India’s Data Centre Boom and Expansion (2026–2030): Capacity, Cities, AI, Power and What Comes Next

This guide explains India’s data-centre expansion, how to interpret market forecasts, where capacity is concentrated, what is driving demand, what could constrain growth and how to assess a project without mistaking an announcement for a live facility.

Contents

1. What counts as a data centre?
2. How large is India’s data-centre market?
3. Why is the market expanding?
4. Where are India’s data centres and new projects?
5. What Tier II expansion requires beyond a land offer
6. What does the buildout include?
7. The bottlenecks that will decide the pace
8. What the boom could mean for companies and the economy
9. How to assess a project, claim or market forecast
10. Read the market through workload, capacity and utilisation
11. From a market announcement to a commissioned hall
12. Why digital growth does not translate one-to-one into data-centre MW
13. How operators balance speed, resilience and efficiency
14. A practical market scorecard for readers
15. Three planning scenarios for the 2026–2030 expansion
16. What public announcements reveal—and what they leave open
17. Frequently asked questions
18. Sources and further reading

What counts as a data centre?

A data centre is a facility that houses computing, storage and networking equipment, together with the electrical, cooling, security, fire-safety and monitoring systems needed to operate that equipment. Facilities range from enterprise server rooms and smaller edge sites to large colocation campuses and hyperscale buildings. The type of facility affects what it is designed to do, who owns the infrastructure and how much power it needs.

Enterprise data centres

An enterprise facility is owned or operated for one organisation’s own workloads. Banks, government departments, manufacturers, hospitals, retailers and technology companies may run private infrastructure to meet performance, control, security or legacy requirements. Some organisations operate their own equipment in a shared colocation facility rather than owning a building.

Colocation data centres

A colocation provider develops and operates a facility where customers rent space, power and related services. Customers may lease a rack, a private suite or a larger block of capacity and generally install or control at least some of their own IT equipment. The operator provides the building and critical infrastructure; exact responsibility depends on the contract.

Hyperscale and cloud facilities

Hyperscalers and cloud providers build or lease large facilities to support cloud services, AI, storage, databases and other digital workloads. A cloud customer consumes a service rather than necessarily managing the physical server. The underlying facility may be owned by the cloud company, a partner or a landlord that leases capacity back to it.

Edge facilities

Edge facilities bring computing closer to users, devices, industrial sites or networks when latency, resilience or local processing matters. An edge site may be a small telecom facility, an enterprise micro data centre or a compact colocation site. Edge does not automatically mean a large new commercial building; some edge capacity sits inside existing infrastructure.

Why these categories overlap

A data-centre campus can serve multiple customer types. A hyperscaler may use leased colocation while building its own campus. An operator may provide cloud interconnection and manage customer equipment. An enterprise may use both a private facility and public cloud. A project tracker should therefore describe the facility, owner, operator, customer and service model separately instead of treating “data centre” as one uniform product.

How large is India’s data-centre market?

There is no single market number that captures all of India’s operating capacity, planned projects and future demand. The numbers differ because research firms use different dates, sources, definitions and forecasts. Some report installed IT capacity; others use facility capacity, pipeline, or a long-term market forecast. The right response is not to average them. It is to show each estimate with its original date and definition.

Selected 2025–2026 capacity estimates

These figures are intentionally presented together because their differences are material. Savills’ 6 GW 2030 forecast and Wood Mackenzie’s 12 GW forecast are not the same estimate, and neither should be converted into a guaranteed outcome. Cushman & Wakefield’s pipeline is not comparable to an operational stock figure. CBRE’s 9M 2025 snapshot predates the 2026 reports and may use a different scope.

Capacity terms matter

• Operational IT capacity: The power available to IT equipment in a live facility, according to the source’s methodology.
• Facility or utility power: The electrical draw or infrastructure capacity serving the whole site, which includes non-IT loads such as cooling and power losses.
• Under construction: A project phase with construction activity underway; it is not yet customer-ready capacity.
• Planned pipeline: Projects that may be proposed, announced, permitted or in development, depending on the source’s definition.
• Announced campus capacity: A target or long-term programme number; it may cover multiple phases and years.
• Policy capacity target: A government objective to attract or mobilise development, not installed capacity.

A 1 GW campus announcement should not be added to operational stock as though it were live. A source that reports “capacity” without a clear metric should be quoted with its own term and caveat. Comparing IT load to total power demand without a conversion basis can also mislead.

Why market forecasts vary

Forecasts depend on assumptions about cloud and AI demand, project execution, power availability, utilisation, construction lead times, efficiency, future leases, financing and which facilities are included. A forecast may assume that planned campuses are delivered; another may include only projects with a stronger level of commitment. A market forecast is a modelled future, not a list of confirmed projects.

When comparing a forecast, check:

1. The baseline year and installed capacity used.
2. Whether the number measures IT load, facility load or another metric.
3. Whether announced, planned and under-construction projects are included.
4. Whether the forecast is for India as a whole or a set of tracked markets.
5. What assumptions are made about AI, cloud, power and project completion.
6. The publication date and any later revisions.

A credible market article should preserve the disagreement and tell the reader why it exists. It should not choose the largest number merely because it sounds more exciting.

Why is the market expanding?

