Fractionova

← All perspectives

Data centres in India

India is building bigger data centres. The cooling chosen now will run for up to twenty years.

By Founder & Principal Prabha Shankar ·

India’s data-centre sector is changing in kind as well as in size. A September 2026 study for the Ministry of Environment, Forest and Climate Change counts about 1.5 GW of IT capacity across about 300 facilities in 2025, and puts the pipeline of announced projects at about 17 GW by 2032, in only around 350 facilities. Cooling can take 30–50% of a data centre’s energy, and the systems chosen now will run for 15–20 years.

That makes cooling, water and power long-term commitments. The World Bank adds a caution about how much capacity India needs. For the companies that supply, build, power and finance this capacity, the findings raise decisions about technology, sites, partners and timing. Those are the decisions to which Fractionova brings AI-native Opportunity Intelligence and principal-led judgement.

Start a Decision

This perspective draws on the Ozone Cell, Ministry of Environment, Forest and Climate Change, Study on Promoting Sustainable Cooling Technologies in Data Centres (September 2026), prepared with the Alliance for an Energy Efficient Economy, with the World Bank’s October 2026 India Development Update and ORF Middle East’s The Hormuz Inflection (October 2026). Facts and judgements are attributed to their authors; the interpretation is Fractionova’s. Fractionova is not affiliated with the Ministry of Environment, Forest and Climate Change, the Alliance for an Energy Efficient Economy, the World Bank or ORF Middle East.

India’s data-centre capacity

IT capacity. The lighter segment shows the study’s view that the 2032 pipeline could reach 20 GW.

The 2032 pipeline is more than ten times the capacity installed in 2025, while the number of facilities rises only from about 300 to about 350.

Ozone Cell, MoEF&CC, September 2026, pp. vii, 7–8

What the Ministry study finds

The study, commissioned by the Ministry’s Ozone Cell, assesses cooling technologies, refrigerants and resource efficiency in Indian data centres. Five findings stand out for business.

Fewer, far larger facilities

Capacity grew from 375 MW in 2020 to 1,500 MW in 2025 and is expected to reach 3,000 MW in 2026. Announced projects put the 2032 pipeline at about 17 GW, possibly 20 GW, while the number of facilities rises only to about 350. The average facility grows from about 10 MW to about 48 MW, built mostly by hyperscalers. The study reports that MeitY expects over US$200 billion of data-centre investment, with nearly US$70 billion of projects entering construction.

MoEF&CC, September 2026, pp. 7–8

A third to a half of the energy

Cooling can account for about 30–50% of a data centre’s energy use. Average reported Power Usage Effectiveness in India is around 1.5 to 1.6, and reporting is largely voluntary. Improving it from 1.60 to 1.25 could save about 37 TWh of electricity by 2032.

MoEF&CC, September 2026, pp. vii–viii, 53

Racks are outgrowing air

Rack densities that were 5–8 kW have reached 30–50 kW with AI and high-performance computing, and are expected to reach 150 kW. That is driving a shift from air cooling to liquid cooling, including direct-to-chip systems. The seven facilities the study assessed mostly used air and hybrid cooling, with relatively low IT capacity utilisation.

MoEF&CC, September 2026, pp. vii, 3

Water is the other constraint

By 2032, annual water use could range from near-negligible for air-cooled systems to as much as 331 billion litres for water-cooled systems, about a fifth of Delhi’s annual water supply. A single 100 MW facility using evaporative cooling can draw close to 2 million litres a day.

MoEF&CC, September 2026, pp. viii, 52

The refrigerant clock

Cooling systems last 15–20 years. Equipment chosen during this build-out will run through the steepest phase of India’s HFC phase-down under the Kigali Amendment, which begins in 2028. The study recommends considering low-GWP and non-ozone-depleting refrigerants ahead of it.

MoEF&CC, September 2026, pp. vii–viii

A note on the source: the study’s executive summary describes capacity as estimated to exceed 17 GW by 2032, while its body puts the pipeline of announced projects at approximately 17 GW. Some figures in the study are drawn from other sources.

Where the capacity sits, and the power it needs

Location, grid access and water now weigh as much as land and connectivity.

Two hubs hold most of it

More than 60% of India’s operational capacity is in Maharashtra and Tamil Nadu, reflecting the early lead of Mumbai and Chennai as connectivity hubs with submarine cable landing stations.

MoEF&CC, September 2026, p. 9

Bring your own power

Grid-connection timelines are long in Mumbai, Chennai, Hyderabad and Delhi NCR. The study sees captive and group-captive renewables, open-access power purchase agreements and battery microgrids as ways to deploy faster, and favours renewable and low-carbon sources.

MoEF&CC, September 2026, p. 22

Beyond the metros

The study suggests that Tier-2 and Tier-3 locations with land, renewable potential and grid capacity may be more resource-efficient than further expansion in metros where power and water are under pressure. It recommends choosing cooling on local energy and water availability rather than efficiency ratings alone.

