India's finance minister announced in February's budget that cloud providers who run workloads from Indian data centres will pay zero taxes on those revenues through 2047 — a forty-year tax holiday designed to pull compute capacity away from Singapore, the UAE, and Ireland and anchor it in India instead. The offer is part of a larger bet: that India can position itself as the third major pole of the AI race, distinct from the United States and China, and use AI to compress the development arc that took other countries generations. Prime Minister Modi's government is not being coy about the ambition. Officials are pitching $200 billion in AI infrastructure investment by 2028. Every major AI CEO attended the India AI Impact Summit in February. The question is whether the ambition is matched by what India can actually deliver, and on what timeline.
Bloomberg's analysis of why the AI industry is "obsessed with India" identifies several converging factors. India has 1.4 billion people and the world's largest population of English-language technology workers — a combination that makes it attractive as both a consumer market and a talent pool. Indian engineers built significant portions of the infrastructure that powers the current AI systems at Google, Microsoft, Amazon, and Anthropic; the country's pipeline of machine learning talent is genuine and deep. The cost of compute in India is low relative to US and European markets, partly because of lower land and energy costs, and the government is actively competing with other jurisdictions for data centre investment in a way that Singapore, the UAE, and Ireland have done for years but at smaller scale. On top of that, Amazon, Microsoft, and Google have collectively pledged $67.5 billion in Indian investments since October, providing the foundation for the infrastructure India's government is promising.
The sovereignty angle is the most interesting part of India's positioning. The Modi government has been explicit that it wants India to be a consumer of AI built on Indian terms — not simply a market for US products or a factory for US-owned compute. This means building domestic model capabilities, retaining data onshore, and developing an industrial policy for AI that parallels the semiconductor industrial policies that South Korea and Taiwan executed for chips in earlier decades. The zero-tax offer is one instrument of that policy; the $1.1 billion state-backed venture fund for AI startups is another. Whether these instruments are sufficient to produce competitive Indian AI companies, rather than simply attracting foreign investment into Indian-headquartered operations, is the harder question. The history of technology industrial policy suggests that subsidy and tax incentive can attract capital but cannot reliably produce the research culture that generates frontier capability.
Adani's announcement of $100 billion in AI data centre investment by 2035 — to be powered by renewable energy — puts Indian private capital alongside foreign investment in ways that make the ambition feel more grounded than a government announcement alone would. But Adani is building infrastructure, not models, and the distinction matters. Data centres attract workloads; they do not automatically produce the research institutions, talent density, or competitive product ecosystems that make a country genuinely consequential in AI development rather than a host geography for foreign companies' servers.
The European and Canadian situation provides a useful contrast. A Washington Post commentary published last week argued that Europe and Canada are "hoping to chart a course independent of China and the United States" but have "no country that can fully decouple from US or Chinese models anytime soon." Europe has regulatory leverage — the AI Act is a serious regulatory framework that the US has not matched — but relatively little product leverage: no European company is in contention for frontier model leadership. Canada's claim to AI relevance rests heavily on the academic research lineage (Hinton, Bengio, LeCun all trained there) that predates the current commercial era by a decade. Neither Europe nor Canada has executed an India-style aggressive industrial policy for AI infrastructure, and neither has India's combination of population scale, English-language technical workforce, and government willingness to compete on subsidy terms.
What makes India's play genuinely interesting, rather than just another country announcing ambitious AI targets, is the specificity of the instruments and the timeline of the commitments. A forty-year tax holiday is not a press release; it is a binding fiscal commitment that will outlast multiple governments. The Indian government has structured this as an infrastructure play designed to attract foreign hyperscaler investment, which is a more realistic near-term target than producing frontier models domestically. If it works, India becomes the data centre hub for a significant share of global AI workloads that are currently run from higher-cost geographies — capturing the economic activity of the infrastructure layer even if the model layer remains American or Chinese. That is a different kind of AI sovereignty than building a domestic GPT, but it is a credible and potentially very large economic prize.