Building consumer AI is hard because of free products, says Sarvam cofounder

Building artificial intelligence for consumers is hard, because the competition is with free products like ChatGPT and Gemini, Sarvam AI cofounder Vivek Raghavan said.

Unless Indian large language models (LLMs) create real value, adoption will remain limited, Raghavan said.

Last week, Sarvam AI launched multiple applications of its LLM at the India AI Impact Summit in New Delhi.


Raghavan compared sovereign AI capability to countries that have nuclear power and those that don’t. He argued that India must choose to build its own foundation models.

“If you pick your battles carefully and focus on problems that are deeply relevant to India and you solve them better than anyone else, things will move," he said, speaking at ET’s AI Conclave & Awards 2025 here on Thursday.

On Sarvam's monetisation plan, Raghavan said: “We are primarily working with enterprises. Our approach is largely B2B. We’re also working with governments and different sectors. These are the primary channels we are focused on right now.”

Sarvam-30B and Sarvam-105B, its first homegrown foundational models unveiled last week, support 22 Indian languages. It also debuted Sarvam Kaze AI-powered smart glasses and a suite of tools, including Sarvam Vision (OCR/multimodal), Sarvam Dub and the Indus beta app for mobile and web.

At the AI summit, the startup announced partnerships with Qualcomm, HMD, Bosch and Nokia to integrate its models.

India’s AI opportunity lies in a well-orchestrated system of small models which are cheaper, more energy-efficient and faster, Raghavan said. “We need to focus on solving real business problems that matter to India, and show that we can solve them better and more effectively.”

While frontier models are good, “the reality is that for 95% of what people actually need to do, smaller models are more than sufficient”, he said.

“It’s very clear that the US and China will be frontier AI powers. There are a few countries in the middle and India is one of them,” Raghavan said.

Raghavan, who was the chief product manager for the Aadhar programme, said the ability to generate intelligence ‘tokens’ at low cost and at massive scale is a weak sovereign capability in India today.

“What ends up happening instead is that we send our data out and import intelligence back in,” he said. “Some people are proud that India is the largest user base for ChatGPT, but what are we really doing?”

India needs people who can train models and run models at scale, he said. “Not everyone needs to do this, but enough people must do it. Otherwise, we risk becoming a digital colony.”