MoEF&CC, September 2026, pp. viii, 55

Power from the corridor

ORF Middle East argues that cross-border electricity suits peak demand and specific industrial loads such as data centres, while baseload should be secured at home. It also proposes co-locating nuclear power, including small modular reactors, with data centres along the India–Middle East–Europe corridor. These are proposals, not programmes.

ORF Middle East, The Hormuz Inflection, October 2026, pp. 27, 50

The World Bank’s caution

The World Bank’s India Development Update places the build-out in a wider frame.

  • A small base. India accounted for about 1% of global installed data-centre capacity in 2025. It ranks second in the world for AI talent but 68th for infrastructure in the Tortoise Global AI Index.
  • Talent and data first. The Bank suggests that India could prioritise AI talent and training data rather than a large number of data centres or advanced chips, and concentrate frontier investment where it has an advantage.
  • Access as well as ownership. It recommends widening access to domestic and international cloud and computing services, rather than relying exclusively on large-scale domestic infrastructure.

The Bank’s view does not contradict the pipeline, which the Ministry study links to AI, cloud computing and data localisation. It does suggest that not every announced megawatt should be taken as demand. We read that as a reason to test which capacity is under construction and contracted before committing to it; that is our inference, not a finding of the reports.

What changes for a supplier or investor

A finding matters commercially when it is connected to a specific business. For a company that supplies cooling, power or services to Indian data centres, or invests in them, these are the questions Fractionova would put first, organised by the dimension of the opportunity they affect.

Technology & capability

Liquid, hybrid or air

Rising rack densities are pushing new capacity towards liquid and hybrid cooling. The study recommends choosing on local energy and water availability, not efficiency ratings alone.

The decision

Which cooling approach fits the sites and densities we are targeting, and can we prove its water and energy performance?

Customer demand

Fewer, larger buyers

New capacity is concentrated in fewer, much larger campuses, built mostly by hyperscalers. As reported by the study, MeitY expects over US$200 billion of investment, with nearly US$70 billion of projects in construction.

The decision

Which campuses are under construction, and which are still announcements?

Partners & counterparties

Developers, builders and technicians

Hyperscalers, colocation developers and their contractors set the specifications. The study calls for training technicians in low-GWP refrigerants and liquid cooling.

The decision

Which developers and contractors should we qualify, and who will install and service our systems?

Economics & commercial potential

Efficiency is worth money

Cooling takes 30–50% of a data centre’s energy, and better efficiency could save about 37 TWh by 2032. The study finds dedicated finance for cooling upgrades limited, particularly for colocation and mid-sized operators.

The decision

What is our offer worth to an operator over fifteen to twenty years, and who will finance it?

Policy & geostrategic conditions

Refrigerants and local manufacture

India’s HFC phase-down begins in 2028. The study recommends standard disclosure of energy and water efficiency and, in the medium term, support for domestic manufacture of key liquid-cooling components. These are recommendations, not schemes.

Local manufacturing as a strategic option

The decision

Will our refrigerants and products meet the rules ahead, and should we make key components in India?

Timing & market readiness

Pipeline versus need

The pipeline is large and growing, but the World Bank cautions against building a large number of data centres ahead of talent and data. Our perspective on the Gulf’s build-out looks at how buyers there now weigh speed and resilience.

Read Building Compute Through Conflict

The decision

What should we commit to now, and what should wait for contracted capacity?

What the reports do not settle

The reports are useful evidence and an incomplete basis for a supplier’s or investor’s decision. Four limits matter.

  • A pipeline is not demand. The 17 GW is the study’s reading of announced projects. It does not show which will be built, filled or paid for.
  • Field evidence is limited. The study assessed seven facilities in Mumbai, Pune and Delhi NCR. Practice elsewhere may differ.
  • Recommendations are not rules. Disclosure formats, water advisories and support for local manufacture are proposals for the coming years.
  • Different measures. The World Bank’s 1% share and the study’s capacity figures measure different things and are not combined here.

Those gaps are where a supplier’s or investor’s decision is made: in which campuses proceed, what they specify, where they are sited and who pays.

Decisions this page is written for

Bring the question in the form your board would ask it.

Which cooling approach should we take to Indian data-centre buyers?

Decision Sprint

Where to back the build-out?

This is a hypothetical illustration, not a client engagement or a claim of achieved outcomes.

A UAE-based investment firm wants exposure to India’s data-centre build-out. It has been offered a stake in a colocation developer, a minority position in a cooling and water-reuse specialist, and a share of a captive renewable power platform serving data-centre campuses. The investment committee needs to decide where its first commitment should go.

Fractionova’s AI engine would connect each option with the campuses it depends on, their construction and contract status, cooling and water requirements, power arrangements, regulation and the partners involved. Principal-led review would establish which route supports the first commitment and what evidence would justify it.

Explore the Data-Centre Cooling Assessment

A

Take a stake in a colocation developerBack the buildings and the customers in them. The questions are which campuses are contracted, at what utilisation and in which cities.

B

Back a cooling and water-reuse specialistBack the technology the new campuses must buy. The questions are which approach buyers will specify, and how fast the specialist can scale.