India’s demand is not produced by a single technology or one customer group. A facility can serve cloud platforms, enterprise applications, online services, payment systems, content, AI workloads, government systems and digital infrastructure. Growth in those services creates more demand for compute, storage, power and connectivity, although not every digital transaction requires a new data centre in India.

1. Cloud adoption

Public cloud lets a company access computing and storage without building every server room itself. As businesses move applications, analytics, backup, development environments and customer services to cloud platforms, providers need infrastructure close to the workloads and users they serve. A cloud region is a collection of physical and network resources that supports cloud availability zones and services; it is not necessarily one building or one capacity figure.

Cloud growth also changes the mix of infrastructure. Customers may use managed databases, software services, container platforms and AI tools rather than renting raw compute. As a result, cloud revenue, data-centre MW and server counts do not move in a simple one-to-one relationship. A cloud company can grow revenue through utilisation and software efficiency even if capacity grows differently.

2. Digital services and online activity

Payments, e-commerce, video, messaging, search, identity, mapping, financial services and workplace tools all depend on digital infrastructure. India’s large online population and expanding digital services contribute to demand for reliable storage, compute and network capacity. The effect is distributed: some workloads are served from local facilities, some from other regions, and some are cached or delivered through content networks.

A rise in users does not automatically equal a proportional rise in data-centre power. Software optimisation, server efficiency, workload placement, caching, compression and utilisation can change the energy needed per transaction. A useful market analysis considers both demand growth and efficiency improvements.

3. Artificial intelligence and accelerated computing

AI training and inference can require dense clusters of accelerators, high-speed networks, specialised cooling and substantial power. Generative AI has raised interest in large-scale compute, model training and serving, enterprise adoption and AI-focused campuses. Google’s Visakhapatnam hub is one example of a public AI-infrastructure plan; Meta and Reliance announced a 168 MW initial Jamnagar facility to be leased by Meta; Digital Connexion announced a 1 GW Visakhapatnam campus plan under an MoU.

These announcements are not proof that all AI demand is contracted, all capacity is funded or every site is operational. AI workloads can be uncertain: model efficiency may improve; hardware generations change; inference demand may grow in different locations; and a customer may optimise or shift workloads. Developers need credible customer signals and phased design, not only broad AI enthusiasm.

4. Enterprise modernisation

Large organisations may migrate legacy systems, build disaster-recovery environments, consolidate data centres, or use hybrid models that combine private infrastructure and cloud. Regulated or latency-sensitive workloads may require specific controls or local services. Some organisations choose colocation to avoid building and operating their own power-redundant sites.

The timing varies by sector. A bank, manufacturer, hospital, media company and software startup have different security, performance, compliance and procurement needs. A new colocation hall may support several customer migrations, but the operator’s lease-up pace depends on actual customer demand and service fit.

5. Content, gaming and network delivery

Video, gaming, social platforms and software downloads need storage, caching and network delivery close to users. Content delivery networks can reduce latency and international traffic by placing caches in multiple markets. These systems may use a mix of data centres, carrier hotels, telecom facilities and edge locations.

The growth of online media can therefore support interconnection and colocation demand, but some capacity may be delivered by existing facilities or distributed infrastructure rather than a new hyperscale campus. The market map should identify the form of infrastructure, not assume every content signal means a new large building.

6. Data resilience and business continuity

Organisations need backup, disaster recovery and resilient operations. A business may distribute workloads across regions or maintain secondary systems that can take over during a disruption. This supports demand for geographic diversity and interconnection. However, a second facility may be used primarily for resilience and not run at the same utilisation as the primary site.

A resilience design should account for the whole system—network routes, power grids, flood and climate risks, staff access and operational procedures. Two buildings in the same vulnerable region may not provide true geographic resilience if they share a network route or utility dependency.

7. AI policy and local infrastructure ambition

Government policy and corporate strategies can encourage local investment, but they should not be confused with confirmed customer demand. National and state initiatives may promote data-centre parks, cloud services, AI adoption, semiconductor ecosystems and clean-energy infrastructure. Their effects depend on implementation, eligibility, grid capacity and the projects that actually reach construction.

A public target can still matter: it may guide planning, investment promotion and infrastructure coordination. But an analyst should distinguish the target from approved projects and operating capacity. See India’s public data-centre investment map for a framework to separate announcements, financing and policy support.

Where are India’s data centres and new projects?

India’s established data-centre markets are commonly associated with Mumbai, Chennai, Delhi NCR, Bengaluru, Hyderabad and Pune, while new large-scale announcements have also drawn attention to Visakhapatnam and Jamnagar. The exact market ranking and capacity in each location vary by source and date. City lists are a useful starting point, not a guarantee that every location has available power or a live campus.

Mumbai and Maharashtra

Mumbai has a mature connectivity and colocation ecosystem and is a key financial-services market. Maharashtra is also the focus of AWS’s public cloud-infrastructure plan for the state. AWS says the US$8.3B Maharashtra plan sits within the previously announced US$12.7B India cloud-infrastructure plan, so the state and national amounts should not be added without an overlap note. A project-level view should identify the actual site, phase, utility connection and customer.

Chennai

Chennai’s network connectivity and regional role make it an important market for enterprise, cloud and content infrastructure. A site assessment should consider fibre diversity, power supply, land, water, climate resilience and customer latency requirements. Market presence does not eliminate the need to verify a new project’s approvals or capacity.

