AI costs are dropping fast—and India stands to gain: OpenAI CEO Sam Altman

NEW DELHI: The cost of deploying and running artificial intelligence (AI) applications is falling at an exponential pace, potentially making AI tools affordable at scale across India sooner than expected, according to Sam Altman, co-founder and chief executive of OpenAI, one of the world’s most valuable startups.

Altman also dismissed the idea that companies like his would “kill” software and technology services firms, even as AI automates large parts of their core work.

“I am optimistic that we are going to drive prices down more than anyone thinks is possible or reasonable or likely. Using AI can become absolutely cheaper for India to use at scale, and we are moving actively in this direction,” Altman said, speaking at a roundtable that Mint was part of.

At the heart of his optimism are sharply declining computing costs and rising investment in artificial intelligence infrastructure in India.

“The cost of inference from two Decembers ago, in about 14 months, has declined by 1,000x. Will it continue to decline so exponentially? I don’t think we can, but I’m sure that we can continue to reduce it and make it absolutely cheaper and more affordable,” he said.

This was Altman’s first visit to India since last year, ahead of which he had said the country was emerging as one of OpenAI’s biggest markets. “India is OpenAI’s largest market outside of the US, and it is rapidly on course to become the biggest market for us globally,” he had told Mint in August last year.

Altman, however, argued that differentiation will increasingly come from building “good software”, which will continue to require skilled engineers.

Analysts broadly agree.

”There will be some near-term disruption as various tasks get increasing degrees of automation, but to be sure, all AI companies will eventually require a layer of tech services before they are deployed in enterprises. This can increase the market for AI firms like OpenAI, and also amplify business for the IT sector,” said Sudarshan Seshadri, partner at consulting firm Tiger Analytics, which itself is a user of OpenAI’s tools.

Speaking about the current state of adoption, Chris Lehane, chief global affairs officer at OpenAI, said India has 100 million weekly active users, “of which 30 million are students.”

For Altman, much lies in how OpenAI fares in competition against its closest rivals—Google’s Gemini and Anthropic’s Claude AI models. While the latter struck a partnership with Infosys this Tuesday, OpenAI fired its own salvo with a partnership with the Tata Group on Thursday.

”A part of this deal is to help one of the biggest companies in the world use and get access to the best of our AI tools. But of course, there is the data centre part that the Tatas are building, and we plan to accelerate the same and use it for our services,” Lehane said.

Neither Altman nor Lehane specified if the Tata data centre plans would support OpenAI’s data localization efforts.

In the longer term, Altman said India’s AI trajectory will depend on policy and infrastructure choices, particularly around power and water consumption, as AI inference remains resource intensive.

Despite that, he argued efficiency gains are already reshaping the economics of adoption.

”It (AI models) will absolutely be cheaper. Energy costs at inference today per line of code, relative to the energy costs of a person running the line of code is already way, way lower. Today’s AI models are already quite efficient in terms of cost per token of their performance, versus the cost of one performance token of human performance. I expect this to continue to get stronger, and the cost of per unit of intellectual capacity will go down, and we’ll use so much of it,” he said.