Power, Water and Fibre: A Location Framework for India’s Data-Centre Growth

India’s data-centre capacity increased from about 375 megawatts in 2020 to around 1,500 megawatts in 2025. The sector’s electricity demand is estimated to reach 13.56 gigawatts by 2031–32, according to a March 2026 statement by the Ministry of Electronics and Information Technology. Industry estimates cited also suggest that cloud data-centre capacity could increase four to five times by 2030.

Data-centre site selection requires evidence at the asset level because infrastructure constraints can outweigh headline cost advantages. A lower land price is attractive but provides limited value when the delivery date for the required grid connection remains uncertain. Network resilience depends on physical route diversity, so two contracts from two different network providers, can offer limited protection if both links share the same trench. Annual efficiency metrics also require seasonal context because they tend to hide higher water consumption during the dry season. City-level averages can support an initial market assessment, but the final site selection decision should reflect the specific substation, water source, fibre routes and land costs.

This article sets out a framework for assessing potential data-centre locations. The first stage examines whether each site can secure sufficient power, water and fibre infrastructure within the required delivery period. Sites that satisfy these requirements are then evaluated across land availability, natural hazards, permitting, construction capacity, workforce access, customer proximity, total cost and schedule. The relative importance of these factors varies by operating model. Network-rich colocation facilities place greater emphasis on connectivity and customer access. Regional cloud zones require reliable infrastructure across a defined service area. Artificial-intelligence training facilities place greater emphasis on power availability and heat removal. Disaster-recovery facilities require geographic and infrastructure separation from the primary site.

Growth changes the level of analysis

At several gigawatts, development affects power, water, fibre, land and disaster planning. The International Energy Agency estimates that global data-centre electricity use will rise from 415 terawatt-hours in 2024 to about 945 terawatt-hours in 2030, and that grid risks could delay about 20 per cent of planned projects. In advanced economies, new transmission lines can take four to eight years, while transformer and cable lead times have doubled (International Energy Agency, Energy and Artificial Intelligence, 2025).

India met a peak electricity demand of 256.1 gigawatts on 25 April 2026, according to the Ministry of Power, and had 548.9 gigawatts of installed generation capacity on 30 June 2026, according to the Central Electricity Authority. These national totals convey an impressive trajectory but do not guarantee capacity at a particular substation. A large, continuous load may require additional substation bays, transformers, transmission lines and protection studies.

The same distinction applies to water and fibre. The review of a specific site must cover both the plot and nearby facilities.

India’s current map reflects networks and customers

Official reporting from TRAI lists 790 megawatts of datacenter capacity in Mumbai and Navi Mumbai, 305 in Chennai, 182 in Bengaluru, 152 in Hyderabad, 76 in the Delhi National Capital Region and Noida, and 15 in Pune and Kolkata combined. Mumbai and Chennai account for about 72 per cent of the total.

Mumbai and Chennai combine customers, submarine cables, network companies and experienced operators. The Telecom Regulatory Authority of India reported 17 cables at 15 stations in five cities in December 2021, concentrated most heavily in these two markets. Shorter distances to landing stations can reduce international delay (Telecom Regulatory Authority of India recommendations).

In March 2026, the government listed four systems being commissioned in Mumbai, one also in Chennai, and three further systems planned for Chennai, Raigad and Mumbai. Inland diversity still depends on backhaul, landing stations, nodes and local access paths.

Concentration supports customers and interconnection, but it also increases shared use of land, utilities and transport corridors. Projects may therefore move reflexively towards a metropolitan edge while retaining acceptable latency and access. Colliers projects that capacity across India’s seven largest markets could exceed 4,500 megawatts by 2030, from 1,263 megawatts in April 2025, while floor space could rise from 16 million to about 55 million square feet (Colliers, May 2025).

The assessment begins with the workload

Four broad workload types describe much of the market, although one campus may combine them.

Network-rich colocation serves several customers and depends on network companies, cloud connections, internet exchanges and nearby enterprises. Network delay and the cost of extending several networks limit how far it can move from a cluster.

Regional cloud and content infrastructure serves a defined market, often across separate availability zones. It needs predictable delay, large power connections and more than one route between facilities. It can use a metropolitan edge when the network still meets those requirements.

Artificial-intelligence training, high-performance computing and batch processing give greater weight to large power connections, dense equipment and heat removal. They need substantial fibre, but not always as many local providers as retail colocation. Interactive inference may need to be closer to users than training.

