voice AI applications using frugal design principles. “India will be the source of applying AI and frugal engineering to solve real-world problems,” Nilekani said. “That will have global implications because many countries cannot afford expensive systems. If we can deliver population-scale voice AI in 22 Indian languages, it becomes a global requirement. Voice AI is the final frontier,” he added.
According to Nilekani, accuracy will no longer be a barrier; he said. He emphasised the need to build population-scale voice AI applications using frugal design principles. “India will be the source of applying AI and frugal engineering to solve real-world problems,” he said.
Dhupur said the focus of voice applications should be on making them seamless, trustworthy and cost-effective. “For all tasks, costs have to come down. All of us know the pipeline cost of doing a transaction. The opportunity for India is to play the large-volume game and bring those costs down,” he said.
AI, Nilekani said, is unfolding as a race in two directions — towards the bottom and towards the top. “The race to the bottom is AI so low cost, and present in people’s mental health. The race to the top is improving lives at scale. That’s where India can lead with foundational models,” he said.
The risks of AI — hallucinations, biased outputs, morphed images, obscene or sexually explicit content — are already well known. Nilekani cited content generated by Elon Musk’s Grok as an example. India, he said, offers a chance to innovate at scale, which is crucial for mass adoption of any technology. He pointed to Aadhaar, India’s unique identification system, and UPI, the world’s largest real-time payments system operated by the National Payments Corporation of India.
In Aadhaar’s case, the real innovation was in minimal design, with just four identification fields. That simplicity enabled the system to scale, enrolling around 1.5 million people a day across millions of stations.
Scaling, Nilekani said, can be achieved using large language models. He reiterated that large language models will become commodities, with companies such as OpenAI, Meta, Google, Anthropic, Mistral AI, and DeepSeek building them across the US and China.
 "Nandan Nilekani, chairman, Infosys & cofounder, EkStep Foundation")