Every time you ask an AI chatbot a question, somewhere a server is working hard — and getting hot. Keeping that hardware cool, at the scale AI now demands, requires enormous amounts of electricity and, in many cases, water. Hyderabad and Visakhapatnam (Vizag) are both emerging as major hubs for exactly this kind of infrastructure — and the numbers involved are worth understanding before the next wave of approvals goes through.
The scale, in real numbers
In Hyderabad, currently operating data centres (Sify and STT) draw about 18.4 MW of IT power capacity. But announced and planned projects — including Yotta, STT Future City, and Sify’s expansion — would push that to roughly 400 MW. Add long-term potential like CtrlS Chandanvelly, and the city could be looking at over 600 MW of IT capacity. At that scale, a 1,000 MW cluster running continuously could consume an estimated ~12,264 million units of electricity a year — equivalent to about 78% of Telangana’s entire domestic (household) electricity consumption for 2024–25.
Vizag’s trajectory is similarly steep. Current projects (Sify Rushikonda, initial AdaniConneX) total around 100 MW. Announced projects — AdaniConneX’s full build-out, a Google-Adani AI hub, and Sify’s Paradesipalem facility — would take that to roughly 800 MW. Longer-term potential, including a proposed 1.5 GW Reliance facility near Bhogapuram, points toward a cluster exceeding 2,500 MW.
Where the water comes in
Data centres don’t just need electricity — nearly all of that electricity eventually becomes waste heat, and something has to carry that heat away. The most common method, evaporative cooling, uses water directly.
Using an illustrative water-use estimate, Hyderabad’s currently operating centres use around 161 million litres a year; planned expansion could push that past 3,500 million litres annually. In Vizag, current operations use an estimated 876 million litres a year, with planned projects pointing toward over 7,000 million litres annually.
To put one figure in perspective: at a 1,000 MW cluster scale, water-intensive cooling could require around 24 million litres a day — enough to meet the daily domestic water needs of roughly 1.6 lakh people. In Vizag, that same daily figure would represent about 6% of the city’s entire drinking-water supply from GVMC.
Both briefings are careful to note these are illustrative estimates based on standard efficiency assumptions, not verified actual consumption — which is exactly why both call for facilities to publicly disclose and independently verify their real water and power use.
What’s actually at risk locally
This isn’t an abstract concern. In Hyderabad, the briefing points to pressure on the Musi river catchment, groundwater recharge areas, lakes and wetlands, and protected zones including KBR National Park, Mrugavani National Park, and Mahavir Harina Vanasthali National Park.
In Vizag, the risks are even more specific: proposed sites sit close to the Mudasarlova Reservoir (a drinking-water source), the Kambalakonda Wildlife Sanctuary’s Eco-Sensitive Zone (home to leopards, pangolins, and over 100 bird species), and coastal ecosystems protected under Coastal Regulation Zone rules.
Neither city currently has a dedicated regulatory category for data centres under India’s 2006 EIA Notification — many large facilities are processed under a category that doesn’t require the conventional public hearing process, and neither briefing found evidence that cumulative impact (multiple facilities together, not just one at a time) is being properly assessed.
Not all cooling is equal
Both briefings lay out a clear hierarchy of cooling technologies, from most to least water-intensive: traditional evaporative cooling towers sit at one end (high water use, and explicitly flagged as something to avoid in water-stressed areas), while direct-to-chip liquid cooling and full immersion cooling sit at the other — near-zero or negligible water use, and specifically recommended as the standard for high-density AI workloads.
The core message from both documents: efficiency gains alone can’t offset unchecked growth in scale, and in water-stressed regions, water-efficient cooling should not be optional.
Jobs vs. resource cost
One consistent, striking figure across both briefings: a 100 MW facility is estimated to create only about 100–200 permanent on-site jobs — roughly 1 to 2 specialised jobs per megawatt of capacity — alongside a larger but temporary construction workforce. Both briefings frame this plainly as a question citizens are entitled to ask: do that many permanent jobs justify the land, grid infrastructure, water allocation, and public subsidies these projects typically receive?
What EJF is asking for
Both briefings converge on the same core demands: mandatory, comprehensive environmental review (not the lighter-touch process many projects currently use) for any data centre above 20 MW; a hard rule against using groundwater or drinking water for cooling; protection of ecologically sensitive catchments and sanctuaries; enforceable limits on heat, noise, and emissions; and a public, independently-verified dashboard reporting each facility’s actual water use, energy source, and compliance — not just projected or assumed figures.
The bottom line
AI’s growth is real, and so is the infrastructure it needs. The question both briefings raise isn’t whether Hyderabad and Vizag should host this growth — it’s whether that growth happens transparently, with water security and ecological limits treated as non-negotiable, rather than found out the hard way after approvals are already granted.
Read the full citizen briefings for detailed data, cooling technology comparisons, and complete demands: Visakhapatnam Data Centres & AI Infrastructure · Hyderabad’s Emerging Data Centre Cluster