Resilience, backup and recovery facilities require geographic separation, independent utilities, separate fibre paths and a tested recovery plan. Distance reduces common exposure only when the secondary site does not share the same floodplain, transmission corridor or cable bottleneck.

The project brief should record rack density, load growth, network delay, traffic, service and recovery targets, operating life and phased capacity. This profile links buildings and utilities to actual use.

Power: connection size, timing and reliability

Power is a basic condition because on-site backup cannot replace a large grid connection over campus life. A power infra review should address five questions.

How much power can the site import? The maximum at the site boundary exceeds the electricity available to computers because cooling, conversion and other systems also consume energy. Power usage effectiveness divides total facility energy by computing-equipment energy, as defined in ISO/IEC 30134-2:2026 and explained by the United States Department of Energy. A 100-megawatt connection supports about 87 megawatts of computing load at a ratio of 1.15, 80 at 1.25, 74 at 1.35 and 69 at 1.45. These are arithmetic examples; actual performance changes with use, weather, cooling and measurement boundaries.

What is the connection status? The review should identify the substation, voltage, capacity, backup, constraints, construction, cost and dates.

How will it be built? Should cover transformers, switchgear, cables, land permissions, manufacturing lead times and commissioning. Staged grid supply can be matched to staged installation of computing equipment.

What happens during interruption? Separate feeds, batteries and generators form one system. Two feeds may share a substation bus or corridor. The review should cover component and regional failures, backup duration, fuel, emissions and planned outages.

Water: cooling design and local sources

Water use depends on computing load, climate, cooling, operating temperatures, treatment and the measurement boundary. Drinking water, treated wastewater, surface water and groundwater have different reliability, quality, permissions and infrastructure.

Water usage effectiveness measures on-site water use against computing energy. ISO/IEC 30134-9:2022 defines the measure and reporting categories. Annual averages hide monthly peaks or dependence on drinking water, so reporting should separate sources and show monthly and peak demand, discharge, quality and cooling mode.

Cooling choices shift demand between electricity and water. Evaporative systems use water and can reduce compressor power. Dry systems use less on-site water but may require more electricity in hot weather. Direct-to-chip and immersion systems remove heat from dense equipment through liquid, but the building must still reject that heat through a tower, dry cooler or hybrid system. The United States Department of Energy reports that direct liquid cooling can improve both electricity and water performance in suitable designs. Singapore’s tropical standard estimates cooling-energy savings of 2 to 5 per cent for each 1-degree-Celsius increase in operating temperature when equipment and practices allow it (Infocomm Media Development Authority).

The Central Ground Water Board classified 4,946 of India’s 6,762 groundwater assessment units as safe in 2025; 758 were semi-critical, 201 critical, 730 over-exploited and 127 saline (Central Ground Water Board).

A site review should map the watershed, groundwater unit, municipal zone and wastewater network; examine monthly supply, drought restrictions, permits, treatment and the final pipe; and test normal, dry-year and interruption cases. Global tools can screen locations, but local transfers, quality and infrastructure can change conditions (World Resources Institute, 2025). Treated wastewater may reduce drinking-water demand. Rainwater depends on storage, groundwater on yield, and seawater on intake, corrosion, discharge and environmental review. A monthly balance should show the electricity or computing effect of a lower-water mode.

Fibre: capacity, latency and physical separation

Network capacity measures the volume of data that a connection can carry. Network latency measures the time required for data to travel between two points.

It is common practice to secure two or more fibre connections from multiple telecoms/internet service providers, but actual physical route separation determines whether those connections remain lit when a cable, conduit or supporting structure fails.

Assessing physical separation requires a complete review of each connection from the server racks to the wider network. Each route should be mapped through network rooms, property exits, ducts, intermediate facilities and long-distance corridors. Two links may leave a building through separate points before converging in the same trench. Different network providers may also lease capacity on the same underlying cable. International connections can face a similar concentration risk when multiple landing routes cross the same bridge or transport corridor. United Kingdom government guidance therefore recommends that customers verify the physical separation of their telecommunications routes. (Cabinet Office guidance).

India’s lit submarine-cable capacity rose from 39,282 gigabits per second in 2016 to 138,606 in 2022. Activated capacity rose from 9,137 to 111,111 gigabits per second, or from 23 to 80 per cent of lit capacity (Telecom Regulatory Authority of India, 2023).

A project record can include route maps, ownership, shared structures, construction exposure, maintenance, latency, common failures and repair periods. Routes need review because leases, road works and layouts change. Inland and recovery sites remain viable when capacity, delay and separate paths are compatible with the intended workload.

