Policymaking for AI Data Infrastructures: Nurturing a Healthy and Sustainable AI Ecosystem in India

Preeti Raghunath

Abstract

India is witnessing an increase in investments into AI data centres, including what is claimed to be the world’s largest. With a strategic vision aimed at incentivising foreign investments in AI data centres, this essay argues for safeguards for local resources, livelihoods, and environmental impacts. It examines the socio-ecological costs of AI technologies and associated data infrastructures, building on ongoing policy analysis and qualitative research in Visakhapatnam, Andhra Pradesh. It maps the policy ecology for AI data centres in India, including the intersections of policies and practices governing energy, water, and land allocation, contracts, and procurement. Situating India’s strategic approach in the wider international context, this essay identifies gaps in impact assessments, social and technical audits, and crisis preparedness, and proposes recommendations for cross-sectoral policies for just and equitable AI futures.

India is witnessing increasing investments in hyperscalers and AI data centres, including what is claimed to be the world’s largest data centre. At the AI Impact Summit hosted in New Delhi in February 2026, Big Tech companies committed to a record USD 200 billion in investments in AI data centres in India, aimed at serving a rapidly growing global AI economy. From Microsoft and Google to OpenAI and BlackRock, major investment deals and projected targets for 2030 were announced. India also joined Pax Silica, a technology alliance spearheaded by the United States to strengthen semiconductor supply chains amidst intensifying geopolitical competition with China. 

If anything, the Summit highlighted India’s ambition to attract and host global co-location data centres and hyperscalers. Just days before the Summit, the Union Budget 2026 introduced measures offering tax holidays and safe-harbour provisions to foreign companies setting up data centres in India until 2047. These developments reflect a clear strategic intent: incentivising foreign investments in data infrastructures, particularly AI data centres. Yet any vision aimed at creating enabling ecosystems for AI capabilities, development, and deployment in India must also build in safeguards for local resources, livelihoods, and broader socio-environmental impact.

This essay examines the socio-ecological costs of AI technologies and associated data infrastructures, building on ongoing qualitative research. The concept of “infrastructure” traverses multiple registers, from its association with early colonial modernity to its present-day role as a key expression of governance in the neoliberal state. In social sciences literature, infrastructure functions both as a singular concept that captures this imperative for modernity and as a collective phenomenon through which its plural workings are manifested, mobilised, and studied. This essay approaches infrastructure as a site where the interests of global tech capital interface with, and at times evade, the Indian state’s regulatory definition. It also adopts a broad definition of sustainability, encompassing both environmental impacts and resource economies as well as questions of social equity and justice. Indeed, social and ecological sustainability are not separate. They intersect to produce multiple marginalities, as resource depletion and inadequate access produce differentiated impacts, often exacerbating existing forms of marginalisation. Drawing on these conceptual foundations, this essay explores the interactions between policies and practices in energy, water, and land allocation, contracts and procurement, as they stand today. In doing so, it draws on international and national practices in developing environmentally sound critical infrastructures across sectors. The following section contextualises these issues within India’s AI strategy and identifies policy gaps, before presenting a set of recommendations for cross-sectoral policies that can foster a healthy and sustainable AI ecosystem in the country.

The recommendations offered in this essay include both immediate and medium-term ones, with the aim of ensuring that investments in AI data centres serve the larger public interest and generate lasting benefits for the country. India should classify AI data centres as a distinct category of critical infrastructure, reflecting their unique resource demands and socio-ecological impacts. It should also move from voluntary reporting to mandatory, independently verified disclosures on energy, water, and biodiversity use to ensure accountability and curb greenwashing. Policy should align fiscal incentives with local economic benefits, including workforce upskilling and protection of vulnerable livelihoods, and strengthen environmental safeguards through rigorous impact assessments, water and ecosystem protections, and innovation in cooling technologies. Over the next few years, India will require an integrated, cross-sectoral regulatory framework spanning energy, land, and environment. This should be complemented by stronger governance of AI hardware lifecycles through improved e-waste management, extended producer responsibility, and global cooperation. Continuous monitoring of community health impacts, together with a broader sociotechnical framing of AI safety that addresses risks along the entire AI supply chain, would position India to play a leading role in developing and advocating for new and improved global standards for sustainable AI infrastructure.

