Dr Vivek Raghavan, Co-founder of Sarvam AI, stated at the Bengaluru Tech Summit 2026 that India must build a sovereign AI stack to prevent becoming a digital colony. He emphasized the importance of domestic hardware, Indic-language models, and locally hosted infrastructure.

India must develop its own artificial intelligence infrastructure, models, and applications rather than depending heavily on global technology providers, according to Dr Vivek Raghavan, Co-founder of Sarvam AI.

Speaking at the 29th edition of the Bengaluru Tech Summit 2026 on Tuesday, Raghavan cautioned that excessive reliance on foreign AI systems risks turning the country into a digital colony. He noted that while India is a large user of tools like ChatGPT, this dynamic results in exporting local data and importing finished intelligence.

Highlighting India's demographics, Raghavan pointed out that the country has nearly a billion active internet users, with over 500 million located in rural areas. He stated that 98 per cent of rural internet users access the internet in their native languages at least occasionally, and nearly 57 per cent prefer regional languages. Given India's linguistic diversity—including 22 scheduled languages and around 6,000 dialects—voice interfaces and Indic-language support must be prioritized.

On the infrastructure front, Raghavan acknowledged that India is not yet close to building its own chips for AI workloads. Consequently, the country will need diversified hardware sourcing alongside a long-term strategy to build domestic capabilities. Furthermore, he stressed that computing infrastructure, including GPUs and cloud facilities, should be located in India and operated by Indian entities, particularly for sensitive functions and regulated industries.

Addressing the data layer, Raghavan called for local datasets, Indic-language corpora, and benchmarks to measure AI performance accurately within India's cultural and linguistic context. While domestic companies should eventually build models from scratch, open-source models can be utilized in the interim if they remain under Indian control. He defined a sovereign model as one where the sovereign authority can audit the data used during training. Publicly funded scientific data, school textbooks, parliamentary debates, court judgments, and media archives were cited as uniquely Indian datasets suitable for training differentiated models.

Sarvam AI has previously built a 105-billion-parameter model using nearly 16 trillion tokens and announced plans to develop a trillion-parameter model trained entirely from scratch.

Looking ahead, Raghavan introduced the concept of universal basic intelligence as an alternative to universal basic income, advocating for personal AI agents that communicate in local languages and help citizens navigate public services and entitlements. He noted that as AI becomes integrated into every job, widespread access will be essential to prepare the workforce for shifting occupational requirements.

"Dr Vivek Raghavan’s perspective at the Bengaluru Tech Summit highlights a critical strategic priority for the Indian technology ecosystem. Building domestic AI infrastructure and sovereign models is not just about technological self-reliance, but also about safeguarding data sovereignty and catering to our linguistic diversity. As Indian startups and enterprises scale artificial intelligence adoption, focusing on Indic languages, localized data, and domestic cloud hosting will be essential for creating sustainable, culturally relevant solutions." — Dr. Shishir Gupta, Founder & CEO, StartupLanes

Recent StartupLanes Articles

Browse through our 30 latest publications on venture capital, startups, and angel investing.