States chase AI gold rush as Visakhapatnam asks: at what cost?
Andhra Pradesh wants to build a 6.5-GW data-centre ecosystem around Visakhapatnam, while Telangana has just landed a 1-GW TCS AI campus. Experience overseas, however, shows that the second phase of the data-centre boom is invariably about water, power, public subsidies and who ultimately bears the cost.
The gathering organised by Green Visakha in Visakhapatnam on Sunday may prove to be about much more than a few proposed data centres. It marks the arrival in India of a debate that has already engulfed some of the world's biggest digital infrastructure hubs: how much public land, electricity and water should governments commit to artificial intelligence infrastructure, and what should communities receive in return?
Speakers at the Green Visakha meeting called for an independent expert committee and a cumulative assessment of projects totalling about 6,500 MW, covering land, water, electricity, environmental impact, government incentives, local employment and data security. Their position was significant: they said they were not opposing technology or investment, but wanted the costs and benefits placed in the public domain.
That distinction is increasingly being made around the world because the AI boom is turning data centres from relatively unobtrusive digital infrastructure into something resembling a new heavy industry.
From software race to compute race
For the past three decades, Indian states competed for IT companies by offering office space, talent and tax concessions. AI changes that equation. Training and running large AI models requires thousands of specialised chips operating continuously in high-density server clusters. The scarce commodities are therefore no longer merely engineers and office space but megawatts of reliable electricity, grid connections, cooling, land, fibre routes and, in coastal locations, submarine cable access.
The International Energy Agency expects global data-centre electricity consumption to roughly double from about 485 TWh in 2025 to around 950 TWh in 2030, with AI-focused data centres growing much faster than conventional facilities. Data centres could account for around three per cent of global electricity demand by the end of the decade.
India is consequently seeing an extraordinary infrastructure build-out. JLL estimates that the country's installed data-centre capacity, about 1.6 GW in mid-2026, could reach 6 GW by 2029, requiring an estimated $110 billion of investment. Hyperscalers themselves are expected to build nearly 30 per cent of the new capacity.
This is the economic prize Andhra Pradesh and Telangana are now chasing.
Andhra wants to leapfrog established hubs
Andhra Pradesh's ambitions have expanded spectacularly. Its Data Center Policy 4.0, notified in November 2024, originally spoke of adding 200 MW and attracting advanced AI-enabled facilities. Less than two years later, the government says it has achieved allocations for a proposed 6.5-GW ecosystem centred on Visakhapatnam.
The anchor is Google's approximately $15-billion AI hub, being developed with AdaniConneX and Airtel, with an initial 1-GW data-centre campus, clean-energy infrastructure and a new international submarine cable gateway. Google began construction in April. Reliance and other operators have added to the proposed pipeline.
Visakhapatnam offers something Hyderabad and Bengaluru cannot: a coastline capable of directly landing international submarine cables. For latency-sensitive cloud and AI services, that can be a major advantage. Andhra Pradesh is effectively attempting to use data centres and cables to recreate the transformational effect that IT services once had on Hyderabad.
But the scale itself creates the case for scrutiny.
IT minister Nara Lokesh told the Assembly last month that a fully developed 6.5-GW ecosystem was estimated to require around 3 TMC of water annually, while maintaining that it would come from the Polavaram project's industrial allocation and would not reduce supplies to residents. Three TMC works out to roughly 233 million litres a day averaged over a year.
That number is particularly sensitive because the government has told Reuters that Visakhapatnam currently receives around 410 million litres a day against demand of about 480 million litres — a present deficit of roughly 70 MLD. The state says water earmarked for residents or rural schemes will not be diverted to the data centres, while Google says its facility will use advanced air cooling to reduce reliance on water.
This does not establish that the projects will cause a water shortage. It does establish why Green Visakha's demand for a publicly auditable water balance is difficult to dismiss.
Telangana builds on an existing digital ecosystem
Telangana is pursuing the same AI economy from a very different starting point. Hyderabad already has a dense ecosystem of technology companies, global capability centres, cloud customers and skilled employees. Its strategy is to turn an established software and enterprise-computing hub into an AI-compute hub.
On Saturday, TCS subsidiary HyperVault and its partners announced plans to invest as much as ₹70,000 crore in a 1-GW AI data-centre campus in Telangana.
That comes on top of AWS's expansion. In January 2025, the Telangana government said Amazon would invest about ₹60,000 crore in additional data-centre infrastructure, and in July this year the government broke ground on an AWS facility at Bharat Future City. Microsoft, meanwhile, has been preparing its biggest Indian data centre in Hyderabad as part of its wider multibillion-dollar AI and cloud investment programme.
Telangana has also attracted a proposed 100-MW AI data centre from the UPC-Volt venture, coupled with a dedicated 100-MW renewable energy project.