Other factors: hazards, land, permits and workforce

Sites that meet the three basic conditions should be assessed for other factors. Natural hazards may affect several systems together. The India Meteorological Department classifies 100 coastal and near-coastal districts by cyclone hazard, including 12 as very highly prone and 25 as highly prone, many along the Bay of Bengal (IMD classification). Coastal and inland reviews should cover flood, wind, heat, drought and air quality, including their effects on drainage, structures, fuel and fibre.

Usable land may be reduced by flood buffers, setbacks, corridors, roads and utility equipment. The review should include title, land use, soil and heavy-equipment access. Because off-site infrastructure crosses other land, rights of way, environmental conditions, fire approvals and utility construction belong in one schedule. Workforce planning should cover operating skills, contractors and suppliers.

The roles of India’s principal markets

Mumbai and Navi Mumbai support network-rich colocation, finance, enterprise systems, cloud access and international traffic. Further growth depends on grid dates, flood and drainage design, land, water and separate fibre paths. Metropolitan-edge sites can retain access when latency remains suitable.

Chennai combines an established base, cable landings, industry and technical workers. It can support international connectivity, cloud and large campuses, subject to seasonal water, flood and cyclone conditions and independent routes to landing stations and network nodes.

The Delhi National Capital Region and Noida serve northern enterprise, public-sector and consumer demand. Regional cloud and colocation benefit from proximity to users, while site reviews need long-distance fibre diversity, summer cooling, water, air-quality compliance and firm power dates.

Bengaluru supports cloud, enterprise, edge and artificial-intelligence services through its technology market. Campus options depend on power, water and uncongested fibre routes; edge locations trade more utility space against network and staff access.

Hyderabad combines a technology market with peripheral campus land. It can serve regional cloud and power-intensive computing if long-distance fibre, local water and supplier capacity develop with committed projects.

Pune can serve western enterprise demand and provide separation from Mumbai for recovery, cloud expansion and selected computing campuses. Its recovery value depends on avoiding the same regional power and fibre failures.

Kolkata, Visakhapatnam and other eastern coastal markets can extend reach and add routes as networks enter service. Reviews should cover power, land, hazards, water, provider choice and the status of announced infrastructure.

Other regional cities may host edge, government, enterprise, recovery or power-intensive workloads. As completed capacity and demand increase, network nodes, contractors and service teams may locate nearby. This is a commercial sequence rather than a location requirement.

Evidence from markets outside India

Foreign markets show recurring patterns, although their systems differ from India’s.

Constraints can move growth to nearby markets. London inventory is expanding near existing cloud zones, including north of the city, while West London faces grid delays. In Northern Virginia, power constraints and limited large sites are pushing consideration towards the Interstate 95 corridor and adjacent power markets (CBRE Global Data Center Trends 2026).

A queue position is not an operating supply. Ireland assesses new connections against location, network constraints and generation adequacy. Britain’s 2026 reform debate followed a rise in applications that included projects unlikely to proceed soon (Ofgem, July 2026). Evidence about land, customers, demand, construction and grid work distinguishes project stages.

Established hubs can develop alongside neighbours. Singapore paused approvals in 2019 and later used a competitive process. Its 2024 roadmap began with more than 1.4 gigawatts and set out at least 300 megawatts of near-term additional capacity with efficiency measures (Singapore Infocomm Media Development Authority). Johor grew nearby, linked by networks, customers and operations.

Measurement becomes more detailed with scale. The European Union requires qualifying data centres to report energy and water indicators, separating total and drinking water and defining electricity, renewable-energy and heat-reuse measures (European Commission).

Across these cases, demand first concentrates near customers and networks. Shared demand can lengthen infrastructure delivery, after which growth may move to connected locations with capacity. Larger projects require more site-level evidence.

A six-stage location assessment

1. Define the service requirements. Record workload, customers, traffic destinations, availability design, equipment density, facility and computing power, growth, cooling, data movement and recovery targets. Separate required conditions from preferences.

2. Check the basic site conditions. Exclude plots with unresolved ownership, incompatible use or unaddressed hazards. Test whether power, water and fibre each have a workable delivery path.

3. Map the supporting infrastructure. For the remaining sites, map the grid connection and upstream work, water balance and sources, and fibre paths to relevant nodes. Test normal conditions, stress and specific failures; identify who controls each asset and when it will operate.