Research on the Socio-Ecological Impacts of AI Data Centres

Current scholarship at the intersection of AI, sustainability and policy studies has largely focused on “AI for sustainability”, even as concerns about the growing energy demands and environmental impacts associated with resource extraction for AI chips, the operation of AI data centres, training of large models, and the use of Generative AI applications have intensified. UNCTAD’s Digital Economy Report 2024 notes that data centres and ICT infrastructures account for 6-12 per cent of global energy use, among other resource demands, and calls upon policymakers, industry leaders and consumers to take urgent action to mitigate their adverse effects. 

This emerging body of work builds on an established tradition of scholarship examining digital media and the environment, and on recent work exploring the environmental impacts of AI. This includes research on the energy requirements for training large AI models, the resource extraction, labour practices, and e-waste associated with AI systems, and the supply chains of data. Scholars and commentators have highlighted the difficulties involved in accurately measuring the environmental costs of AI technologies, as well as the risks of greenwashing. While insights on the relevant use cases for harnessing AI for positive climate impacts remain limited, there is burgeoning evidence of the carbon footprint associated with AI systems, particularly in developing LLMs (large language models). Recognition of these implications is beginning to emerge within parts of the technology sector itself. Companies such as Hugging Face, for example, now track the carbon footprint of AI development and/or usage. These concerns are becoming more pressing as industries with significant environmental footprints – such as oil and fossil fuel companies, and defence manufacturers – are now deploying AI in large-scale operations. 

Already vulnerable communities are rendered even more marginalised in the ongoing geopolitical race for AI and data infrastructure supremacy. Rising temperatures in regions already experiencing climate stress, growing water scarcity in urban areas with receding water tables, and increasing energy demands are lived realities for communities on the ground. Yet these experiences remain largely disconnected from high-level technology debates and environmental policy conversations. Centring issues and questions of socio-ecological sustainability in the rapidly unfolding saga of AI and its infrastructures is thus critical to India’s medium- and long-term AI strategy. Drawing on this body of academic scholarship and informed by policy analysis, this essay identifies strategic priorities for the development of an equitable and sustainable AI ecosystem in India.

Policies for AI Data Centres: An International Overview

The international policy landscape for AI data centres has rapidly emerged as a critical site of contestation, situated at the complex intersection of industrial strategy, digital sovereignty, and resource regulation. Governments increasingly categorise large-scale compute infrastructure as strategically important but are caught between the instinct for aggressive expansion and the need to bear its socio-ecological costs, which their populations ultimately shoulder. Reports from intergovernmental and regional bodies suggest that the demand for concentrated compute power necessitates a fuller re-evaluation of cyber-risks, energy grid stressors, and the availability of water and other resources.

In the United States, policy has increasingly favoured acceleration, with a federal push aimed at minimising the bureaucratic barriers to the deployment of hyperscale facilities. However, this deregulation and incentivisation of infrastructure development has been met with pushback from local communities that bear the environmental costs. Protests in Virginia, mobilised using hashtags like #NotInMyBackyard, are a prominent example. Similarly, the United Kingdom’s efforts to embed AI in public sector organisations and develop “AI Growth Zones” also reflect a growth-first logic aimed at facilitating investment in what are considered nationally significant infrastructure. 

Conversely, the European Union offers a more conditional model, framing its expansion within a green industrial policy that links compute capacity to mandatory energy-efficiency reporting and sustainability standards. While not anti-growth, the EU’s approach emphasises a nascent form of digital sovereignty, seeking to reduce dependence on non-European cloud and chip ecosystems while monitoring the heavy water and energy footprints of these facilities. In China, this pursuit of sovereignty is even more starkly pronounced, with state-led strategies prioritising domestic chip integration and maintaining strict government control over the digital stack to reduce dependence on foreign suppliers.

These dynamics are also evident across the Global South and among middle powers, where the policy conversation is increasingly centred on building countries’ compute sovereignty. In Brazil and several countries in the Middle East, sovereign capital and energy advantages are being leveraged to host hyperscalers at scale, positioning these regions as new nodes in the global data hierarchy. In Southeast Asia, the ASEAN Trusted Data Corridors (TDCs) facilitate regional safe harbours for Big Tech. However, this hyper-growth frequently collides with the material realities of resource scarcity. For instance, Singapore’s stringent green mandates have triggered a spillover effect into Malaysia’s Johor region, where authorities recently halted non-AI projects to protect local water and power reserves. Across Africa, countries like South Africa and Kenya are navigating the dual challenge of seeking local infrastructure to reduce dependence on offshore clouds while contending with severe constraints relating to power reliability and capital costs. Despite their differences, these experiences point to a common policy dilemma. Governments are seeking to position themselves within the geopolitical race for AI capability and technological sovereignty while simultaneously managing the environmental, social, and resource costs associated with AI data infrastructures. The challenge lies in balancing these objectives in ways that ensure both long-term competitiveness and sustainable resource stewardship.