Hyderabad

Hyderabad combines technology-sector demand with a growing cloud ecosystem. Microsoft has announced the operational status of its India South Central cloud region. That status is separate from its broader India investment pledge and does not publicly specify all facility-level capacity or capex. Vendors should map procurement to the facility owner or package, not infer a tender from a cloud-region launch alone.

Bengaluru and Pune

Bengaluru and Pune have technology companies, enterprise users, engineering talent and growing digital infrastructure needs. A market report may track facilities and pipelines in both locations, but a city’s technology-sector growth is not itself data-centre capacity. Investors should verify the parcel, power node, operator, tenant and pipeline stage.

Delhi NCR and Noida

Delhi NCR serves government, enterprise and consumer demand. Noida and Greater Noida are in Uttar Pradesh, whose 2026 data-centre policy sets investment and ecosystem ambitions with eligibility conditions. Projects within the wider NCR should be attributed to the right state and policy jurisdiction. Delhi, Haryana and Uttar Pradesh incentives should not be treated as one scheme.

Visakhapatnam

Visakhapatnam has attracted major AI-infrastructure plans. Google announced approximately US$15B over 2026–2030 for a broader AI hub and reported groundbreaking in April 2026. Digital Connexion, a JV including Brookfield, Reliance and Digital Realty, announced a US$11B, 1 GW campus plan by 2030 under an MoU. These are distinct programmes with different partners and public status. Neither should be treated as the same project or as fully operational capacity.

Jamnagar and Gujarat

Meta and Reliance announced a 168 MW initial facility in Jamnagar that Meta will lease, with an option to scale. Gujarat’s Data Centre Policy 2026–29 sets a state-wide target of 7.5 GW and an investment ambition of ₹6 lakh crore. The project-specific lease announcement and state policy target belong in separate records; neither proves that the policy amount has been spent or the facility is live.

How to choose a location

Data-centre site selection is a multi-variable problem. A city may have a strong customer base but constrained power; another may offer land and state support but need more fibre or a longer construction supply chain. A site can look attractive in a press release and still face a long utility connection process.

What Tier II expansion requires beyond a land offer

Interest in a secondary city often starts with a land parcel, a state incentive or a headline about lower-cost power. Those are inputs, not proof that a sustainable data-centre cluster exists. A new location needs a complete ecosystem: a customer case, feasible grid connection, resilient fibre, development approvals, water and climate planning, contractors, equipment support and a workforce. If one layer is missing, the project may need years of enabling investment before it can compete with an established market.

Customer demand should lead the location thesis

A developer should identify the workloads that benefit from the location. Possible reasons include lower latency to a regional user base, a customer’s enterprise footprint, a second disaster-recovery region, access to a particular network route, or a partner’s existing infrastructure. “Land is available” does not answer who will lease capacity or why customers will choose the site over a Mumbai, Chennai, Hyderabad or NCR facility.

Demand can be distributed across more than one customer. An anchor tenant may make an initial phase financeable, while enterprise, government, content and cloud workloads create additional use cases. The project should not assume that a large state target translates into contracted demand. It should test customer interest, required latency, data movement costs, service expectations and the timing of workload migration.

Ecosystem depth changes the operating model

An emerging market may have fewer specialist contractors, experienced data-centre operators, carriers, equipment technicians and commissioning engineers. Some capabilities can be brought in from another city, but travel time, spares, emergency response and staff retention affect operating cost and reliability. A project may have to invest in training, local suppliers, transport and service depots alongside the campus.

Connectivity needs special attention. A city with a cable landing station or a new fibre announcement still needs diverse paths from the actual site, multiple carrier choices, route protection and interconnection arrangements. A single route or carrier dependency can weaken resilience even if the regional network is expanding. Operators should test route diversity at the facility boundary, not rely on a city-level label.

A staged entry can reduce market risk

A developer may begin with a smaller, customer-backed phase and retain land or utility options for later expansion. The first phase can prove construction, service and power delivery; expansion then depends on usage and additional customer contracts. This can limit idle capital, though it does not remove the risk of an undersized site or delayed grid upgrades. Any phased approach needs an integrated campus design, so later growth does not require disruptive reconstruction.

A fictional example illustrates the logic: a regional operator finds a parcel near an industrial customer and a proposed transmission upgrade. Instead of describing the ultimate site capacity as imminent, it validates the customer’s load, confirms the connection schedule, identifies fibre routes and tests the operating-service model. If the first phase is viable, it advances; if power timing or customer demand remains uncertain, it keeps the later phase as an option rather than a forecast.

Separate proposal counts from delivery evidence

A state may receive proposals or expressions of interest that add up to several gigawatts. Those proposals are useful signals of investor attention, but they can overlap, change size, or fail to secure land, power and customers. A market map should track the proposal date and status, then update the record when a project obtains approvals, financing, a connection milestone, construction and commissioning.

For policymakers, progress measures should include approved and energised capacity, timelines, power delivery, water use, project jobs and local supplier participation—not only the total MW of proposals. For investors and operators, the question is not whether a state has many announcements but whether its infrastructure and commercial ecosystem can deliver a specific project on schedule.

What does the buildout include?

A data-centre boom is not simply a row of buildings. It includes real estate, electricity, renewable generation, transmission, cooling, fibre, servers, software, security, water management, operations and talent. Different companies may own each layer, and the timing of each investment can differ.

