/images/ppid_59c68470-image-178903502976412940.webp)
/images/ppid_59c68470-image-178885761451747054.webp)
What is the story about?
India can emerge as a global leader in artificial intelligence over the next decade, with the country now building the data, computing capacity, talent and investment needed to drive adoption as well as innovation, Sarvam AI co-founder Pratyush Kumar said.
Speaking to CNBC-TV18 on the sidelines of Day 2 of the Global Fintech Festival 2026, Kumar said India is at an early stage of its AI journey, drawing a parallel with the country's rise in digital payments.
“I think that same thing is possible in AI,” he said, pointing to the AI summit hosted in India, the rollout of the country's first models, growing GPU compute capacity and rising investments.
“I believe that this is the first year we start thinking about population-scale AI,” Kumar said. “So I am very bullish that the next ten years, India will absolutely decimate the numbers in terms of usage and adoption of AI.”
However, he said adoption alone will not be enough for India to establish leadership in AI. Unlike fintech, AI requires continued investment in innovation and research, given the need for new advances in foundational technology.
Kumar said India needs to develop capabilities across the AI value chain, including GPUs and data centres, software infrastructure, models and applications.
Sarvam announced the general availability of a “token factory” in India, which allows banks, regulated industries, government bodies and startups to use models hosted on Sarvam's GPUs with full data protection, he said.
Kumar also rejected the idea that India should choose between smaller models focused on adoption and larger frontier models.
While smaller models can handle a large share of tasks, he said larger models are required for frontier applications such as software engineering and cybersecurity.
“We should not let go of R&D and pushing the frontier as we think about adoption. Both need to go hand in hand,” Kumar said.
He added that India should ensure the AI value chain remains within the country, including infrastructure, software, models and applications.
Kumar said Sarvam's focus on sovereign AI does not mean decoupling from global technology ecosystems.
“Sovereignty is not a tax. It’s a specialisation,” he said.
Sarvam is hosting Chinese models on its stack, working with American hyperscalers, hiring talent outside India and working with international enterprises, Kumar said. Through its partnership with HCLTech, the company is also working with international clients.
“There’s a real opportunity for us to build, do the research, have the tech in India, and then build for the world,” he said.
Kumar said Sarvam's existing 105-billion-parameter model competes favourably with models of a similar size while being cheaper. The company is also developing products in areas including agentic AI, cybersecurity and coding, with the aim of offering models and applications as a complete stack.
He said global models such as ChatGPT and Claude remain benchmarks for Sarvam, but the company views the competition as a long-term effort.
“This is not a race to be looked at in quarters. It is a multi-decade thing,” Kumar said.
Kumar acknowledged that AI-driven productivity gains could reduce the number of people required for certain functions, but said the technology could also enable India to build more across sectors such as healthcare, education, agriculture, governance and banking.
“What it needs is a realignment of how we think about skills. People have to take agency to upskill themselves, no matter which stage of the career they are in,” he said.
On enterprise adoption, Kumar said “the hockey stick has started”, with companies increasingly using AI to make repetitive processes more efficient, cheaper and faster.
He said adoption is visible across employee productivity, sales and field forces, background processes and customer support, while full end-to-end autonomy remains limited because governance requirements are strong and liability frameworks are still evolving.
Kumar also said company boards have moved beyond simply listing AI use cases and are increasingly thinking about what the underlying AI stack means for their businesses over the next five years.
Kumar expects AI adoption to expand further as costs fall, eventually opening up a much wider range of consumer applications.
He said the technology is currently being consumed largely through enterprise use cases, but falling token costs could make consumer applications viable at scale.
He expects this to include applications such as AI tutors, AI doctors and AI financial advisers, with the “democratisation of AI” putting the technology in the hands of more Indians over the next three to four years.
Speaking to CNBC-TV18 on the sidelines of Day 2 of the Global Fintech Festival 2026, Kumar said India is at an early stage of its AI journey, drawing a parallel with the country's rise in digital payments.
“I think that same thing is possible in AI,” he said, pointing to the AI summit hosted in India, the rollout of the country's first models, growing GPU compute capacity and rising investments.
“I believe that this is the first year we start thinking about population-scale AI,” Kumar said. “So I am very bullish that the next ten years, India will absolutely decimate the numbers in terms of usage and adoption of AI.”
However, he said adoption alone will not be enough for India to establish leadership in AI. Unlike fintech, AI requires continued investment in innovation and research, given the need for new advances in foundational technology.
India needs to build the full AI stack
Kumar said India needs to develop capabilities across the AI value chain, including GPUs and data centres, software infrastructure, models and applications.
Sarvam announced the general availability of a “token factory” in India, which allows banks, regulated industries, government bodies and startups to use models hosted on Sarvam's GPUs with full data protection, he said.
Kumar also rejected the idea that India should choose between smaller models focused on adoption and larger frontier models.
While smaller models can handle a large share of tasks, he said larger models are required for frontier applications such as software engineering and cybersecurity.
“We should not let go of R&D and pushing the frontier as we think about adoption. Both need to go hand in hand,” Kumar said.
He added that India should ensure the AI value chain remains within the country, including infrastructure, software, models and applications.
Sarvam wants to build in India for the world
Kumar said Sarvam's focus on sovereign AI does not mean decoupling from global technology ecosystems.
“Sovereignty is not a tax. It’s a specialisation,” he said.
Sarvam is hosting Chinese models on its stack, working with American hyperscalers, hiring talent outside India and working with international enterprises, Kumar said. Through its partnership with HCLTech, the company is also working with international clients.
“There’s a real opportunity for us to build, do the research, have the tech in India, and then build for the world,” he said.
Kumar said Sarvam's existing 105-billion-parameter model competes favourably with models of a similar size while being cheaper. The company is also developing products in areas including agentic AI, cybersecurity and coding, with the aim of offering models and applications as a complete stack.
He said global models such as ChatGPT and Claude remain benchmarks for Sarvam, but the company views the competition as a long-term effort.
“This is not a race to be looked at in quarters. It is a multi-decade thing,” Kumar said.
AI and jobs
Kumar acknowledged that AI-driven productivity gains could reduce the number of people required for certain functions, but said the technology could also enable India to build more across sectors such as healthcare, education, agriculture, governance and banking.
“What it needs is a realignment of how we think about skills. People have to take agency to upskill themselves, no matter which stage of the career they are in,” he said.
On enterprise adoption, Kumar said “the hockey stick has started”, with companies increasingly using AI to make repetitive processes more efficient, cheaper and faster.
He said adoption is visible across employee productivity, sales and field forces, background processes and customer support, while full end-to-end autonomy remains limited because governance requirements are strong and liability frameworks are still evolving.
Kumar also said company boards have moved beyond simply listing AI use cases and are increasingly thinking about what the underlying AI stack means for their businesses over the next five years.
Consumer AI could be the next phase
Kumar expects AI adoption to expand further as costs fall, eventually opening up a much wider range of consumer applications.
He said the technology is currently being consumed largely through enterprise use cases, but falling token costs could make consumer applications viable at scale.
He expects this to include applications such as AI tutors, AI doctors and AI financial advisers, with the “democratisation of AI” putting the technology in the hands of more Indians over the next three to four years.
/images/ppid_59c68470-image-178879003160795863.webp)
/images/ppid_59c68470-image-178893752656920849.webp)
/images/ppid_59c68470-image-178894009299893925.webp)
/images/ppid_59c68470-image-178894016129596581.webp)
/images/ppid_59c68470-image-178886508102642669.webp)
/images/ppid_59c68470-image-178877256413242742.webp)