The Data Centre Mirage, Why Storage Alone Won’t Deliver India’s AI Sovereignty
As Policymakers Court Global Tech Giants, a Critical Question Emerges: Are We Building the Wrong Infrastructure for the AI Revolution?
There is a seductive appeal to data centres. They are vast, gleaming cathedrals of the digital age, humming with the promise of technological prowess. They represent investment, jobs, and a tangible presence for the abstract world of cloud computing. When global tech companies announce plans to build data centres in India, it feels like validation—proof that the world recognises India’s importance in the digital economy.
But seduction is not the same as substance. And as India charts its course in the artificial intelligence revolution, a growing number of analysts are asking a uncomfortable question: are policymakers over-estimating the benefits of hosting the world’s data on Indian soil?
The answer, increasingly, appears to be yes. A proliferation of data centres that store proprietary data in Indian locations is neither necessary nor sufficient to achieve the country’s priority aims in AI. To develop genuine AI capability, India must devote resources to what is vital: indigenous chip design, fabrication, assembly, and testing on the hardware side; and the creation of homegrown AI models on the software side. Without these, the data centre boom may prove to be a mirage—a visible symbol of progress that masks a deeper failure to build sovereign capacity.
The Allure and the Illusion
India’s market for AI services is estimated to be the world’s third largest. The country generates diverse troves of data that AI businesses hunger for—from the digital footprints of hundreds of millions of smartphone users to the vast archives of government records, from the linguistic diversity of a billion conversations to the transaction histories of the world’s fastest-growing digital payments ecosystem. This data is valuable. It is the raw material from which AI models are refined and improved.
But value is not the same as control. Several cutting-edge AI developers have offered their services to India either cheaply or free, precisely to acquire a massive user base that would give them untold giga-loads of data to refine their models further. This is not charity; it is strategy. Social media platforms have demonstrated the pattern: India is among the planet’s most prolific generators of data, but the value of that data has been captured largely by foreign companies. The users produce; the platforms profit.
The same dynamic threatens to play out in AI. Building data centres to store this data does not change the underlying equation. Storage is not sovereignty. It does not give India control over the algorithms that process the data, the models that learn from it, or the applications that deliver value to users. It makes India a storage utility provider to the rest of the world—a role with little strategic upside and significant environmental cost.
The Environmental Price
Data centres consume huge quantities of electricity. While mega projects often claim they will run on renewable energy, the scale of consumption is staggering. A boom in overall power demand from data centres is likely to make it that much harder to decarbonise India’s economy—a challenge that is already daunting given the country’s development needs and climate commitments.
Mere data storage is not the most power-intensive activity. The disks that hold bits need energy, but the demands are modest compared to what comes next. Today’s AI race requires a great deal more to be done with data—from training models to inference operations—that guzzles electricity. Training a large language model can consume as much energy as a small city over weeks or months. Each query to a deployed model requires computational resources that add up across billions of interactions.
High-end processing is presumably what the latest data projects are aiming for. These would require advanced processors of the sort made by Nvidia—its Rubin family of chips, for example—apart from high-bandwidth memory chips that are essential for moving data quickly between processors and storage. But as various scenarios of India’s AI evolution plotted by the Economic Survey 2025-26 show, the country faces the same hurdle in each case: a chip shortage.
This is the critical bottleneck. Without chips, data centres are empty shells. Without indigenous capability to design and fabricate the processors that power AI, India remains dependent on foreign suppliers who can—and will—prioritise their own strategic interests.
The Chip Challenge
The semiconductor supply chain is one of the most complex and concentrated in the global economy. Design tools come from a handful of companies, primarily in the United States. Fabrication is dominated by Taiwan Semiconductor Manufacturing Company (TSMC) in Taiwan and a few other players in South Korea and the United States. Assembly and testing are more dispersed but still concentrated in East Asia.
India has made efforts to enter this ecosystem, with mixed results. The government has announced incentives for semiconductor manufacturing, but the economics are challenging. Building a leading-edge fab costs tens of billions of dollars and requires years of construction before a single chip is produced. The talent pool for chip design is growing but remains small compared to the demand. And even if India succeeds in building fabrication capability, it will be years—perhaps a decade—before it can produce the most advanced processors needed for cutting-edge AI.
