GFF 2026: India’s AI advantage could be cost, not just capability, says Sarvam AI CEO

India’s biggest advantage in artificial intelligence may not be its ability to build the most powerful models, but its ability to deploy AI at a cost that works for a billion people, Sarvam AI co-founder and CEO Pratyush Kumar said at the Global Fintech Fest 2026.

Kumar said India was still at the beginning of its journey in building AI, even as the technology itself has entered its fourth year since OpenAI launched its chatbot ChatGPT. The country now has its own AI models, growing access to GPUs and an expanding pool of AI talent, he said.

Also Read: Voice AI startup Navana raises Rs 40 crore from Ronnie Screwvala, others


“I believe the next 10 years, India will absolutely lead in AI, both adoption and building of AI,” Kumar said.

But the opportunity for India, he said, goes beyond simply developing indigenous models. The country needs to build the infrastructure to produce and run AI at scale and at a cost structure that can support its massive population.

“India will not only do it at scale, but also at the cost structure that works for a billion people. That’s the opportunity India offers to the global effort in AI,” Kumar said.

Sarvam, which is building its own large language models as well as deploying open models, is said to already seeing the economics of that approach play out.

Kumar said Sarvam is running models on GPUs in India at prices that are “equal or cheaper” than those available globally. The company is also serving models built by other providers on its infrastructure, he said.

Sarvam’s experience, he said, points to the emergence of what he called “token factories” — infrastructure that can serve AI models at scale, much like factories produce physical goods.

“The first thing is we should be able to manufacture intelligence ourselves, just like we manufacture steel, manufacture cars, manufacture agriculture produce,” Kumar said.

The company is also betting on smaller, specialised models rather than relying entirely on frontier models for every task. Kumar cited Sarvam’s 105-billion-parameter model as an example, saying it costs about one-eleventh as much as models such as GPT Mini or Gemini Flash for a particular use case, while performing better on metrics relevant to voice AI.

The claim highlights a broader shift in enterprise AI, where the most capable model may not necessarily be the most useful one if it is significantly more expensive to deploy.

Also Read: GFF 2026: Axis Bank, J.P. Morgan see banking move from AI that knows customers to AI that acts

Kumar said India’s opportunity could therefore lie in building models and AI systems that are customised for specific tasks, languages and operating environments rather than simply trying to replicate the largest frontier models.

The company is already seeing demand for such applications in India. Over the past 12 months, Sarvam has handled 325 million minutes of voice AI from India, while its stack is currently processing around 400 million API calls a day, Kumar said.

The company is also working on document AI, including with Life Insurance Corporation of India, where it is digitising more than three crore paper forms generated annually, he said.

For financial institutions, Kumar argued that this combination of lower-cost infrastructure, specialised models and sovereign deployment could become particularly important. Banks and other regulated institutions need to be able to run AI without necessarily sending sensitive data outside their control, he said.

That is why, he said, India needs an AI stack spanning domestic compute infrastructure, models, platforms, agents and governance rather than simply relying on a handful of global AI vendors.

The larger risk, according to Kumar, is that India could repeat its earlier pattern of becoming a consumer of technology rather than a builder of it.

“We are frankly more or less, as my co-founder says, digital colonies right now,” Kumar said. “In the next decade, we have the opportunity, at least in this critical space of AI, to be builders and improvers rather than users.”