Land, zoning and campus planning

A campus needs land with a suitable shape, access, environmental and planning approvals, and room for substations, buildings, roads, cooling infrastructure and future phases. A developer may secure land years before the entire campus is built. The public capacity target can refer to the ultimate site plan rather than the first financed phase.

Large campuses need a phased master plan. Each phase should match expected demand and utility access. If every shell is built before customer commitments or grid upgrades are available, capital can sit idle. If capacity is built too slowly, the operator may fail to meet customer delivery dates. Developers balance construction lead times, financing and market demand.

Grid connection and electrical systems

The facility depends on the grid or other energy sources, as well as substations, transformers, switchgear, backup generators, uninterruptible power systems and controls. Grid upgrades may be outside the data-centre company’s direct control. The project’s announced timeline must account for utility approvals, equipment lead times and commissioning.

A site’s available MW may be different from its IT load because the facility also uses energy for cooling, lighting, security and electrical losses. Power usage effectiveness (PUE) is one way of describing facility efficiency, but PUE varies by design, climate, utilisation and measurement. It is not the only sustainability metric and should not be used without a clear reporting boundary.

Cooling and water systems

Servers produce heat that must be managed. Cooling approaches include air-based systems, chilled water, evaporative methods, direct-to-chip liquid cooling and other designs. The choice depends on climate, rack density, water access, energy costs, reliability and workload. AI clusters can have higher rack densities and may need different cooling design than conventional enterprise workloads.

Water use is a local issue. A water-cooled system can have high efficiency but require careful sourcing and management. An air-cooled or liquid-cooled design may shift rather than remove resource impacts. Project teams need to report water source, withdrawal and consumption clearly, distinguish potable from non-potable water, and consider local watershed stress. A national claim about water is not a substitute for site-specific assessment.

Renewable power and storage

Operators are looking at renewable procurement, open-access supply, captive or group-captive projects, storage and grid balancing. A renewable-energy contract must match the facility’s reliability and operating profile. Annual energy certificates or matching do not necessarily mean every hour’s power is physically delivered from renewable sources. Reporting should explain the contract, matching period, transmission constraints and residual grid use.

A company may announce a clean-energy portfolio that serves multiple sites. Do not assign the full renewable project capacity to one data centre unless the source establishes that allocation. Also distinguish a renewable target from a signed power purchase agreement or an operational power asset.

Fibre, subsea connectivity and interconnection

Data centres need diverse network links to connect customers, cloud platforms, content networks, carriers and other campuses. Fibre route diversity can improve resilience; interconnection helps workloads exchange data without relying solely on public internet routes. Subsea cable systems connect India to global markets, while terrestrial and metro fibre link cities and facilities.

A facility’s location should be evaluated for latency, carrier availability and route diversity. A nearby cable landing station or metro fibre route can be helpful but does not prove the facility has the right contracts or connections. A market report’s connectivity advantage should be translated into a site-specific route map and verified with carriers.

Servers, accelerators and network equipment

A building can be complete before all IT equipment is installed. Servers and accelerators have their own procurement, delivery, power and cooling requirements. AI infrastructure may require high-speed networking, specialised racks and different power distribution. Hardware generations can change rapidly, making the timing of an equipment purchase important.

This is one reason a project’s building capacity and live compute capacity differ. The operator might commission a hall in stages as customer equipment is deployed. Market reports that track data-centre space do not necessarily track accelerator capacity or compute availability.

Operations, maintenance and security

Once live, a facility requires skilled operators around the clock. Teams monitor power, cooling, access, fire systems, network connections and equipment. Operators test redundancy, maintain generators, manage incidents, update security controls and plan service windows. The operational model affects availability and customer trust.

Security includes physical access, monitoring, visitor management, asset protection and coordination with customers’ own cybersecurity practices. A facility’s certifications or audit reports should be checked for scope and date. No single certificate means every service or customer workload is automatically compliant.

The bottlenecks that will decide the pace

India’s data-centre opportunity is large, but capacity cannot be built by investment announcements alone. Several physical and commercial constraints determine whether a planned facility becomes useful, customer-ready capacity.

Power availability and grid readiness

Power is one of the most important constraints because a facility must secure a reliable supply at the right site and time. A grid connection can involve land acquisition, approvals, substations, transmission equipment and coordination with utilities. A city’s aggregate generation is not the same as firm capacity available at a particular project node.

Power cost matters as much as availability. Tariffs, demand charges, grid fees, open-access rules, backup fuel, renewable procurement and currency exposure can influence operating economics. A developer must assess both the physical connection and the long-term cost structure. “Green power available” is not a complete procurement plan unless the volume, profile, contracts and delivery route are clear.

Water, heat and local environmental conditions

Data centres can be energy-intensive and may use water for cooling, depending on design. Heat waves can increase cooling demand, while drought and water stress can limit local supply. Flooding, cyclones, earthquakes and other climate hazards also affect site risk. The project must assess conditions specific to its location and design.

Water stewardship is not only a technical issue. Communities and local industry also need water. Reuse, recycling, non-potable sources, cooling design and transparent reporting can help reduce pressure, but the plan needs independent validation and monitoring. A press release claiming sustainability should be assessed against defined measures.

