Artificial intelligence looks to be housed within a screen. A user inputs a question and, moments later, receives a reply. This has the feel of something immaterial and instantaneous, something that barely exists outside of code. But behind the response lies physical reality.
Specialized chips, data-center infrastructure, cooling systems, power supplies, fiber optics, personnel, and land. These are the reasons why HCLTech’s proposed artificial intelligence data center in Bhubaneswar, which will cost ₹14,257 crore to build, is significant. The project fits into a broader Indian push to exert more control over the infrastructure and data that make up modern artificial intelligence (HCLTech). The competition in artificial intelligence will be fought through models and applications. Most of the value will be in data centers full of graphics processing units.
Intelligence Requires Compute
To get an idea of why, it’s useful to think of how modern artificial intelligence works. Artificial intelligence models require an extensive amount of mathematical calculations to train. The best processors for performing these calculations, GPUs, were initially used for graphics rendering but have since found a second life as the foundation for training and operating large-scale artificial intelligence models. The code underlying a particular model can be replicated relatively easily.
What’s far more valuable is the ability to train the model on a large amount of data using specialized silicon, which is a scarce resource. A promising but small startup with a compelling use case and sufficient data may lack the resources to train its model on available GPUs. The company would then have to rely on renting compute power from large cloud infrastructure providers, who in turn have an incentive to raise prices. Compute becomes analogous to industrial capacity. An automobile manufacturer needs factories; an airline needs airplanes; an artificial intelligence company needs access to chips inside data centers. The final product remains intangible, but the means of its creation are physical.
India Is Building Shared Compute
India aims to reduce its reliance on foreign infrastructure through a program called IndiaAI Mission. India allocated ₹10,372 crore to the mission over five years, and it has seen success so far. As of March 2026, over 38,000 GPUs were already onboarded across the common-compute system, which provides startups, researchers, universities, and government organizations with access to GPUs at subsidized rates. The government has also earmarked another 20,000 GPUs for the shared system (Press Information Bureau). The shared model effectively addresses the economic challenge of limited compute availability. At the individual level, few small firms will be able to purchase and maintain a large-scale GPU cluster. Even if capital is available, the utilization of such an asset may be low if there are periodic spikes in demand.
A shared system allows multiple users to rent compute capacity when needed. The system is similar to how cloud computing already works for standard enterprise applications, but India’s program treats affordable AI compute as an essential public good. IndiaAI Mission has positioned compute as a “strategic enabler” that should be available to a wide pool of users rather than concentrated within the reach of a few large technology companies (Press Information Bureau). The goal is not just to have more chips; it’s to ensure that the right people are the ones using them.
What Sovereign AI Actually Means
The phrase “sovereign AI” has a certain ring to it but may involve several meanings at once. At the most basic level, economic sovereignty implies control over different parts of the supply chain. Where is the data stored? What laws apply to the infrastructure? What company has the ability to restrict or allow access and change costs? Can the model understand India’s multilingual environment and institutions? Can it continue functioning if there is a geopolitical or commercial dispute?
An AI model that was mostly trained on English data may not work as well on India’s diverse range of languages. An AI model that was only deployed via foreign infrastructure may raise questions about privacy, procurement, or continuity. India’s sovereign-model push aims to create systems that utilize local data and Indian languages for areas such as healthcare, farming, education, and government. Twelve teams were shortlisted for the IndiaAI foundational-model program’s first phase, and Sarvam AI and BharatGen have already presented their models at the 2026 India AI Impact Summit (Press Information Bureau). Sovereignty does not always mean self-sufficiency. It seeks to minimize the risk that critical choices will be made elsewhere.
The Chips May Still Come From Abroad
A data center in Odisha is not the same as a self-sufficient supply chain for semiconductors. Fewer companies control the cutting-edge chips at the heart of modern AI. India’s own efforts at IndiaAI acknowledge this reality, noting that “advanced GPUs are primarily produced outside India, and access to them is a strategic challenge” (Press Information Bureau). The data center can be built in Odisha. The chips can still be produced abroad. This is an important distinction, as it highlights the limits of economic sovereignty.
India does not need to have complete control over every aspect of the AI supply chain to benefit from it. The country’s position on AI is one of relative leverage: having more influence than a small tech firm but less control than a global semiconductor manufacturer. Sovereignty is not a binary but rather a spectrum. Having one’s own data center is better than relying on a foreign one, but neither is as desirable as building an entire semiconductor foundry. Each step reduces dependencies but does not eliminate them. A company that rents servers from a local provider still has fewer options than one that owns the servers.
A Data Center Is a Capital-Intensive Bet
The ₹14,257 crore investment that HCLTech has planned for its data center is a substantial one. The company has to consider the long-term viability of the project. Data centers are capital-intensive infrastructure that require power, cooling, land, and networking equipment. The most valuable assets, the GPUs, become obsolete quickly. A data center’s economics are ultimately driven by utilization.
The moment a GPU is turned off, it stops generating revenue. There needs to be a guarantee that there will be enough organizations, both private and governmental, that will need the center’s capacity. Long-term contracts and government contracts help stabilize the outlook. Having an AI park also helps attract affiliated businesses and organizations. HCLTech’s center is set to be built with the support of Sarvam and the government of Odisha and will serve as a focal point for industry-specific applications for private and public-sector organizations while also nurturing a local developer ecosystem (HCLTech). The bets that HCLTech and other companies are making are not only on the value of data-center rentals. They are also hoping that compute capacity will draw in an entire industry around it.
Cheap Compute Can Determine Who Innovates
Having access to affordable compute power can transform the landscape for innovation. The high cost of experimentation creates an environment where only established firms can afford to develop and test numerous ideas. If a startup has only one shot at success, it needs to be acutely aware of its chances before investing significant resources. Large technology companies, by contrast, can afford to experiment and iterate. Access to cheaper, shared compute resources bridges this gap, enabling a broader range of organizations to engage in research and development. Researchers can explore the creation of models that perform specific tasks, such as translation of India’s diverse languages. Hospitals can engage in medical-domain research. Startups can develop applications for local farming and weather needs. Government agencies can build their own solutions without relying on foreign infrastructure for every initiative. Not every endeavor will be successful, but that is the nature of experimentation at scale. Innovation thrives when a variety of entities have the opportunity to fail at a manageable cost.
The AI Cloud Has a Physical Address
Artificial intelligence is often conceptualized in abstract terms, as something that emerges from the “cloud.” But the cloud is, in fact, a network of physical servers housed within data centers. These facilities exert a certain amount of control over who can train models, where data can reside, how much processing power is affordable, and which regulations apply. In this way, data centers are akin to more traditional infrastructure projects, such as ports or power plants. They are physical manifestations of economic infrastructure. Software defines the capabilities of artificial intelligence, but raw computational power dictates the extent of experimentation that can occur. India’s AI future will be shaped not only by the skill of its programmers and the richness of its datasets but also by the presence of large-scale data centers willing to provide the necessary compute resources. Artificial intelligence may be spoken of in ethereal terms, but its strength is generated within concrete walls.