Infosys co-founder and chairman Nandan Nilekani on Friday said that large language models (LLMs) have become commodities, and with the cost of building them dropping significantly, it is the right time for Indian companies to invest in developing their own AI models.
“What DeepSeek has done is show that you don’t need billions of dollars to create a reasonably good large language model. That’s a big breakthrough,” Nilekani said at an event by the All India Management Association (AIMA).
Nilekani’s remarks mark a shift from his earlier stance, where he suggested India should focus on AI applications rather than building its own foundational models. Now, with costs falling—potentially as low as $50 million for an LLM—he sees an opportunity for Indian firms to step in.
“The technology is moving so quickly that it’s dropping in price and becoming more efficient. Would I spend a billion dollars to build an LLM? No. But if I can do it for $50 million, sure,” he said.
Beyond affordability, Nilekani highlighted the need for India-specific LLMs to address biases in global AI models.
“…A very good argument for building India’s own LLMs is that we can train them on Indian datasets, Indian context, ensuring accuracy,” he noted, adding that the next year will likely see significant AI models emerge from India.
In line with this, the government has announced plans to develop foundational AI models over the next 8-10 months. The initiative, backed by the Rs 10,000 crore IndiaAI mission, aims to build AI systems rooted in Indian languages, culture, and datasets. To support this, the government is setting up a computing infrastructure with 18,693 GPUs, with startups, academia, and researchers being able to access it at subsidised rates.
Nilekani also addressed concerns about AI’s impact on employment, stating that while certain tasks will be automated, very few jobs will be entirely eliminated. Instead, AI will enhance productivity and create new job opportunities that don’t yet exist.