Land, permits and construction capacity

Large projects need suitable parcels, approvals, roads, utilities, construction labour and materials. Complex projects can be delayed by land, zoning, environmental approvals, transformer availability, cooling equipment lead times or commissioning. A delay in one system can hold up the entire phase. Planning schedules should include contingencies and dependency tracking.

Construction quality matters because data centres are mission-critical facilities. Commissioning has to test electrical redundancy, cooling, fire protection, monitoring, security and controls. Accelerating work without adequate testing can create operational risk. A project’s groundbreaking date is not its customer-ready date.

Financing and customer demand

A project may require large upfront capital before recurring revenue begins. Lenders and investors need confidence in customer demand, power access, schedule, operator capability and exit options. If demand is overestimated or AI adoption changes, a campus can face under-utilisation. If the project is under-sized, it may not meet customer requirements.

Pre-leases can help validate demand, but terms matter: capacity, start date, price, power pass-through, expansion rights, service levels, termination and customer credit. A press release naming a customer does not reveal every commercial risk. An investor should review actual contracts and financial models, not only public claims.

Equipment supply and skills

Transformers, switchgear, generators, chillers, pumps, batteries, servers and networking equipment may have long lead times. Projects need local installation, integration and maintenance capacity. A shortage of trained engineers can affect commissioning and operations. Companies may need to build apprenticeships and supplier ecosystems alongside new facilities.

A strong local vendor base can improve resilience and reduce delivery risk, but localisation should not be treated as a substitute for quality and certification. Vendors need clear specifications, testing, warranties, service response and cybersecurity controls.

Policy coordination

States may introduce incentives and investment targets, while energy, water, land and network infrastructure require coordination across agencies. Policy certainty and transparent application processes can help projects plan. But incentives cannot by themselves create electricity, water or demand. Projects should model the base economics independently and confirm eligibility before including a policy benefit.

Uttar Pradesh’s 2026 policy targets more than ₹2 lakh crore in investment mobilisation. Gujarat’s 2026–29 policy sets a 7.5 GW capacity target and ₹6 lakh crore investment ambition. These are policy aspirations, not proof that every named or proposed project is funded. An investment map should keep policy targets in a separate column from project capex and operational stock.

What the boom could mean for companies and the economy

Infrastructure investment and construction activity

Data-centre development can generate demand for land, engineering, construction, electrical and mechanical systems, cooling, security, fibre and commissioning services. These opportunities can benefit a broad ecosystem, including Indian contractors and manufacturers that meet technical and service requirements. The value of supplier opportunities is not necessarily disclosed in a headline corporate pledge.

A project can also strengthen regional digital infrastructure and attract customers, but benefits depend on deployment. The campus must reach operation, maintain service quality and connect to customers and networks. A planned campus does not create the same economic activity as an operating facility.

Jobs and skills

Construction and operations require different skills. Project design, electrical engineering, HVAC, network operations, cybersecurity, facilities management, procurement and site operations all have roles. A data centre can be capital-intensive and may not create as many ongoing jobs per unit of investment as some other industries. Claims about jobs should specify whether they are construction, permanent, direct, indirect, temporary or economy-wide estimates.

Skills programmes should be matched to actual employer demand and qualifications. Training technicians without local jobs or clear career pathways can disappoint students. Operators, equipment suppliers, colleges and policymakers can cooperate on apprenticeships, certifications and practical training.

Digital resilience and cloud access

Local data-centre capacity can support cloud and enterprise services closer to customers. It can improve latency, availability options and data-processing choices. Resilience depends on diverse sites, networks, utilities and tested failover—not on having a data centre in the same city alone. An operator must design for failure and test recovery.

Energy demand and climate goals

The sector’s electricity demand creates both a challenge and a chance to invest in clean power, storage, grid upgrades and efficiency. Data-centre developers may become significant anchor customers for renewable-energy projects and advanced grids. At the same time, renewable matching and power availability need careful verification; a headline sustainability claim does not automatically prove low hourly emissions or limited water impact.

Local economic development

A campus may support tax revenues, construction activity and service providers. It may also compete for land, water and power with local users. Policymakers should evaluate the public benefits alongside the infrastructure costs and environmental impacts. Transparent reporting of power use, water, jobs, incentives and project status can help communities assess the trade-offs.

How to assess a project, claim or market forecast

A project-status checklist

Before repeating a project announcement, ask:

• Who announced it, and is there a primary source?
• Is the entity an owner, developer, operator, tenant, investor or government partner?
• Is the amount a commitment, capex budget, loan, policy target or estimate?
• Does it apply to one project, multiple sites or all company businesses?
• Is the facility proposed, at MoU stage, financed, under construction, commissioned or live?
• What capacity metric is used, and which phase does it describe?
• What facts remain undisclosed, such as costs, customer contracts or operational MW?
• Does the project overlap with a broader commitment or a state/national announcement?

A capacity-forecast checklist

A forecast should be read with its date, methodology and definition. Check the base capacity, forecast horizon, operational/pipeline split, inclusion rules and assumptions. If two reputable forecasts disagree, report both with attribution rather than selecting a convenient one. The forecast is useful as a scenario for planning; it is not proof that a particular project will be built.