Current Policy Landscape for AI Data Centres in India 

India does not have a unified policy approach to data centres as yet. Instead, these infrastructures are governed through a combination of AI- and data centre-related policies at both the national and state levels. At the national level, the government introduced a draft National Data Centre policy in 2020, which was revived in 2025 amid the AI boom. The 2025 draft policy aims to grant infrastructure status to data centres and establish dedicated data centre economic zones. Importantly, the draft classifies data centres as an “essential service,” potentially bringing them under the ambit of the Essential Services Maintenance Act (ESMA). In addition, the Reserve Bank of India has enforced strict data localisation requirements under the DPDP Act of 2023 in the realm of fiscal data. These requirements have prompted the setting up of data centres domestically to comply with these clauses. Such measures are also in line with wider debates and aspirations to enshrine data sovereignty through the DPDP Act and the India AI Governance guidelines, which seek to create an enabling environment for a domestic AI economy while strengthening sovereign capacity.

Beyond data centre-related policies, the Ministry of Electronics and IT (MeitY) released India’s AI Governance Guidelines in late 2025. These guidelines encourage companies operating in India to voluntarily publish their socio-economic and environmental impacts, and transparency is stated as desired. However, these provisions remain non-binding. The IndiaAI Mission has also committed to 34,000 GPUs of public compute capacity, backed by production-linked incentive (PLI) schemes for IT hardware manufacturing and server production. In addition, MeitY is developing AI-specific standards for operations, maintenance, power consumption (which can be five times that of conventional needs), and water use (up to ten times higher), in consultation with Telecom Regulatory Authority of India (TRAI). 

Several states, including Maharashtra and Telangana, introduced dedicated data centre policies in the 2010s, but these have undergone revisions due to the growth of AI data centres. Several other state governments have adopted similar policies since 2021. As noted earlier, the Union Budget 2026-27 provides major tax incentives, including holidays until 2047, for foreign cloud providers using Indian data centres, fixed-profit safe harbours to avoid disputes, and lower taxes on electronics warehousing. At the state level, competitive federalism is evident in the range of incentives offered to attract data centre investments. These include capital and land subsidies, exemptions from registration duties, and special dedicated power grid supplies. Several state governments have actively courted Big Tech companies to bring their data infrastructure campuses to their cities and towns. Coastal states like West Bengal and Andhra Pradesh have even added undersea cable landings systems to the offer. Taken together, these policies reflect a concerted effort to position India as a global cloud and AI infrastructure hub. 

Contextualising India’s AI Data Centre Policies 

Even as India seeks to build an AI economy that caters to the country’s unique needs, use cases and linguistic diversity, a number of competing contextual factors constrain or divert India’s resources and impinge upon nascent capacities. Geopolitical contests and wars that wager on critical minerals acquisition and associated supply chains, constraints on acquiring AI chips that are essential for training models, and AI market concentration in the Global North all actively shape the choices India is able to make to have a stake in the global AI economy. For instance, moratoriums such as in Singapore or regulatory conditions like in the UK’s recent litigation from Foxglove have led companies to seek alternative locations for large-scale AI infrastructure in other countries. India is seen as a convenient destination given its policy incentives and relatively permissive regulatory frameworks. This, however, places India in a difficult position: the country must now build its sovereign capabilities while also being perceived as a stable business environment to acquire technology and compute infrastructure. Although India has emerged as a leader in Digital Public Infrastructure (DPI) among Global South countries, it is constrained by the unequal political economy of global AI. Renewed efforts must be made at international cooperation, collective action, and the pursuit of more equitable terms of participation in the global AI infrastructure ecosystem.

Gaps and Challenges 

While the preceding discussion outlines the many policies governing data centres in India, significant gaps remain in practice. 

