The Turning Point
Five years ago, an Indian enterprise that wanted serious computing capacity looked to Singapore, Hong Kong or Amsterdam. Today the traffic runs the other way. India has moved from roughly 520 MW of installed capacity in 2020 to approximately 1.8 GW of operational IT load by mid-2026, a fourfold expansion in six years.
Savills India reports 258 MW added in the first half of 2026 alone, 59 per cent above the same period of 2025; CBRE places total capacity near 1,700 MW at end-2025 and projects roughly 500 MW of fresh supply in 2026. India now hosts about 271 facilities and holds close to 92 per cent of South Asia's colocation market. Agencies measure capacity differently, so headline figures vary by 200 to 300 MW, but the direction is not in dispute: forecasts converge on 4.5 to 6.5 GW by 2030, with AI-accelerated scenarios beyond 7 GW.
For a country generating a fifth of the world's data, the surprise is not that this is happening. It is that it took so long.
What Is Driving the Demand
Four forces operate simultaneously, and it is their convergence that explains the pace.
Data localisation. The Digital Personal Data Protection Act, RBI mandates on payment data, and IRDAI and SEBI requirements have made onshore storage a legal necessity for regulated entities: a compliance floor, not a preference, creating demand that cannot be served from offshore.
Consumption at scale. Mobile broadband subscriptions crossed 905 million by mid-2024, with per-capita consumption at roughly 32 GB per month, the highest in the world. UPI, OTT, gaming and vernacular content have turned a low-ARPU market into a high-volume one.
Artificial intelligence. This has reset the investment case. AI workloads demand rack densities of 40 to 130 kW against 5 to 8 kW for conventional colocation; pre-committed hyperscale capacity was about 82 per cent of H1 2026 absorption, and project blocks have grown from 20 MW to 50 MW and above.
Sovereign compute ambition. The IndiaAI Mission's ₹10,000 crore outlay and the intent to build domestic GPU capacity add a policy tailwind. Hyperscalers now contribute 60 to 70 per cent of incremental colocation demand.
The Competitive Landscape
The market is consolidated, capital-intensive and contested by three categories of player. Established platforms, ST Telemedia Global Data Centres India, Sify, Yotta, Nxtra by Airtel, NTT DATA and CtrlS, hold over 1 GW installed against a pipeline exceeding 4.5 GW. Conglomerate entrants, AdaniConneX, Reliance's Digital Connexion and Tata Communications, bring land banks, power relationships and execution capability that pure-plays struggle to match. Global hyperscalers are increasingly choosing to self-build rather than lease, and are expected to account for close to 30 per cent of new capacity, committing over USD 50 billion towards 1.4 GW of owned capacity by 2029. Often underrated are the mid-market specialists: ESDS, Web Werks, Pi Datacenters, RackBank and Anant Raj, often first movers in Tier-II geographies, and where much of the interesting deal flow now sits.
The Investment Picture
Announced commitments and deployed capital are two very different things, and the distinction matters for anyone underwriting this sector. Cumulative announced commitments exceed USD 126 billion, with USD 56.4 billion in calendar 2025 alone. CEEW, more conservatively, places 2019–25 committed investment near USD 95 billion, and notes candidly that only about half has a clearly identified recipient state. Actual market revenue is far smaller: about USD 9.79 billion in 2025, projected to reach USD 21.03 billion by 2031.
Marquee commitments of the past eighteen months:
| Investor | Commitment | Focus |
|---|
| Google / Alphabet | USD 15 billion over 5 years | 1 GW AI hub at Visakhapatnam |
| Amazon Web Services | USD 35 billion announced | Multi-region, Hyderabad and Mumbai |
| Microsoft | USD 17.5 billion | Cloud and AI infrastructure |
| Reliance / Digital Connexion | USD 11 billion | Visakhapatnam and related projects |
| AdaniConneX | USD 5 billion | Visakhapatnam AI campus; 100 MW Chennai |
| OpenAI with Tata Group | 100 MW scaling to 1 GW | Dedicated AI compute |
| Anant Raj | ~USD 1.5 billion | 300 MW facility |
Google's Visakhapatnam project is the largest AI hub Alphabet is building outside the United States; Andhra Pradesh estimates 188,000 jobs. Average capital cost has risen from ₹40–45 crore per MW to ₹60–70 crore per MW on higher density, advanced cooling and Tier III/IV redundancy; every 1 GW added generates an estimated USD 4–6 billion of capex and opex.
Policy and the Incentive Architecture
Central measures
Finance Act, 2026. Section 124 inserted Sl. No. 13C into Schedule IV to the Income-tax Act, 2025, exempting a notified foreign company's Indian income from procuring services from a “specified data centre” up to the tax year ending 31 March 2047. The company must be notified by the Centre, must neither own nor operate the facility, and must sell through an Indian reseller; the facility must be set up under an approved MeitY scheme and be Indian-owned. In substance, a twenty-year fiscal certainty guarantee, and its effect is on the cost of capital, since certainty of this kind can compress WACC by 200 to 400 basis points.