A site-investment checklist

A site-specific diligence package should cover:

1. Land ownership and permitted use.
2. Firm power availability and connection schedule.
3. Power tariff, redundancy and renewable options.
4. Water availability, reuse and local environmental impact.
5. Fibre route diversity and carrier interconnection.
6. Climate and disaster exposure.
7. Customer demand and lease conditions.
8. Project finance, cost overrun and currency exposure.
9. Construction partners, equipment delivery and commissioning.
10. State policy eligibility, incentives and application status.
11. Operating staffing, maintenance and security.
12. Expansion plan and dependencies for later phases.

A procurement and vendor checklist

A supplier looking at the boom should identify the relevant buyer and stage rather than selling into a market headline. The project owner may use an EPC, consultant, system integrator or approved vendor list. Procurement can be packaged by phase. A developer may award some packages well before groundbreaking or after detailed design. A vendor should verify the tender, technical requirements, budget, qualification and payment terms.

Public project data can help prioritise accounts, but it cannot confirm a tender date unless a source says so. The safest way to communicate an estimated buying window is to label it as an estimate and explain the basis. A vendor should not commit inventory or staff to a project based only on a speculative schedule.

Read the market through workload, capacity and utilisation

Capacity is a supply-side measure; it does not tell the whole story about demand or the value of a facility. A data centre may have high installed capacity but low utilisation, or a smaller site with high occupancy and strong customer value. Conversely, a new AI hall may be heavily pre-committed but still depend on equipment delivery before it can provide compute.

Workload types have different infrastructure needs

The table is a broad guide, not a specification. Actual workloads vary, and a single company may run several types. The operator needs customer discovery and engineering before deciding what to build.

Utilisation is different from capacity

Capacity may refer to a facility’s technical ability to deliver power to IT equipment. Utilisation describes how much of that capacity is being used. Occupancy can refer to rented space or contracted capacity; it may not equal consumed power or active compute. A facility can be leased before the customer fills every rack, and a cloud region can serve workloads with variable demand over the day.

When market sources report MW, readers should ask whether the number is designed, energised, commissioned, contracted, occupied or consumed. If these statuses are not disclosed, do not assume. Likewise, revenue per MW depends on service mix, lease structure, density, power pass-through, utilisation and contract terms. A public capacity number is not enough to infer revenue or valuation.

AI creates density and flexibility questions

AI training can require large clusters to operate together, so network topology, power delivery, cooling and physical layout all matter. Inference has a different profile: models may serve many smaller requests distributed across locations, and model efficiency or demand patterns can change over time. A new campus may be marketed as AI-ready, but the buyer must verify rack density, cooling capability, network fabric and electrical design.

A flexible facility may accommodate a range of IT densities or allow phases to be upgraded as equipment changes. Flexibility has a cost: more robust systems or spare capacity can raise initial capital expense. The right design depends on a credible customer and workload view, not the assumption that every AI product will use a full gigawatt campus.

From a market announcement to a commissioned hall

The public lifecycle of a data-centre project usually contains several milestones, and each tells a different story. A headline announcement can occur before detailed engineering; a lease may be signed before construction; groundbreaking can precede installation of critical equipment; a completed building may still need commissioning and customer fit-out.

A practical stage map might look like this:

1. Strategic plan: The company identifies a market, customer segment or national investment programme.
2. Site or partnership announcement: A location, joint venture or MoU is made public; legal, land and power details may remain incomplete.
3. Feasibility and approvals: The parties work through site control, permits, grid connection and commercial design.
4. Customer and financing validation: A lease, customer commitment or financing is secured, subject to contract terms and conditions.
5. Design and procurement: Technical drawings and equipment packages are finalised; long-lead orders begin.
6. Groundbreaking and construction: Site work and building construction proceed; the project may continue to depend on utility equipment and systems integration.
7. Mechanical and electrical completion: The major systems are installed, but the facility may not yet be accepted for customer service.
8. Commissioning: Integrated tests verify redundancy, failure modes, controls, security and operating procedures.
9. Customer fit-out and energisation: IT equipment is installed and brought online in phases.
10. Commercial service: The operator confirms that a particular phase is available or in operation.

A project tracker should not compress these ten milestones into “announced” and “live.” For example, Google's reported groundbreaking is evidence that construction started on the broader AI-hub programme; it is not evidence that all gigawatt-scale capacity is already commissioned. The first live phase may come earlier than later campus phases, and the source should identify which one.

Why digital growth does not translate one-to-one into data-centre MW

Forecasting often begins with an appealing chain: more people online, more data, more AI, therefore more data-centre power. The relationship is real but not linear. Several forces can increase or reduce capacity needs per unit of digital activity.

Efficiency and software optimisation

New chips can deliver more computation per watt, while software and model optimisation may reduce the resources needed for a particular task. Operators can improve server utilisation, consolidate workloads, automate cooling and tune power systems. These efficiency gains can support more services from less incremental capacity than a simple extrapolation suggests.

Workload location and architecture

A workload may run in a public cloud, private cloud, enterprise facility, edge site or another country. The choice depends on latency, cost, regulation, customer contracts, data movement, skills and availability. A company may use several locations and shift workloads over time. Local demand does not mean every byte or AI process must be hosted in a new domestic campus.

Utilisation, peak demand and redundancy

A facility is designed to handle peaks and failures, not only average use. Resilience requires spare capacity or redundant systems. Cloud workloads can be moved, autoscaled or scheduled, so the load profile may change by time of day and season. Peak power needs can exceed average consumption. A forecast based only on annual data growth misses the design margin and resilience requirement.

