'AI divide may be much worse than the digital divide, if we are not careful,' says Sarvam AI's cofounder Vivek Raghavan

bengaluru_techsummit

India must build its own foundational AI systems to keep pace with global advances and ensure equitable access, industry leaders said at the Foundational Models for India panel at the Bengaluru Tech Summit 2025.

"The existing models have sub-1% Indian data," said Vivek Raghavan, cofounder of Sarvam AI, which is developing large-scale Indian LLMs.

Sarvam plans to release a 120-billion-parameter model "with more than 17 trillion tokens, of which 15–20% come from fully Indian data," he noted, calling it a critical step in understanding and building AI “from scratch.”

Raghavan warned that the gains from AI must reach all citizens.

If not, "the AI divide, if we are not careful, will be much worse" than the existing digital divide, he said. While India has created scalable public digital infrastructure such as Aadhaar and UPI, he argued that most systems remain far from user-friendly. "We can have significant efficiencies in what we do, and it should go to everyone," he said.

Sarvam was earlier announced as the first startup selected to build India’s foundational AI model under the mission.

Also Read: Sarvam likely to roll out India’s first LLM by early next year, 40,000 GPUs to be empanelled by Centre

Founded in July 2023 by Vivek Raghavan and Pratyush Kumar, who previously worked at Infosys co-founder Nandan Nilekani-backed AI4Bharat, Sarvam develops a full-stack offering for Generative AI, ranging from research-led innovations in training custom AI models to an enterprise-grade platform for authoring and deployment.

Raghavan, an entrepreneur and technologist who was instrumental in building Digital Public Goods (DPGs) like Aadhaar, said that Sarvam will work with Indian enterprises to co-develop domain-specific AI models leveraging their data.

On funding, Raghavan said deeptech founders need patient capital because “building foundational models takes time and money,” adding that India must avoid “bandaid revenue thinking” and instead back long-term innovation.

Evaluating AI systems remains another challenge. Ananth Nagaraj, cofounder of voice-AI startup Gnani.ai, said enterprises increasingly rely on field trials to assess accuracy and end-user impact.

Raghavan added that global benchmarks often fail Indian realities and argued for “much more domain-specific and India-specific benchmarks.”

Sashikumar Ganesan, founder of deeptech firm Zenteiq and professor at IISc, said benchmarking must be continuous, adding that safety is a shared responsibility of developers, users and policymakers.