This is not to say the effort is not worth making. Strategic autonomy in semiconductors is essential for any country that does not want to be at the mercy of foreign suppliers. But it is to say that the timeline is long, the investment is enormous, and the path is uncertain. In the meantime, India must work with what it has.
The Frugal AI Alternative
Given these constraints, can India make headway with what might be called “frugal AI”? The Economic Survey’s chapter on AI recommends a focus on small language models and specialised applications rather than large foundation models. This is a sensible approach that plays to India’s strengths.
Sarvam, a local model for Indian language text and voice applications, is a good example of what is possible. By focusing on the specific needs of Indian users—multiple languages, voice interfaces for low-literacy populations, integration with local context—such models can deliver value without requiring the massive computational resources of a GPT or Gemini. They can be trained on smaller, curated datasets, run on less powerful hardware, and deployed at scale across the diverse environments of the subcontinent.
Open-source models from the United States and China could also be adapted for local needs. The open-weight models released by Meta, Mistral, and others allow developers to fine-tune and customise without starting from scratch. This reduces the barrier to entry and enables innovation that builds on global research.
But none of this will suffice for purposes of national security. Even if an open-source model has open weights that can be freely tweaked, its developer might still be able to hold users ransom. The terms of use can change. Updates can be withheld. The underlying code may contain hidden vulnerabilities or backdoors. For applications that touch on defence, intelligence, and critical infrastructure, dependence on foreign models is a risk that no sovereign state should accept.
The Sovereignty Imperative
For AI sovereignty to be more than a buzzword, India must indigenously develop the key models it cannot do without. This does not mean building everything from scratch. It means having the capability to build what matters, when it matters, without depending on others.
This requires investment in people. India is one of the few countries in the world with the potential talent for innovative AI: a large supply of young people who can be trained in advanced linear algebra, calculus, and probability. These are the mathematical foundations of modern AI. With the right education and training, this demographic dividend can become a strategic asset.
Another group could work on developing the chips that perform parallel data processing for AI use. The design of AI accelerators—specialised processors optimised for the matrix multiplications that underlie neural networks—is a skill that can be developed in India. The tools for chip design are available, even if fabrication remains elsewhere. By building design capability, India can participate in the higher-value parts of the semiconductor ecosystem while working toward eventual fabrication capacity.
China’s success with fewer resources than the United States shows that determination can make a difference. Despite sanctions and export controls, Chinese companies have developed advanced AI chips and models. They have done so through sustained investment, strategic focus, and a willingness to prioritise long-term capability over short-term returns. India can learn from this example.
Rethinking Incentives
The current policy approach has emphasised attracting foreign investment through subsidies and partnerships. Global tech giants are offered incentives to build data centres, establish operations, and train local talent. These arrangements have benefits, but they also have costs—both financial and strategic.
The money spent on subsidies to attract global majors could instead be deployed to fund local research and startups that hold genuine promise. A thousand crore rupees distributed across a hundred promising AI startups would create a very different ecosystem than the same amount given as a tax break to a foreign company. The startups would be owned and controlled in India. Their intellectual property would remain here. Their success would build domestic capability rather than foreign dependency.
This is not to say that foreign partnerships have no role. They do. Indian researchers and companies can learn from collaborating with global leaders. Indian users can benefit from world-class AI services. But the balance must shift from attraction to cultivation, from hosting to creating.
The Inspiration Gap
India clearly needs to catch up on AI. The gap between where the country is and where it needs to be is substantial. But catching up requires more than building infrastructure; it requires original thinking.
The infatuation with data-centre proliferation must yield to inspiration in the realm of original thinking. Data centres are visible, tangible, and easy to announce at press conferences. They generate headlines and ribbon-cutting ceremonies. Original thinking is harder to measure, harder to celebrate, harder to take credit for. But it is ultimately what matters.