The attraction for investors is obvious. Telangana's data-centre policy provides for uninterrupted power arrangements, renewable-energy access, land and fiscal incentives and infrastructure including water and fibre.
Yet gigawatt-scale AI centres are radically larger consumers than the facilities for which many older state policies were designed. Telangana therefore faces essentially the same questions as Andhra Pradesh: how much new generation and transmission will be needed; who will pay for it; what water will be used; and whether subsidies remain justified at hyperscale.
Every major state wants a share
The competition is national. Mumbai remains India's dominant data-centre market because of financial-sector demand, internet exchanges and submarine cables. CBRE estimated that Mumbai accounted for 53 per cent of India's roughly 1,530 MW capacity in September 2025, followed by Chennai with 20 per cent, Delhi-NCR with 10 per cent and Bengaluru with seven per cent.
Uttar Pradesh has gone further by rewriting its policy specifically for the AI era. Its 2026 Data Centre Policy targets more than ₹2 lakh crore of investment, offers an “AI Compute Booster”, land and electricity-duty concessions, and proposes that new facilities move towards 50 per cent renewable-energy use by 2030.
Maharashtra, Tamil Nadu, Karnataka and Odisha also have policy frameworks intended to attract data centres. What was once a contest involving perhaps four major metropolitan markets is increasingly a state-level industrial policy race.
But resistance is also spreading within India. Residents in Thane have raised objections to a proposed 422-MW Amazon hyperscale facility over electricity, water, backup diesel generators and the cumulative impact on an already crowded urban area.
Vizag is therefore not an isolated case.
The world has already reached this argument
Perhaps the most instructive experience is Singapore. Faced with land and electricity constraints, the city-state temporarily paused growth of new data centres in 2019. When it reopened the sector, it did not return to unrestricted approvals: applications were selected according to energy efficiency and decarbonisation criteria. Singapore's Green Data Centre Roadmap now seeks at least 300 MW of additional capacity, but ties expansion to better efficiency and green energy.
Ireland went further after data centres grew from five per cent of national electricity demand in 2015 to 22 per cent in 2024. Its regulator now requires new facilities seeking grid connections to provide generation or storage corresponding to their maximum import demand and to meet at least 80 per cent of annual consumption with additional renewable generation. Data-centre consumption could otherwise reach 31 per cent of Irish electricity demand by 2034.
In the European Union, operators are being pushed towards disclosure. EU rules require reporting of energy performance and water-footprint indicators, with the European Commission developing a rating system and minimum performance standards.
In Virginia, home to the world's largest concentration of data centres, regulators have created a separate electricity tariff for very large users and imposed long-term payment commitments so ordinary consumers are less likely to subsidise grid infrastructure built for hyperscalers. New large users can be required to pay at least 85 per cent of contracted transmission and distribution costs regardless of actual utilisation.
Water has become equally contentious. In Chile, an environmental court ordered the water implications of Google's proposed Santiago data centre to be reconsidered in light of climate change and prolonged drought. Google subsequently abandoned its original approval process and decided to redesign the project around air cooling.
And only this week, civil-society groups in South Africa sought a pause in additional data-centre approvals until their implications for water, electricity and land were better understood. The country's Human Rights Commission has received more than 250 submissions on the subject.
Vizag raises the right question
The emerging global position is therefore neither “stop AI” nor “approve everything”. It is increasingly build, but disclose; expand, but pay the infrastructure costs; use scarce resources only after demonstrating efficiency; and subject very large clusters to cumulative rather than project-by-project scrutiny.
That principle has particular relevance to Andhra Pradesh.
A 50-MW data centre and a proposed 6.5-GW regional cluster cannot sensibly be assessed in the same way. At that scale, questions about transmission corridors, additional generation, reservoir allocations, wastewater, diesel backup, biodiversity, land acquisition and heat and noise become regional-planning questions rather than merely individual project clearances.
Employment claims also deserve closer examination. Data centres generate substantial construction activity and can anchor cloud, AI, telecom and hardware ecosystems, but the facilities themselves are highly automated. International research finds that permanent operational employment is much smaller than construction employment. Even government estimates for the Google Vizag project have varied depending on whether direct, indirect and induced jobs are counted.
The economic case for Andhra Pradesh and Telangana remains powerful. Owning large-scale compute capacity could place them closer to the next generation of AI companies, cloud services, semiconductor demand, network infrastructure and digital exports. Missing the build-out entirely could also leave India dependent on computing infrastructure elsewhere.
But the experience from Singapore, Ireland, Virginia and Chile suggests that states gain little by treating environmental scrutiny as an obstacle to the AI economy. Transparent rules may instead be what gives the industry its social licence to operate.
Seen in that context, Green Visakha's intervention is not necessarily a revolt against the AI economy. It could be the beginning of India's next debate about it: not whether the country should build the computing infrastructure of the future, but how much it should build in one place, what resources it may consume, who should pay for those resources, and how local people share in the gains.
Comments