India will lead in AI adoption, scale over next decade: Sarvam cofounder

India's leading advantage in artificial intelligence may not reside in constructing the most potent models, but rather in deploying AI at a cost accessible to a billion people, asserts Sarvam AI co-founder and CEO Pratyush Kumar at the Global Fintech Fest 2026. Even as AI has entered its fourth year since the launch of OpenAI's ChatGPT, India is only at the nascent stage of its AI development journey.

This emerging nation now boasts its own AI models, increasing access to Graphics Processing Units (GPUs), and a burgeoning pool of AI talent, Kumar said.

Kumar believes that over the next ten years, India is set to become a global leader in AI, both in terms of adoption and creation. However, the opportunity for India transcends merely creating indigenous models. It necessitates the establishment of infrastructure capable of producing and running AI at scale while maintaining a cost structure that caters to India's vast population. This holds the potential to become the unique selling point for the global AI mission, according to Kumar.

Sarvam AI, which is concurrently developing its own large language models and deploying open models, is reportedly already witnessing the profitability of this approach. The company's models, when run on Indian GPUs, cost "equal or less" than those available globally. Additionally, Sarvam is facilitating models developed by other providers on its infrastructure.

Kumar further emphasized the emergence of "token factories" - infrastructure that can serve AI models at scale, akin to factories manufacturing physical goods. India should strive to manufacture its intelligence similarly to how it manufactures steel, cars, and agricultural produce. He cited Sarvam's 105-billion-parameter model as evidence, stating that it costs about one-eleventh as much as models like GPT Mini or Gemini Flash for specific use cases, while outperforming them on voice AI metrics.

This reflects a broader trend in enterprise AI, where the most capable model may not always be the most practical one if it is prohibitively expensive to deploy.

India's potential, Kumar argues, lies in building models and AI systems tailored to specific tasks, languages, and operational environments rather than merely attempting to replicate the largest frontier models. Currently, Sarvam is already experiencing demand for such applications in India. Over the past year, the company has processed 325 million minutes of voice AI from India, and its platform handles approximately 400 million API calls daily.

Furthermore, Sarvam is working on document AI, including a collaboration with Life Insurance Corporation of India, where they are digitizing over three crore paper forms annually.

For financial institutions, this combination of lower-cost infrastructure, specialized models, and sovereign deployment could prove especially beneficial. Banks and other regulated entities need to be able to run AI without transmitting sensitive data outside their jurisdiction, Kumar said. Therefore, India needs an AI stack encompassing domestic compute infrastructure, models, platforms, agents, and governance, rather than merely depending on a handful of global AI vendors.

Kumar warns that India risks becoming a mere consumer of technology instead of a builder. "We are, frankly, digital colonies right now," he said. "In the next decade, we have the opportunity, at least in this critical space of AI, to become builders and improvers, not just users."