Infrastructure status. Data centres entered the Harmonized Master List on 11 October 2022, subject to a 5 MW IT load threshold, unlocking long-tenor lending and better ECB terms. A live technical issue: the Finance Act 2026 definition is broader and carries no MW threshold, so a facility may qualify under one and not the other until MeitY issues its scheme.
Draft policy and FDI. The draft National Data Centre Policy proposes up to 20 years of conditional tax exemption, electricity duty exemption, input tax credits, single-window clearance and pre-allocated land in dedicated zones; it awaits notification. 100 per cent FDI is permitted under the automatic route, save for restricted-country investors.
State policies
At least eight states, Maharashtra, Tamil Nadu, Telangana, Karnataka, Uttar Pradesh, Odisha, Rajasthan and Haryana, have notified dedicated policies. The common stack: capital subsidy, stamp duty and electricity duty relief for five to ten years, preferential tariffs, concessional land, “Essential Service” classification permitting 24×7 operation, and single-window clearance. Tamil Nadu's 2021 policy was due to expire in April 2026 with no extension notified, which investors should verify.
Where the Capacity Is, and Where It Is Going
Mumbai and Navi Mumbai remain dominant at 41 to 50 per cent of national capacity, on multiple cable landing stations, BFSI proximity, mature grid and scalable land in Panvel and Thane. Chennai holds 18 to 20 per cent as the second cable gateway, Hyderabad benefits from Telangana's ten-year electricity duty exemption, Delhi-NCR is the fastest-growing challenger, Pune has emerged as an AI compute location, and Bengaluru draws SaaS and GCC demand despite water stress.
The story of 2026 is geographic diffusion. Tier-II markets hold an estimated 60 to 80 MW today, expected to exceed 100 MW by end-2026. Visakhapatnam has become India's most significant new hyperscale cluster, Ahmedabad and GIFT City offer policy support and renewable proximity, and Nava Raipur targets 80 MW in phase one. The drivers are latency, land cost, water and renewable proximity: Gujarat, Rajasthan, Tamil Nadu and Karnataka offer ISTS-exempt PPAs enabling a 50 to 70 per cent renewable share, increasingly a hard requirement for tenants with net-zero commitments.
The Real Constraints
Capital is not the binding constraint in India today. Power, water and skilled execution are.
Power is the single most important variable. A 100 MW facility needs firm, redundant supply from preferably two independent grid sources with on-site backup. Data centres took roughly 0.5 per cent of national electricity in 2024, projected to more than double by 2030, absorbable against the 500 GW non-fossil target, but only where transmission and substation infrastructure exist. Power approvals, not land, gate most projects; treat the sanctioned load timeline as the critical path from day one.
Cooling and water. Cooling can be up to 40 per cent of facility energy use, and a 100 MW site can consume around 2 million litres of water a day. Water-stressed locations need dry or hybrid systems despite a PUE penalty, and AI workloads are pushing direct-to-chip and immersion cooling.
Land, certification and talent. Campuses need 15 to 25 acres per 100 MW with clear title, and land assembly remains the commonest cause of slippage. Uptime Tier III is the minimum, increasingly Tier IV for BFSI; IGBC/LEED certification, zero liquid discharge and ESG disclosure are now standard. Critical facility engineers and certified operations staff remain in short supply: a genuine bottleneck.
Why India, and Why Now
India's case rests on advantages difficult to replicate elsewhere in Asia. Demand is domestic and structural, generated by India's own users and a digital public infrastructure stack processing billions of transactions monthly, and localisation makes a proportion of it non-substitutable. Land, construction and operating costs sit materially below Singapore, Japan, Hong Kong and Australia: comparable engineering capability at a fraction of the cost is the fundamental arbitrage. Singapore's moratorium and Malaysia's grid constraints have redirected regional demand, ISTS waivers make genuinely green facilities viable, and infrastructure status has opened long-tenor debt and InvIT structures. The exit market works too, as Carlyle's investment in Nxtra and Brookfield's in Digital Connexion show.
The Honest Risks
It would be poor advice to present only the upside. Power availability and transmission adequacy remain the hardest constraint, with connection timelines varying enormously by state. Water scarcity is a live siting risk in several hubs, and land title continues to cause delays. Localised oversupply is possible if pipelines are delivered simultaneously into markets where absorption lags, though pre-commitment above 80 per cent suggests this is not an immediate concern. And the AI demand assumption underpinning much of the pipeline, while robust today, may not hold at the same intensity through the decade.
None of these argue against investment. They argue for disciplined site selection, conservative underwriting, and contractual protection on power cost pass-through.
Conclusion
India's data centre sector has crossed from a story about potential to one about execution at scale. The capacity base has quadrupled in six years, the capital is committed, the tax framework is settled for two decades, and demand is domestic, regulated and growing.
What remains scarce is the ability to execute: to secure power at scale, to assemble land cleanly, to build to hyperscale standards on schedule, and to structure returns that survive the pass-through of Indian power costs. For promoters and investors who bring that capability, the next five years represent one of the most significant infrastructure opportunities available.