Hardware refresh and changing density

Servers and accelerators are replaced over time. A data hall can host more compute after a refresh, but the upgrade may be limited by power and cooling systems. A campus may expand capacity in phases, or replace equipment rather than construct a new building. A reliable forecast should model both new space and improved compute per unit of existing infrastructure.

Demand uncertainty and investment timing

Projects have long lead times, while digital demand can shift quickly. Developers must decide whether to build ahead of customer demand or wait for commitments. Building early can capture a market but increases financing and utilisation risk. Waiting can reduce risk but lose a customer to another location. Phasing, flexible design, anchor tenants and credible power plans help manage this timing problem.

These factors explain why a data-centre forecast can differ from a general forecast of internet traffic or AI users. A forecast should connect workload assumptions to facilities and power rather than extrapolate demand from a single headline metric.

How operators balance speed, resilience and efficiency

A data-centre operator has to make design choices that are not captured by a single capacity number. More redundancy can improve availability but add equipment, space and energy overhead. High-density AI racks may require liquid cooling and different service procedures. Renewable power can reduce emissions but may require storage, grid balancing or flexible operations. A facility built for one specific workload may become less flexible if customer needs change.

Redundancy and availability

Power and cooling systems are designed with backup and maintenance in mind. Redundancy may be built at component, system or site level. The design goal depends on customer requirements and contract commitments. A marketing claim such as “highly resilient” should be backed by system architecture, maintenance procedures and independently tested performance. A second facility only improves disaster recovery if it has sufficiently independent power, connectivity and operational dependencies.

Efficiency metrics and boundaries

PUE is a familiar metric comparing total facility energy with IT energy, but its value changes by climate, utilisation and measurement method. PUE does not measure all environmental impacts. Water usage effectiveness, carbon intensity, renewable matching and equipment reuse provide additional views, each with its own measurement boundary. Comparing one operator’s annual average to another’s design target may be misleading.

A useful sustainability report should state the period, site scope, accounting method, data coverage, energy source and whether the figure is measured or estimated. It should also disclose trade-offs. A cooling design with lower water consumption may use more electricity; a renewable supply claim may reflect annual matching rather than hourly clean energy. The site and local grid conditions matter.

Modular phases and operational readiness

Phased construction allows an operator to align capital with demand, but each phase must integrate with the campus-wide power, water, network and security systems. Modular construction can accelerate delivery in some cases, but does not eliminate the need for site preparation, utility connection, equipment testing and commissioning. The project schedule should identify critical-path equipment and utility milestones, not just the building completion date.

A site can be technically complete yet not commercially ready if contracts, security reviews, carrier links or customer equipment are not in place. Therefore, a market update should differentiate construction completion, commissioning, customer availability and live utilisation whenever sources provide the information.

A practical market scorecard for readers

A reader comparing cities, companies or project announcements can use a scorecard without pretending to know confidential information. Rate each dimension as verified, partly verified, unverified or not applicable, and attach source links. Avoid turning the score into a precise investment rating unless the method is transparent and tested.

This scorecard is a screening aid, not due diligence. Investors and customers need access to contracts, engineering studies, utility records, permits and site data before making decisions.

Three planning scenarios for the 2026–2030 expansion

Scenarios are more useful than a single confident prediction when several infrastructure dependencies remain uncertain. The following are qualitative frameworks, not forecasts or estimates of capacity.

Scenario 1: demand grows, but delivery is paced by power

Cloud and AI customers continue to expand, but the critical path is site-specific utility connection, transmission equipment and approvals. Operators phase construction more carefully, prioritise locations with credible power access and reserve later phases until grid milestones are visible. In this scenario, announced capacity can remain much larger than commissioned capacity for a period. The practical indicators to watch are substations, utility agreements, energisation and customer availability—not only land acquisition or MoUs.

Scenario 2: AI demand accelerates and specialised capacity becomes scarce

AI adoption expands across cloud providers, enterprises and research organisations. Customers compete for high-density, liquid-cooled and well-connected space. Operators retrofit existing halls where feasible and develop new AI-ready phases. The key risks are accelerator availability, high-density power delivery, workload uncertainty and pricing. A headline AI campus might be significant, but useful capacity depends on the first deliverable phase, grid supply and customer utilisation.

Scenario 3: efficiency and staged investment moderate the buildout

More efficient hardware and software increase compute output per unit of power, while customers optimise or shift workloads. Operators still build, but projects are phased against contracted demand. Capacity grows without every initial campus target being reached on its first announced timeline. In this scenario, higher utilisation and improved efficiency can coexist with slower physical expansion.

These scenarios can overlap across regions. One city may face a power bottleneck while another sees strong AI demand, and a single operator may expand while another pauses. A responsible market view tracks project-level signals and explains which scenario assumptions are being used.

What public announcements reveal—and what they leave open

Public plans are valuable because they reveal strategic priorities, likely locations, potential partners and the direction of capital. They help suppliers, investors and policymakers decide what to investigate. But the announcement usually does not disclose a full construction budget, land title, customer contract, project debt terms, utility schedule, water plan, operational status or the amount spent to date.

Consider the difference among a company commitment, a project MoU, a lease agreement and a state policy. A commitment can span several business units and years. An MoU can support a campus plan but leave financing or approvals unresolved. A lease may disclose capacity and tenant but not price or completion status. A policy can create eligibility rules but does not guarantee that a project receives incentives. Each is meaningful evidence, but for a different question.

