India could start exporting artificial intelligence (AI) agents to the world in the next five-seven years, while providing the lowest-cost access to AI to every Indian if it reduces the cost of inference, said Rajeev Saluja, vice-president, 5G Radio at Reliance Jio.
“I think as Indians, the biggest challenge for all of us is how do we reduce the cost of inference,” Saluja said. He was speaking at a panel discussion at the pre-event business day of India Mobile Congress on Wednesday. Reliance Jio, which is likely to go public in potentially one of the largest IPOs in India, in the coming weeks, is investing heavily into AI and has pegged AI as one of its future growth pillars.
Saluja said the company views AI from a twin-pronged perspective: As an enterprise that could produce and leverage AI, and from a consumer perspective, to industrialise it and lower costs for every Indian.
“For the past 25 years, India has exported software and services to the whole world,” he said, while talking about the need for India to transition towards becoming an AI exporter.
“Today, when we talk of AI infra, we talk of global graphics processing units (GPUs) and we talk of watts. The transition that we are trying to do from that mindset is to look at inference per rupee or inference per watt and, third most important, inference per time,” Saluja added.
Inference cost is the cost of running data through a trained AI model to get an output.
Saluja said that Jio is working to optimise inference costs across these three parameters — per rupee, per watt and per time — to reduce the cost of running AI and build more economical models in India.
Return on investment, Saluja said, is a key challenge to AI adoption in India. And, overcoming it could mean that India becomes an exporter as well as a beneficiary of the technology.
Further, Jio is seeing more demand for AI from rural areas and is looking to expand its reach in these regions, he said.