4. Compare sites for the workload. Network-rich colocation gives greater weight to fibre and customers. Artificial-intelligence training gives greater weight to firm power and heat removal. Recovery gives greater weight to separation from the primary site’s hazards and infrastructure. The weights are stated assumptions.

5. Compare the cost of a working system. Include land and electricity as well as substations, fibre construction, water treatment, cooling, flood protection, schedule uncertainty, tax, staff and financing during delay. Compare the cost per unit of computing capacity that is connected, commissioned and operating.

6. Link later phases to updated evidence. A campus can open in stages. Later buildings proceed when demand, utility construction, actual water performance and fibre separation remain within the stated conditions. Final campus plans should remain distinct from first-phase capacity.

The decision record should classify each project or asset as operational, contracted, under construction or conceptual. It should preserve each figure’s unit, date and boundary, show ranges where needed, name the evidence and leave unverified statements as assumptions.

Measurement after the facility opens

Operating data test the assumptions used in selection. Power records should cover imports, computing energy, power usage effectiveness, outages, batteries, generators, fuel, power quality and later connection phases. Water records should separate sources and cover water usage effectiveness, monthly and peak demand, treatment, discharge, dry cooling, quality and interruptions. Fibre records should cover traffic, latency, packet loss, route changes, shared maintenance, outages, repairs and backup tests.

Other records include floods, cooling limits, equipment lead times, staff, safety, commissioning defects and permits. Portfolio measures can include confirmed grid dates, operating versus announced capacity, water by source, checked fibre routes, time to first power and shared exposure.

What the framework indicates for India

India can develop a portfolio of locations serving different workloads. Established coastal markets retain customer, network and cable advantages. Their edges can add space without losing those ecosystems. Inland technology markets can support cloud and larger computing campuses when power, cooling and long-distance fibre are ready. Eastern, coastal and regional markets can add reach and separation as operating infrastructure and demand develop.

The framework begins with evidence for the grid connection, water system and physical fibre paths. Suitable sites are then compared by delivered cost, schedule, hazards, workforce and customers. It distinguishes announced projects from operating assets and uses operating measurements before later phases proceed. This approach addresses both concentration within established clusters and premature movement to locations whose supporting infrastructure is not yet available.

Source notes

1. Government of India, Ministry of Electronics and Information Technology, data-centre capacity and estimated electricity demand for 2031–32, 13 March 2026: View source

2. Government of India, operating capacity by city and submarine-cable pipeline, 18 March 2026: View source

3. International Energy Agency, Energy and Artificial Intelligence, executive summary, 2025: View source

4. International Energy Agency, energy supply for artificial intelligence, 2025: View source

5. Colliers India, The digital backbone: Data center growth prospects in India, 28 May 2025: View source

6. CBRE India, India’s Data Centre Market in a New Era, 12 November 2025: View source

7. CBRE India, India Alternate Sectors Outlook 2026, 18 March 2026: View source

8. CBRE, Global Data Center Trends 2026, 17 June 2026: View source

9. Telecom Regulatory Authority of India, recommendations on data centres, content delivery networks and internet exchanges, 18 November 2022: View source

10. Telecom Regulatory Authority of India, recommendations on the licensing framework and regulatory mechanism for submarine-cable landing in India, 19 June 2023: View source

11. United States Department of Energy, cooling-water efficiency opportunities for federal data centres: View source

12. International Organization for Standardization and International Electrotechnical Commission, standard 30134-2:2026 on power usage effectiveness: View source

13. International Organization for Standardization and International Electrotechnical Commission, standard 30134-9:2022 on water usage effectiveness: View source

14. Central Ground Water Board, National Compilation on Dynamic Ground Water Resources of India, 2025: View source

15. World Resources Institute, grounding global water-risk assessments in local data, 2025: View source

16. United Kingdom Cabinet Office, guidance on telecommunications resilience: View source

17. Ireland Commission for Regulation of Utilities, data-centre electricity connection policy, 12 December 2025: View source

18. Singapore Infocomm Media Development Authority, Green Data Centre Roadmap: View source

19. Singapore Infocomm Media Development Authority, Tropical Data Centre Standard, updated 23 September 2025: View source

20. European Commission, energy performance and reporting of data centres: View source

21. India Meteorological Department, cyclone-hazard proneness of coastal and near-coastal districts: View source

22. Ofgem, proposed data-centre connection reforms and evidence on the demand queue, 29 July 2026: View source

23. Government of India, Ministry of Power, all-India peak electricity demand met on 25 April 2026: View source

24. Central Electricity Authority, installed generation capacity as at 30 June 2026: View source