On July 24, HCLTech signed a deal with the Odisha government and the AI startup Sarvam to build a data centre in Bhubaneswar. The number attached to it, ₹14,257 crore, about $1.48 billion, is the part every headline led with. What actually matters is what that money buys: not more computing power, but control over where that power sits.
Here's what "sovereign AI" doesn't mean. It doesn't mean India is building its own chips or training models from scratch without foreign technology. The servers inside this data centre will almost certainly run on Nvidia hardware, like almost every AI data centre on earth. What it does mean is narrower: the physical building, the data stored inside it, and the servers processing that data will sit on Indian soil, under Indian law, instead of on a server owned by a foreign company.
That's a bigger difference than it seems. Today, if an Indian ministry or a public bank runs an AI system, it is typically hosted on servers owned by Amazon or Microsoft, anywhere those companies decide to put them. The data can be subject to the laws of another country; the terms can be changed with little input from India. This kind of facility removes that dependency for anything sensitive, government records, defence-linked research, financial data. It's the same reason countries built their own oil refineries rather than just buying refined fuel from someone else. Ownership of crude oil is meaningless if another country controls the refinery and dictates who gets fuel and when.
The deal needs three different partners because no single one of them has everything required. HCLTech will fund the physical building with its own funds of up to ₹3,500 crore. What makes the system useful for Indian government and business use, and not just another global chatbot, is that Sarvam brings the AI models trained specifically on Indian languages. Odisha provides land and subsidised power, the unglamorous stuff that makes the whole thing financially possible. But take away one piece, and the project falls apart.
Now the part that gets skipped over: jobs. The project will build a technology centre with 5,000 seats by 2028. Divide that into the total investment and you get something like ₹28 crore for every eventual job. And that's before you factor in the years of construction before anyone gets a job. It is not a mistake of the project. Data centres are designed to run on machines, not people. But it's worth saying because the politics around projects like this often suggest a lot more job creation than the economics actually provide. The jobs will be real but few and highly skilled, in line with a pattern seen across India's industrial data this year: output growing faster than hiring.
The bigger question isn't whether this one data centre gets built. It will. It's whether other states start offering the same deal, land and subsidies in exchange for private capital, and whether that turns into a race that only a few states can actually afford to enter.
