Odisha and Tamil Nadu have both signed deals with Bengaluru-based Sarvam AI to build their own compute facilities, train their own models, and deploy AI tools in their own languages. On February 6, Odisha signed an MOU with Sarvam at the Black Swan Summit with President Droupadi Murmu present to develop a 50-megawatt AI-optimised facility. Tamil Nadu went first—on January 13, the state formalised a ₹10,000 crore deal (approximately $1.2 billion USD over five years) for what's being called Digital Sangam, India's first Sovereign AI Research Park.
The Tamil Nadu project is a public-private partnership. Sarvam will bring the investment while the government provides land and support, according to Tamil Nadu's Industries Minister Dr. TRB Rajaa. The park will integrate a 20-megawatt data centre, research labs run by IIT Madras, and startup incubation spaces. Sarvam described it in a blog post as a dedicated physical ecosystem "where advanced compute infrastructure, frontier research, and startup incubation co-locate to attract and grow the world's best AI talent."
Both states are focused on keeping sensitive data within their borders. Sarvam's blog post explains: "This sovereign infrastructure will ensure sensitive data remains within our boundaries and that our AI systems reflect the deep pride in Tamil language and culture."
Emmanuel Oyedeji
What's Actually Getting Built
Odisha's facility has a dual purpose. According to Sarvam: "First, it will act as the state's own AI public utility and apply to the state's core strengths: mining, heavy industrials, and skilling. Vision AI will ensure world-class safety and compliance at industrial sites, while Odia-to-English voice tools will scale job readiness for youth."
The second purpose connects to a broader national strategy. "Odisha is positioning itself as India's compute backbone," Sarvam stated. "Our hub will not only serve Odisha but will anchor a nationwide compute grid. We will offer reliable, production-grade AI capacity to other states and national platforms."
Tamil Nadu has specific applications in mind. "With AI, Vivasāya Nanban assistant will provide 24×7 digital advisory to 79 lakh farm households," the company wrote. That's 7.9 million households. The state also plans a unified AI-powered citizen helpline and tools for welfare and public services.
Chief Secretary Anu Garg noted that Odisha's shift from a "mine-driven economy" to a "mind-driven economy" aligns with the "Viksit Bharat 2047" vision.
The Digital Public Infrastructure Layer
Sarvam's larger ambition goes beyond individual state projects. The company wrote: "India stands at a historic crossroads where the 'data dividend' of our population meets the necessity of digital autonomy. The economic implications are substantial. Choices made now about AI infrastructure will shape whether intelligence compounds within the country or accrues elsewhere."
The technical architecture relies on what Sarvam calls a Digital Public Infrastructure layer. According to the company: "A Digital Public Infrastructure layer that allows intelligence to be shared, reused, and adopted across systems. Together, these layers ensure that learning, value, and capability remain within national boundaries."
Think of it as common infrastructure with different applications on top. States can build their own AI tools while still connecting to a shared grid.
Sarvam AI was selected by the Indian government's IndiaAI Mission to build India's first sovereign large language model. Founders Pratyush Kumar and Vivek Raghavan, who started the company in 2023, are building what they describe as a "full-stack sovereign AI stack end to end. From models grounded in Indian data to applications deployed across population-scale use cases."
Their Sarvam Vision OCR tool recently hit 84.3% accuracy on the olmOCR-Bench, outperforming ChatGPT and Google Gemini 3 Pro on Indian scripts.
The Open Questions
The timing matters. India's AI Impact Summit runs February 16–20 in New Delhi. The government has deployed 38,000 GPUs and is adding 3,000 next-generation GPUs for strategic use cases. Sarvam's state deals give them something concrete to showcase: AI infrastructure that's actually running, not just planned.
But the interoperability challenge remains unresolved. If Odisha builds one stack, Tamil Nadu builds another, and 26 other states eventually want their own, do they actually work together? Sarvam's DPI layer is supposed to solve this by creating shared infrastructure while preserving state control, but the technical details on how that coordination works across different vendors, models, and governance structures haven't been spelt out publicly yet.
There's also the funding question. Sarvam has raised approximately $53 million to date, according to Tracxn. The Tamil Nadu project alone requires ₹10,000 crore over five years. How the company plans to scale from tens of millions in venture funding to over a billion dollars in infrastructure investment hasn't been publicly disclosed beyond Rajaa's statement that "Sarvam will bring in the investment."
Still, the state deals are moving forward. Whether this distributed approach creates a functional national AI grid or fragments into disconnected systems will depend on execution in the coming months.
Kelechi Edeh
