Sanjeev Ahluwalia | Why India’s Sovereign AI Requires Global Collaborations

Sovereignty, defined tightly, includes the digital infrastructure — software, datacentres, electrical power, cooling including water and data — should all be Indian, located in India regulated under Indian law with commercial disputes settled only in India. This resembles the Chinese approach to data sovereignty where direct control equals power. It maxes security and resilience. But does it ensure spontaneous innovation and efficiency and at what cost?

Geopolitics vindicates this approach. Nothing is certain unless it is under a country’s direct control. For more than eight decades post-World War II, Japan had outsourced its national security to the United States — an arrangement, which today, provokes unease. Even US support to Nato now appears conditional.

The problem with drawing a moat around AI capacity is that typically the science around it has evolved collaboratively since the second quarter of the 20th century when the computer was being developed. There is some lack of unanimity around who made the first computer depending on whether you qualify hardware with electro-mechanical components — low efficiency gears and rotating drums — or only more efficient electronic circuit design.

Similarly, purpose matters. Specific purpose machines — like World War II code breakers — were not programmable for general tasks. Computers are flexible because it is software which targets specific tasks to be accomplished. Incidentally, during the WW-2 years, women were ace programmers. Men dominated the hardware innovations. In today’s gender-neutral environment such “allocated” gender roles are abhorrent — a social advance not to be sniffed at. AI, for instance, makes no sense if it reverses social advancement like equity and environmental sustainability.

The race for AI dominance is about securing the critical minerals and the capacity to refine them, invest in fabs to make chips, ensure adequate power supply and cooling alternatives to scarce water for data centre capacity, develop the R&D facilities which feed into hardware and software, cultivate the entrepreneurship to manage big enterprises and put in place supportive regulations.

China defines next steps in an investment centric “six networks” vision. Till 2030, it will spend three per cent of its cumulative GDP till then or US$3.7 trillion (slightly less than India’s annual GDP). About one half is earmarked for upgrading water supply distribution in the arid West where data centres are planned, strengthening the grid with dispersed battery storage, additional compute capacity and communication facilities. The purpose is not just to be ahead in AI. Big investment spends in logistics and underground urban pipe infra are all intended to create domestic demand to soak up geopolitically sensitive trade surpluses.

Unlike China, India is constrained by a trade and current account deficit and must keep fiscal allocations severely targeted. It must service a population larger than China with one-fifth its resources. This fiscal constraint will define how India builds its AI strategy. Collaborations are an option. The existing operational capacity of data centres is about 2.2 MW across 185 establishments and 29 providers. An additional 1.6 MW is under construction with a possible total capacity of 5 MW by 2030.

Scaling up will be constrained by water supply. India ranks 125th out of 185 countries on per capita availability of fresh water, well below Iran though better than Iraq. Nearly 85 per cent of water is used for agriculture but very inefficiently. Enhancing the efficiency of use and crop diversification away from water hungry crops like rice and sugar is a thorny road to freeing up supply. Water supply rates are minimal at best, so overuse is rampant as is the use of cheap fertiliser.

About 40 per cent of workers are in the agricultural economy. Till this share drops below 20 per cent — via absorption in manufacturing or services — political economy factors will constrain agricultural reform. India’s political parties are sensitive to the loss of vote share. This points to a careful selection of data centre clusters, along the coastal regions where desalinated water supply is an option along with waterless cooling options. In any case, proximity to global subsea data transmission links make such locations desirable.

“Local” India tends to intrude into “Global” India. State governments are the primary partners for project completion. In May 2026 Andhra Pradesh issued a local distribution licence to Google’s 1GW data centre being built near Visakhapatnam. Consider now a contrary experience from Haryana. The state government put on hold an order (the first of its kind) of the local State Electricity Regulatory Commission granting a parallel distribution licence to a private licensee, in addition to the existing state-owned distribution company, in the Gurugram and Nuh area. State electricity workers promptly went on strike forcing the government to stay operation of the order.

Note that the regulator had overruled the objections of the state distribution company regarding potential loss of jobs and reduction in revenue if the creamy layer of consumers — industry and large consumers, who cross-subsidise other consumers — were weaned away by the improved supply and competitive rates of the future private supplier. New demand is plentiful. About 0.1 million applications for an additional load of 1.6 GW are pending, which the state-owned distributor is unable to supply. Rather than support efficiency enhancement and promote growth by meeting electricity demand the state government preferred to maintain the status quo.

This is notwithstanding that enhancing efficiency and competition are key objectives of the regulator. The Union government also endorses competition in distribution supply as an index of productivity. The matter will now be decided judicially in appeal. But for the private licensee and its hardware suppliers, which include start-ups supplying Battery Energy Stored Support, it is a bitter lesson, how domestic regulatory uncertainty increases cost and lowers profit.

The lesson is that for implementing sovereign AI infrastructure in India, it must be costless for the average citizen, with enhanced efficiency eking out the required returns on investment. Given the level of pricing distortions in the economy already financed by government, this points to the investment burden falling on the Indian private sector or foreign direct investment.

“Sovereignty” then must be expanded to include safeguarding the rights of private investors and collaborators via explicit guarantees and regulations, pre-tested for meeting exacting legal challenges in India and abroad. Ditto for safeguard and security measures. For forging a consistent global approach, plurilateral working groups with potential partners across the supply chain — the US, the UK, Germany, Japan, South Korea and Australia are recommended.

Sanjeev Ahluwalia is distinguished fellow Chintan Research Foundation and was previously in the IAS and the World Bank.