The best public-source analysis uses an evidence ladder: primary announcement; official policy or filing; independent reporting; market research; then clearly labelled inference. When details are private, say “not publicly disclosed.” Do not fill gaps by assuming an amount from MW, an owner from a project manager, or a delivery date from a promotional target.

A good article also updates the story when evidence changes. If a planned project reaches construction, change the status but preserve the earlier announcement date. If a company revises a target, record the new source and note the prior target was superseded. This creates a useful public history rather than silently rewriting the project record.

Frequently asked questions

Is India’s data-centre boom real?

There are public capacity reports, operating facilities, cloud expansions, new project announcements and large infrastructure commitments. The expansion is real, but not every announced project is financed or live. The pace and scale depend on power, construction, customer demand, capital and execution.

How much data-centre capacity does India have in 2026?

There is no single figure that all research firms report using the same definition. Savills and Cushman & Wakefield each report about 1.8 GW operational capacity in their H1 2026 updates, while other sources use different baselines and scopes. Always state the source, date and whether it refers to IT load or another measure.

What is India’s data-centre capacity forecast for 2030?

Forecasts differ. Savills’ H1 2026 report forecasts 6 GW by 2030. Wood Mackenzie’s July 2026 forecast is 12 GW, using its stated 2.2 GW 2025 baseline. These are separate models; neither is a guaranteed outcome and they should not be averaged without methodological work.

Is the 3.9 GW pipeline operational?

No. Cushman & Wakefield’s H1 2026 report describes nearly 3.9 GW in a pipeline of under-construction and planned projects. It is not operational capacity and includes project stages that may not all reach completion.

Where are most Indian data centres located?

Industry reports commonly track Mumbai, Chennai, Delhi NCR, Bengaluru, Hyderabad and Pune among established markets. New public announcements include major plans in Visakhapatnam and Jamnagar. Market rankings vary by source and date; city-level project status and available power need separate verification.

Why are companies building in India?

They are responding to cloud, AI, enterprise, digital-service, content and resilience needs, as well as long-term market growth. Some companies are also building local cloud or AI infrastructure. An individual project’s rationale can differ, and announced investment does not prove how much capacity customers will use.

Does every cloud region mean a new data centre?

A cloud region depends on physical data-centre infrastructure, but a region may comprise multiple facilities and does not necessarily correspond to one publicly identified campus. Cloud-region status is not the same as published MW or an open procurement opportunity.

Are the new AI campuses already live?

No, not based on the cited announcements. Google reported groundbreaking for its Visakhapatnam AI hub in April 2026, while Digital Connexion’s 1 GW campus is an announced MoU-backed plan. Meta and Reliance announced a 168 MW initial Jamnagar facility to be leased by Meta; operational status should be verified separately.

Is data localisation driving all the investment?

No single explanation applies to every project. Data residency, privacy, security, sector rules, customer contracts and latency can influence location decisions, but they vary by workload and service. Avoid claiming a blanket rule requires all Indian data to remain in India unless a specific law or regulatory requirement supports that statement.

Are state data-centre policies enough to guarantee a project?

No. Policies can offer a framework and incentives, but projects must meet eligibility requirements and obtain necessary land, power, water, permits, financing and customer demand. State targets are ambitions, not guaranteed capacity.

Do data centres always use large amounts of water?

Water use depends on cooling design, climate, operating conditions and the water source. Some cooling methods use more water directly; others may use more electricity or have different trade-offs. A project’s water performance should be assessed at the site level, not inferred from a generic claim.

Will AI make every data centre obsolete?

No. AI changes workload mix, rack density, network needs and cooling design, but conventional cloud, enterprise, storage and content workloads remain important. New facilities may need flexible design to accommodate different densities over time. AI hardware and software evolve quickly, which also creates technology and investment risk.

How can a company sell into India’s data-centre market?

Identify a specific buyer, package and project stage. Understand technical, safety, security, procurement and service qualifications. Use public sources to prioritise accounts, then verify demand and tender status directly. See seven business models for profiting from the data-centre boom for vendor, service, software, energy and infrastructure routes.

Does more capacity automatically mean better digital services?

Not necessarily. Capacity must be connected to reliable power, networks, customers and skilled operations. Quality, resilience, price, location and service matter. A large announced capacity figure alone does not prove affordability or customer value.

Sources and further reading

• Savills — Data Centre Market Watch, H1 2026
• Wood Mackenzie — India data-centre capacity forecast, 27 July 2026
• Cushman & Wakefield — India Data Centre Market, H1 2026
• CBRE — India’s Data Centre Market in a New Era
• Google — India AI hub investment announcement
• Google — Visakhapatnam AI hub construction update
• Microsoft — India cloud and AI investment
• AWS — Maharashtra cloud infrastructure plan
• Digital Connexion — Visakhapatnam campus MoU
• Meta — Reliance AI data-centre announcement
• Gujarat Data Centre Policy 2026–29
• Uttar Pradesh Data Centre Policy 2026
• Gridvik: investment map, profit-pool analysis, business models, power and policy guide, Resources

This article is for general information and does not provide investment, legal, tax, engineering or environmental advice. Market data, forecasts, project plans and government policies can change. Check the linked primary sources and qualified advisers before making decisions.