India’s rapid expansion of Artificial Intelligence (AI) data centers is fundamentally reshaping the nation with digital infrastructure. However, this growth is predicated on two invisible subsidies: the extraction of water from over-exploited aquifers and the dispossession of land from marginalized communities, particularly Dalits. This article examines the intersection of technological advancement and structural inequality. By analyzing water consumption mechanics, land acquisition patterns, and the energy-caste nexus, this research demonstrates that the current hyperscale infrastructure boom reproduces historical caste-based resource extraction. Drawing upon Dr. B.R. Ambedkar’s framework of land as dignity and state socialism, the argument here is that without structural reforms, India’s AI ambitions will deepen the very inequalities it claims to bridge.
The Invisible Subsidies of the Digital Age
The narrative of India’s technological dominance is often framed as a triumph of modernization and economic growth. However, if we go into the details and the on-ground reality, the nation's data center capacity has quadrupled from 375 MW in 2020 to 1.5 GW by 2025, with projections indicating a surge to 13.5 GW by 2031-32. However, this digital architecture is just the tip, if we try to dig deeper, rooted in the physical world, demanding immense quantities of land, water, and energy. Dr. B.R. Ambedkar observed that “Democracy in India is only a top-dressing on an Indian soil which is essentially undemocratic”. This observation remains profoundly relevant when examining the foundational resources fueling the AI revolution.
The infrastructure required for a global AI future is being built in India, but the costs are disproportionately borne by those who benefit least from its outputs. Let’s analyze the state of resource extraction and how what is driving India’s data center boom, while focusing on the displacement of Dalit communities and the depletion of vital water resources in already stressed regions.
The Metabolism of a Data Center: Water Extraction and Scarcity
Every AI interaction carries a resource signature that remains hidden from the end-user. The thermal intensity of AI workloads requires substantial cooling, predominantly achieved in India through evaporative cooling systems. These systems lose approximately 70-80% of their intake water to evaporation, permanently removing it from the local hydrological cycle.
The scale of this consumption is weirdly surprising. India’s data centers consumed an estimated 121.74 billion litres of water in 2024, a figure projected to reach 358.66 billion litres by 2030. This trajectory must be contextualized against India’s broader groundwater crisis. The nation extracts 25% of the world’s groundwater, and nearly 60% of its non-renewable groundwater is consumed annually.
The geographic concentration of these facilities exacerbates the crisis. Data centers are predominantly located in cities identified as potential “Day Zero” candidates like urban centers that could effectively run out of usable water.
| City | Status | Water Stress Context |
|---|---|---|
| Mumbai | Largest Market (~25%) | Severe; 50% of the population in informal settlements competes for supply. |
| Chennai | Second Largest Hub | Experienced near-Day Zero conditions in 2019; aquifers are severely depleted. |
| Hyderabad | Third Largest Hub | Groundwater is declining; Amazon data center regions show significant drops. |
| Bengaluru | Growing Hub | Facing the “worst water crisis in five centuries” (2024); data centers consume 26+ million litres/year. |
| Delhi-NCR | Growing Hub | Groundwater is critically overexploited. |
In Visakhapatnam, where Google is constructing a 1-gigawatt campus, groundwater levels declined from 19.62 meters below ground level (mbgl) in April 2025 to 28.71 mbgl in April 2026. This localized depletion highlights the stark reality i.e. a facility drawing millions of liters daily from an over-exploited aquifer creates immediate, localized stress while manifesting as deepened borewells, reliance on water tankers, and crop failures for surrounding communities.
Land Dispossession: Reversing Ambedkarian Reforms
Dr. Ambedkar emphasized that land is not merely an economic asset but a crucial marker of social status and dignity. He identified the intersection of caste and landlessness as foundational to India’s inequality structure. Contemporary data validates this analysis, with 70% of Dalits remaining landless labourers. Dr. Ambedkar’s proposals for agrarian reform, including the allocation of government land for Dalit settlements, resulted in modest redistribution programs in the 1960s and 1970s. Tragically, it is precisely this redistributed land that is now being targeted for data center construction.
The Google-Adani project in Visakhapatnam exemplifies this reversal. The project involves acquiring 480 acres, of which 200 acres in Tarluvada belong to Dalit families. This land was allotted to landless Dalit families in the 1970s by the Andhra Pradesh government. Pyla Kondamma, a farmer and former village council head from Tarluvada, articulated the discriminatory nature of this acquisition: “They are not touching land owned by dominant castes. Only Dalit land”. The renaming of the village to “Tarluvada IT Hub” on Google Maps serves as a symbolic erasure of the community’s identity, preceding their physical displacement.
When we look at the surface level issues, we generally associate the violence of this dispossession as a bureaucratic issue but if we look beyond, it is far beyond that, it is often physical. Recently, in June 2024, bulldozers demolished Jai Bhim Nagar in Powai, Mumbai, a settlement of approximately 650 mostly Dalit households to make way for the Hiranandani Group's data center expansion. The eviction, carried out during the monsoon season in violation of protective laws, resulted in arrests and injuries among the residents. The Bombay High Court later found that the government commission ordering the demolition lacked the authority to do so.
The Energy-Caste Nexus and Environmental Racism
The energy demands of AI data centers are reversing coal retirement timelines, compounding environmental injustice. In Ennore, Chennai, a fishing community that has spent decades absorbing pollution from coal plants, now faces a new facility approved in 2025 to meet data center power demands. The health costs and higher rates of childhood cancer, respiratory illness, and infertility are overwhelmingly borne by Scheduled Caste and fishing communities, populations with the least political power to resist such siting decisions.
Furthermore, the reliance on diesel generators for backup power introduces a hidden emission source. The two Visakhapatnam data center projects alone involve approximately 354 backup diesel generators with a combined capacity of 971.5 MW. These generators emit significantly higher levels of nitrogen oxides and particulate matter compared to natural gas plants, directly contributing to respiratory diseases in surrounding communities.
The Distribution of Digital Gains
The distributive case for India’s data center boom rests on the premise that AI tools will reach populations bypassed by previous technology waves. Initiatives like the Kisan e-Mitra chatbot and BharatGen demonstrate the potential for vernacular AI applications. However, these programs sit alongside a structural reality i.e. the infrastructure is built in water-stressed megacities, on land acquired from marginalized farmers to process requests primarily benefiting global enterprises and urban professionals.
Without deliberate redistribution of AI’s gains through public compute access, vernacular AI tooling, and infrastructure siting that prioritizes proximity to need rather than existing grid infrastructure, the data center boom will deepen existing inequalities. As Dr. Ambedkar argued in States and Minorities, “The real remedy is not in the hands of the individuals but in the hands of the State. The State should take the responsibility of industrialization”. The current policy architecture, which grants data centers so called “infrastructure status” and bypasses detailed Environmental Impact Assessments, prioritizes investment attraction over social equity and environmental sustainability.
India’s AI infrastructure expansion represents a modern iteration of caste-based resource extraction. The invisible subsidies of water and land are drawn from the most vulnerable segments of society, reproducing the structural inequalities that should be dismantled. To build a genuinely equitable digital future, the Indian government must implement mandatory Environmental Impact Assessments, independent water auditing, and enforceable consent mechanisms for land acquisition. The infrastructure choices made today will lock in impacts for decades; they must be guided by constitutional morality and a commitment to social justice, ensuring that the architecture of the future does not become a monument to inequality.