Blitz Bureau
NEW DELHI: IBM CEO Arvind Krishna pushed back against claims that India lags in technology innovation, defending the country’s role in enterprise AI during a high-stakes panel at Davos with Union Minister for Electronics and IT Ashwini Vaishnaw.
Speaking with NDTV CEO and Editor-In-Chief Rahul Kanwal, Krishna dismissed criticism that Indian firms lack fundamental, breakthrough research. “That criticism is not fair,” he said. “The majority of work in India isn’t about creating new applications from scratch, it’s about maintaining and evolving them.” That, he said, does not involve deep enterprise and AI innovation.
Krishna mapped how tech revolutions shift value from infrastructure like semiconductors to operating systems, and eventually to applications. “In less than ten years, 90 per cent of value moves to innovative apps,” he said. “If India taps into that transition, it can unlock a boom in innovation and capital.”
On India’s approach to Large Language Models (LLMs), Krishna argued that trillion-parameter models will become commoditised. “They’ll have value but no switching cost, and price will drop to bare minimums,” he noted. Instead, he advocated building smaller, effective models infused with sovereign data.
Minister Vaishnaw reinforced this, revealing that sovereign AI models had already been deployed in a recent major defence scenario. “The results were phenomenal,” he said. “We must build a bouquet of smaller, sector-specific models. That’s how we start seeing economic gains.”
Krishna emphasised that innovation isn’t about size. “All 10 of the largest LLMs today perform within a narrow band,” he said. “The next leap won’t come from scale.” Vaishnaw pointed to Indian startups leading unique approaches, including one that has bypassed text entirely to create sound-to-sound AI models.
On Artificial General Intelligence (AGI), Krishna split from Silicon Valley orthodoxy. “Belief in AGI is like religion,” he said. “Some think more training and scale will get us there. I disagree. We don’t yet know how to truly infuse knowledge into models. That’s the scientific gap.”
He concluded that building efficient, specialised models like 50-billion parameter LLMs is the pragmatic path. “They’re 100 times cheaper to run. That opens up massive possibilities for innovation.”