(a) There remains limited recognition of AI’s sustainability implications in policy frameworks. As discussed earlier, GPUs and AI chips are energy-guzzlers. AI data centres generate vast amounts of heat and require large volumes of water to cool them, much of which is lost through evaporation, while the remaining amount of leftover water is contaminated and sent back into local water bodies. Further, the question of land acquisition for data centre campuses in India is complex, with land being tied to historical patterns of land ownership governed by caste and communitarian politics. Yet, there remains a clear policy gap in bringing the intersecting areas of energy, water, land as well as other local factors under one umbrella. India currently lacks a holistic policy ecology for AI data infrastructures in which policies for each of these impacted sectors are brought to bear on and speak to each other. 

(b) India’s pursuit of data and AI sovereignty has meant that companies are incentivised to store and localise the digital data of the Indian population within its territory. However, this turns into a double-edged sword, especially as India accounts for 20 per cent of global data generation but houses only 3 per cent of global data centre capacity. Missing from these figures is the question of whether the country’s resource base can sustain the growing demands of privately owned data centres while delivering meaningful public benefits. Existing inequalities are exacerbated when development jeopardises access to water, land holdings for housing or as an occupational source, and electricity, especially as distribution companies often function at or below capacity. India is faced with the unenviable task of balancing a domestic AI data economy and quest for sovereignty with attracting foreign capital and AI data centre investments, even while catering to local resource economies that are stretched and bursting at the seams. The wider geopolitical race to acquire or own parts of the AI supply chain only amplifies these challenges.

(c) India’s AI and data centre policies do not hold data centre providers accountable for the adverse environmental impacts of data infrastructures. As discussed earlier, policies see transparency disclosures of AI use and their corresponding resource use as desirable, but do not mandate this disclosure. This lack of transparency is concerning because of the enormous repercussions AI data centres have on local social ecologies, their health and biodiversity. By providing incentives to Big Tech companies with no accountability requirements, India’s current strategic approach to AI is marred by the lack of any corresponding binding commitments on these companies. Stronger commitments are needed, including meaningful livelihood opportunities that go beyond a few token low-paying jobs, the restoration of ecosystems, appropriate compensation, and adequate efforts in transparent reporting.

(d) Issues of land acquisition, resource allocation, and the distribution of economic as well as other resource benefits raise larger questions about procedural and distributive justice. In India, land is deeply intertwined with questions of livelihoods, social equity and justice. Land acquisition is an especially fraught issue, given that 50 per cent of the country’s workforce continues to be dependent on agriculture. The National Rehabilitation and Resettlement Policy, 2007 and the Right to Fair Compensation and Transparency in Land Acquisition, Rehabilitation and Resettlement (RFCTLARR) Act, 2013, both mandate social and environmental impact assessments, as well as rehabilitation measures alongside adequate compensation and benefits. However, recent reports suggest that legal due process has not always been followed in AI data centre projects in Telangana, Andhra Pradesh, and other states. Such practices further unlawful expropriation and have a ripple effect on livelihoods, migration, and urban poverty. Similarly, the Draft National Water Framework Bill 2016 proposes that industrial units use water only after meeting a “water for life” requirement. These industries must take efforts to reduce their water footprint using a combination of incentives, rules, and penalties. It also recommends that industries with high intensity of water usage not be located in drought-prone or water stress-prone regions, and encourages transparency regarding water usage in annual reports. However, these measures remain unimplemented in practice, with drought-prone areas hosting AI data centres, local water bodies seeing increasing contamination, and groundwater in these regions risking depletion. 

(e) It is also imperative that Environmental Impact Assessments (EIA) be closely analysed as a mechanism and process in India. In 2020, the Ministry of Environment, Forest and Climate Change (MoEFCC) published a new notification that permitted infrastructure projects to obtain environmental clearance after construction had started. This meant changes in the public hearing mandated to be held before the commencement of the project by the EIA process of 2006. The new draft of 2020 reduced timelines for public consultation and permitted hearings to be held after the start of the project. This dilution of the EIA as a process in enshrining public deliberation was met with widespread criticism but has also allowed for little to no scrutiny of large-scale AI data centre projects – as has been the case in Visakhapatnam’s eco-sensitive zone, where projects have proceeded with no environmental clearance.