The young people learning linear algebra in classrooms across India today are the ones who will design the algorithms, build the models, and create the applications that define India’s AI future. They need not just infrastructure but inspiration—the sense that they are part of something important, that their work matters, that the country is counting on them.
Conclusion: The Hard Path
The path to AI sovereignty is hard. It requires sustained investment over decades. It requires building educational systems that can produce world-class talent. It requires creating research environments where innovation can flourish. It requires accepting that the results will not come quickly, that there will be failures along the way, that the easy shortcuts—like building data centres and calling it progress—lead nowhere.
But the hard path is also the only path that leads to genuine capability. India has the potential to be not just a user of AI but a creator of it. The demographic dividend, the linguistic diversity, the entrepreneurial energy, the democratic institutions—these are assets that no other country can replicate. They are the foundations on which an Indian AI future can be built.
The data centre boom will continue. It will provide jobs, investment, and a certain kind of visibility. But those who understand what is truly at stake will look beyond the gleaming buildings to the harder questions: Who designs the chips? Who builds the models? Who controls the algorithms that will shape our lives?
The answers to those questions will determine whether India’s AI story is one of sovereignty or dependency, of creation or consumption, of genuine progress or illusory achievement.
Q&A: Unpacking India’s AI Sovereignty Challenge
Q1: Why does the author argue that data centres are not the answer to India’s AI ambitions?
A: The author contends that while data centres are visible symbols of technological progress, they are neither necessary nor sufficient for achieving AI sovereignty. Data centres store data, but storage does not equal control over algorithms, models, or applications. India already generates vast amounts of data that foreign AI companies use to refine their models—often by offering services cheaply or free to acquire users. Building data centres does not change this dynamic; it merely makes India a storage utility provider. True AI capability requires indigenous chip design and fabrication, along with homegrown AI models, not just facilities that house foreign-owned infrastructure.
Q2: What are the environmental costs of a data centre boom?
A: Data centres consume enormous quantities of electricity. While many projects claim they will use renewable energy, the scale of consumption makes decarbonisation more difficult. Mere data storage is not extremely power-intensive, but AI requires far more: training models and running inference operations guzzle electricity. Training a large language model can consume as much energy as a small city over weeks. As India pursues both economic growth and climate commitments, a proliferation of power-hungry data centres could significantly complicate efforts to reduce carbon emissions.
Q3: What is the “chip shortage” challenge India faces in AI development?
A: Advanced AI requires specialised processors, particularly graphics processing units (GPUs) like Nvidia’s Rubin family, along with high-bandwidth memory chips. The Economic Survey 2025-26 scenarios show that India faces the same hurdle in each case: a chip shortage. The semiconductor supply chain is highly concentrated, with fabrication dominated by TSMC in Taiwan and a few other players. Building indigenous fabrication capability costs tens of billions of dollars and takes years. Without chips, data centres are empty shells, and without indigenous chip capability, India remains dependent on foreign suppliers who can prioritise their own strategic interests.
Q4: What is “frugal AI,” and how might it help India make progress despite constraints?
A: Frugal AI refers to focusing on small language models and specialised applications rather than massive foundation models. This approach plays to India’s strengths: developing models for Indian languages, voice interfaces for low-literacy populations, and applications tailored to local contexts. Sarvam, a local model for Indian language text and voice applications, exemplifies this approach. Open-source models from the US and China can also be adapted for local needs. However, the author notes that frugal AI will not suffice for national security applications, where dependence on foreign models—even open-source ones—creates unacceptable risks of being held ransom by developers who control updates, terms, or underlying code.
Q5: What should India do instead of focusing on data centre proliferation?
A: The author argues for redirecting resources to what truly matters: indigenous AI chip design, fabrication, assembly, and testing on the hardware side; and creation of homegrown AI models on the software side. This requires investing in people—training young Indians in advanced linear algebra, calculus, and probability, which are the mathematical foundations of AI. It also means developing chip design talent. Rather than spending subsidies to attract global majors, India should deploy that money to fund local research and promising startups. China’s success despite sanctions shows that determination can make a difference. The infatuation with data centres must yield to inspiration in original thinking.
