Synopsis
Snowflake CEO Sridhar Ramaswamy believes India's AI advantage lies in its talent and innovation under constraints, not just infrastructure. He advises focusing on engineering prowess and efficient systems, cautioning against solely measuring AI adoption by token usage and emphasizing business outcomes over consumption metrics.Listen to this article in summarized format
Speaking during a media briefing at Snowflake Summit 2026 in San Francisco, Ramaswamy said countries such as India are unlikely to outcompete larger economies purely on the basis of power generation and compute capacity as AI models become increasingly resource-intensive.
“It is hard for a nation that large and that tightly constrained with respect to things like power to outcompete on the basis of just raw power generation,” Ramaswamy said while answering ET’s question.
Instead, India should focus on areas where it already has a natural advantage, including engineering talent, open-source innovation and building efficient systems under constraints, he said.
India is seeing a surge in investments aimed at building AI and data centre infrastructure, with major technology companies including Microsoft, Google and Amazon committing billions of dollars to expand their cloud and data centre footprint in the country. At the same time, concerns are growing globally about the energy and computing resources required to support increasingly powerful AI models.
While acknowledging those challenges, Ramaswamy argued that constraints often drive innovation rather than limit it.
“Operating within constraints can be as liberating as not having constraints, because you just end up thinking differently,” he said.
The debate around AI spending has intensified in the past weeks as companies grapple with rising token costs and growing AI bills. Some firms, including Uber and Microsoft, have reportedly tightened controls around AI usage after costs increased faster than expected.
However, Ramaswamy also pushed back against the growing obsession with measuring AI adoption through token usage. “First of all, I think token maxxing is a terrible idea,” he said, referring to the practice of encouraging employees to maximise their use of AI tools.
According to him, companies should focus on business outcomes rather than consumption metrics.
“The presence of good AI usage numbers does not indicate that you’re being productive with AI, but a complete absence of numbers certainly indicates you have no clue,” he said.
The data and AI company is already seeing AI agents significantly reduce the time required for some tasks, with projects that once took months being completed in hours. However, he cautioned that higher AI usage alone should not be mistaken for productivity.
The company is also working to lower AI costs by using smaller models for routine tasks while reserving more advanced models for complex reasoning and planning.
Snowflake also unveiled new capabilities for CoCo, its AI coding agent formerly known as Cortex Code, and launched Datastream, a managed streaming service designed to bring real-time data into AI applications.
“Anyone that thinks the trend line is fake simply isn’t experiencing AI at the depth that it needs to be experienced,” he said.
For enterprises, the real question is no longer whether to adopt AI, but how to use it effectively and efficiently, Ramaswamy added.
The reporter was in San Francisco at the invitation of Snowflake.