(f) Another lacuna in India’s AI and data centre policies is the absence of a distinct regulatory identity for data centres. While the draft Data Centre Policies 2020 and 2025 classify data centres as “infrastructure”, they do not recognise them as specific kinds of infrastructure. AI data centres are a new kind of infrastructure, serving as more than data storage units or industrial units. The AI chips that they house place extraordinary demands on resources, which existing definitional remits of “infrastructure” do not capture. Some commentators liken AI centres to nuclear power plants in terms of the need to classify them as a separate category of critical infrastructure. Emerging proposals involving dedicated renewable-energy systems, green hydrogen, and small modular nuclear reactors (SMRs) for AI infrastructure further highlight the need for updated regulatory categories that reflect the distinctive nature of AI infrastructure.

(g) Finally, questions remain regarding the employment and livelihood opportunities generated by AI centres. Data centres typically require smaller workforces, concentrated primarily in security and maintenance operations. Further, these data centres are often tied to regional/global AI training and deployment needs and are not necessarily set up to cater to a domestic AI economy. The International Energy Agency’s review of India’s energy policy highlights the fact that coal and biofuels continue to be our primary source of energy, and projects a doubling and potential tripling of energy demands by 2040 to account for increased consumption, appliance ownership, and cooling needs. Against this backdrop, the costs of bringing in foreign AI data centres with long-term subsidies and the allocation of scarce resources to such projects should be carefully assessed against the wide-ranging socio-environmental benefits they would potentially offer. 

Recommendations

Immediate Measures

1. India should pursue the classification of AI data centres as a distinct category of critical infrastructure in national and/or international governance arenas. As AI ecosystems expand rapidly, such a classification would move beyond existing “infrastructure” definitions to recognise the unique resource demands and socio-ecological impacts of data centres on human ecosystems. 
2. India should move beyond voluntary disclosures and incentivise accountability and transparency as a larger strategic approach. Companies involved in developing data centres should submit mandatory disclosures on energy, water, and biodiversity impacts, verified by third-party auditors and public hearings. This would strengthen real-time accountability, enable benchmarking against global standards, and curb greenwashing in AI infrastructure expansion.
3. In line with the tax incentives introduced in the Union Budget 2026-27, India should incentivise data centre developers to contribute to continuous upskilling and to creating optimal opportunities for local economies to flourish, especially where generational livelihoods that are tied to landholdings and other environmental resources are under threat. 
4. Existing legal frameworks for Environmental Impact Assessments must be implemented, with additional safeguards put in place to ensure that water tables in populous areas are not depleted, eco-sensitive zones are not adversely impacted, and R&D investments are made to improve heat-control and cooling technologies. In addition, AI impact assessments should be integrated across the AI supply chain, introducing wider accountability and opportunities for judicial oversight. 

Medium-term Measures (2-5 Years)

5. India should develop cross-sectoral policies for governing AI data centres that bring energy, water, land, and environmental regulations under one umbrella. Such an integrated policy ecology would help overcome existing policy silos, ensuring that sectoral policies are compatible and account for each other rather than generating fragmented incentives that overlook socio-ecological costs.
6. AI chips are estimated to have a lifespan of one to three years, contributing to rapidly growing volumes of e-waste. Much of this waste is disposed of in countries of the Global South, often with significant consequences on the health and environmental safety of populations exposed to radiation and polluted soil and water. These countries should collectively advocate for and negotiate relatively sustainable and equitable practices for AI data centres, including extended producer responsibility, recycling targets, and partnerships with certified handlers. Such measures would counter the effects of mounting e-waste from AI hardware lifecycles, limiting toxic leaks into local ecologies and reducing landfill burdens on vulnerable communities.
7. The government should commission regular surveys and track the health of communities living in the vicinity of data centres to ensure that inadvertent impacts do not lead to pronounced long-term effects. These assessments should examine issues such as the effects of radiation, rise in temperatures, and the creation of heat islands, and the impact of contamination on humans, animals and the biodiversity of local ecologies.

Finally, AI safety should be understood as a sociotechnical challenge encompassing questions of socio-ecological safety across the AI supply chain. This includes critical mineral mining, chip production, and their deployment in data centres for AI development. India has the opportunity to play a leadership role in advancing end-to-end assurance frameworks, contributing to the development of global standards while operationalising them through national policy and regulatory institutions.

Acknowledgement: I would like to thank Laavanya R Nair for her research assistance in identifying the data centre policies of Indian states.