C

Join a captive power platformBack the power the campuses need to connect faster. The questions are grid rules, offtake terms and the share of renewable supply.

Three ways to commission the intelligence

You do not need to choose a format first. The starting point is the decision your business needs to make.

Decision Sprint

Concentrated intelligence for a consequential choice.

A focused investigation around a defined decision, culminating in a decision brief with the recommended direction, unresolved questions and conditions for commitment.

For example: Decide which cooling approach to take to Indian data-centre buyers, and on what terms.

Market & Partner Mission

A coordinated mandate to develop market access and commercial relationships.

An integrated investigation into market priorities, entry structures and counterparties, with prioritised partners and preparation for engagement.

For example: Qualify the developers, contractors and campuses building capacity in Mumbai, Chennai and beyond, and prepare engagement.

Fractionova Intelligence Office

A retained intelligence capability around your strategic agenda.

Recurring investigation and monitoring, supported by agreed business context and structured principal-led reviews.

For example: Track India’s data-centre pipeline, refrigerant rules and power arrangements as the build-out proceeds.

Principal-led, from Chennai and Dubai

Founder & Principal Prabha Shankar leads the technical-commercial judgement behind Fractionova’s work. An engineer and international business specialist, he brings nearly three decades of experience across automotive engineering management, advanced mobility, new energy, international market development and complex programme execution.

At Fractionova, that experience shapes the questions put to the AI engine, the assumptions challenged during investigation and the commercial significance assigned to its findings.

ExperienceIndia, Europe and the Gulf, with automotive companies, institutions and government stakeholders
MotorsportSecretary of the Event for three editions of the Formula One Indian Grand Prix
EducationElectrical & Electronics Engineering; postgraduate study at Stevens Institute of Technology

Questions this page answers

How fast is India’s data-centre capacity growing?

Fast. The Ministry study counts about 1.5 GW of IT capacity in 2025, expects about 3 GW in 2026, and puts the pipeline of announced projects at about 17 GW by 2032, possibly 20 GW.

Why does cooling matter so much?

Cooling can account for about 30–50% of a data centre’s energy use, and the systems chosen now will run for 15–20 years, through India’s HFC phase-down from 2028. Rising rack densities are also pushing new capacity from air to liquid cooling.

How much water could India’s data centres use?

It depends on the cooling. The study estimates that by 2032 annual water use could range from near-negligible for air-cooled systems to as much as 331 billion litres for water-cooled systems.

Is India building more data centres than it needs?

The reports do not say so. The World Bank notes that India held about 1% of global installed capacity in 2025, and suggests prioritising AI talent and data over a large number of data centres. Fractionova’s interpretation is that businesses should test which capacity is under construction and contracted before committing to it.

How does Fractionova help?

Fractionova brings AI-native Opportunity Intelligence and principal-led judgement to cross-border decisions: which buyers and campuses to pursue, which partners to engage, what to make locally and where to invest, for businesses working between India, the Gulf and other markets.

More perspectives

  • Data centres & intelligent infrastructure

    Building Compute Through Conflict

    MEI finds the Gulf kept building data centres after the March strikes. What resilience, speed and origin checks mean for suppliers to the AI build-out.

    Middle East Institute
    October 2026
  • Trade & market access

    India’s Way Around Hormuz

    India rerouted its oil within months of the Hormuz closure. ORF maps the slower routes, from power links to green ammonia, and what they mean for business.

    ORF Middle East
    October 2026
  • Trade & market access

    Trading Around Hormuz

    The WTO finds Gulf trade moving to ports outside the Strait of Hormuz. What rerouting, lasting costs and uneven demand mean for businesses selling into the Gulf.

    WTO
    October 2026

All perspectives

Bring the decision that deserves deeper intelligence.

A technology to position. A campus to qualify. A component to localise. An investment to test.

Start with what you are considering, what is at stake and what you need to determine.

Start a Decision

Sources and notes. The capacity, facility, investment, cooling, rack-density, efficiency, water, refrigerant, siting, power, skills and finance findings are from the Ozone Cell, Ministry of Environment, Forest and Climate Change, Study on Promoting Sustainable Cooling Technologies in Data Centres, prepared with the Alliance for an Energy Efficient Economy (September 2026), pages vii to viii, 3, 7 to 9, 22 and 52 to 56. The investment figures are the Ministry of Electronics and Information Technology’s, as reported in the study, and the capacity estimates draw on AEEE research. India’s share of global capacity, the AI readiness rankings and the policy priorities are from the World Bank, India Development Update (October 2026), pages 36 and 45. The cross-border and nuclear power proposals are from ORF Middle East, The Hormuz Inflection (October 2026), pages 27 and 50. Figures from different publishers measure different things and are not combined. Fractionova is not affiliated with the Ministry of Environment, Forest and Climate Change, the Alliance for an Energy Efficient Economy, the World Bank or ORF Middle East.

Fractionova equips clients with intelligence and considered recommendations to inform their decisions. Investment and implementation decisions remain with the client; execution and outcomes depend on the organisations involved and prevailing conditions.

Back to top