A Comprehensive LinkedIn Deep-Dive | June 2026
Every time you stream a video, run a Google search, send an email, prompt an AI chatbot, or tap ‘Pay Now’ on your phone, a data center somewhere quietly does the heavy lifting. These facilities are the unsexy but utterly indispensable backbone of the modern digital economy. Yet most people — including many in business and finance — have only a vague idea of what they actually are, how enormous they have become, or how much power they consume.
This article is a deep dive into the world of data centers: what they are, how they work, how much power they need, how they acquire that power (and why that is becoming one of the most complex infrastructure problems of our era), and where the market is heading over the next decade. We zoom in on India — one of the fastest-growing data center markets on the planet — and survey the global landscape.
1. What Is a Data Center?
A data center is a purpose-built physical facility that houses computing and networking equipment — servers, storage arrays, network switches, routers, power distribution units, cooling systems — used to collect, process, store, and distribute data and digital services at scale. Think of it as a city-block-sized brain: constantly running, always-on, and requiring tremendous amounts of electricity, cooling, and physical security to function.
Data centers come in several flavors:
Enterprise / Private Data Centers: Owned and operated by a single organization for its own use. A bank running its core banking software or a hospital hosting patient records would use these. Example: Tata Consultancy Services (TCS) operates its own multi-site enterprise data center infrastructure across India.
Colocation (Colo) Facilities: Third-party facilities that rent out floor space, power, and connectivity to multiple tenants. Tenants bring their own servers. Example: Equinix, the world’s largest colocation operator, runs over 260 data centers across 72 metros globally.
Hyperscale Cloud Data Centers: Massive facilities built and operated by cloud giants — Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform (GCP), Meta — to power their global cloud services. These can span hundreds of thousands of square metres. Example: AWS’s Northern Virginia campus in Ashburn, VA — the ‘Data Center Capital of the World’ — has over 70 individual data center buildings.
Edge Data Centers: Smaller, distributed facilities positioned close to end users to reduce latency for real-time applications. Example: A telco placing a micro data center inside a mobile tower site to serve 5G users in that area.
Managed Service / Hosted Data Centers: The provider manages both the space and the hardware/software on behalf of clients.
The Uptime Institute Tier Classification
The most widely used framework for categorizing data centers by reliability is the Uptime Institute’s Tier Standard, the global benchmark:
| Tier | Description & Uptime SLA |
|---|---|
| Tier I — Basic Infrastructure | Single path for power and cooling; no redundancy. 99.671% uptime (~28.8 hrs downtime/year). Suitable for small offices. |
| Tier II — Redundant Capacity Components | Redundant components (UPS, generators, chillers) but still single distribution path. 99.741% uptime (~22 hrs/year). |
| Tier III — Concurrently Maintainable | Multiple active+passive paths; any component can be maintained without shutdown. 99.982% uptime (~1.6 hrs/year). Most new enterprise/colo builds target Tier III. |
| Tier IV — Fault Tolerant | All paths fully active; any single failure has zero impact. 99.995% uptime (~26 min/year). Used for banking, defense, stock exchanges. |
Example: The Bombay Stock Exchange (BSE) and National Stock Exchange (NSE) both operate Tier IV-equivalent primary data centers because even a second of downtime can cost crores of rupees in lost trades.
2. How Big Are They? The Staggering Scale
Size in the data center world is measured in two primary ways: physical footprint (square metres of raised floor space) and IT power capacity (megawatts available for servers and storage). The latter is increasingly the dominant metric because AI and high-performance computing workloads have radically increased the power density per square metre.
From Closets to Campus Cities
A mid-size colocation facility typically covers 5,000–20,000 sq m and supports 20–50 MW of IT load.
A hyperscale data center campus can cover 100,000+ sq m (the size of 14 football fields) and draw 200–500+ MW.
Meta’s data center in Eagle Mountain, Utah spans 158,000 sq m with an eventual planned capacity of over 970 MW — equivalent to the output of a medium-sized power station.
Microsoft’s campus in Boydton, Virginia covers roughly 500,000 sq m across multiple buildings — one of the largest technology campuses on Earth.
The AI Inflection: Density Goes Through the Roof
Traditional server racks draw 5–15 kW. AI training and inference infrastructure tells a completely different story:
| Infrastructure Type | Power Density per Rack |
|---|---|
| Standard enterprise server rack | 5–10 kW |
| High-performance computing rack | 20–40 kW |
| AI/GPU rack (Nvidia H100 cluster) | 40–80 kW |
| Nvidia GB200 NVL72 AI rack (2025) | Up to 120 kW |
| Nvidia Rubin generation (projected 2027) | 180–360 kW |
| Projected ‘1 MW rack’ (early 2030s) | ~1,000 kW |
This density explosion is why new hyperscale campuses are designed around total megawatt draw — not square footage — and why power has replaced real estate as the binding constraint in data center development.
Global Numbers at a Glance (2025–2026)
Total global installed IT power capacity: ~122.2 GW (Q1 2025)
Global data center electricity consumption: ~415 TWh in 2024, projected to reach 860 TWh by 2025 and 1,587 TWh by 2030
Number of hyperscale data centers globally: ~1,189, with 642 in the United States
AWS alone: 2.3 GW of active IT capacity; Meta: 1.5 GW; Microsoft Azure: 1.2 GW
Scale Check: At 1,587 TWh/year by 2030, global data centers would consume more electricity than India’s entire current annual consumption (~1,500 TWh). Every AI query on ChatGPT uses roughly 10× more electricity than a Google search.
3. How Do They Operate? The Four Pillars
Running a large data center is an exercise in precision engineering across four interdependent systems. Failure in any one can bring everything down.
Pillar 1: Power Infrastructure
Power enters the building from the grid at medium-voltage (typically 33 kV or 11 kV in India; 12–15 kV in the US), then passes through a cascade:
Main Substation & Transformers: Step-down transformers reduce voltage for distribution. High-availability facilities have dual feeders from independent substations — if one grid supply fails, the other carries the load instantly.
Uninterruptible Power Supplies (UPS): Double-conversion online UPS systems sit inline, continuously converting AC to DC and back — ensuring zero transfer time during grid glitches. They bridge the 10–15 seconds until diesel generators reach operating speed.
Diesel Generator Sets (DGs): Multiple large generators (500 kVA to 3 MVA each, with banks of 20–30 units in a hyperscale facility) provide standby generation. They must reach full voltage and frequency within 10 seconds. Most sites keep 48–72 hours of diesel on-site.
Power Distribution Units (PDUs): Step-down and distribute power at the row and rack level, often with real-time metering for energy management.
Static Transfer Switches (STS): Automatically switch between redundant power feeds in under 4 milliseconds — faster than most electronic equipment can detect.
Pillar 2: Cooling Systems
Servers generate enormous heat. Cooling typically consumes 30–40% of total facility power and is the second-largest operational cost after electricity itself.
CRAC/CRAH Units: Traditional raised-floor hot-aisle/cold-aisle architecture uses precision air conditioning. Effective up to ~20 kW/rack.
Chilled Water Systems: Central chillers produce chilled water circulated through Computer Room Air Handlers (CRAHs). More energy-efficient at scale.
Free Cooling / Air-Side Economizers: In cooler climates, outside air replaces mechanical cooling for hundreds of hours per year. Example: Facebook’s Lulea, Sweden campus uses near-freezing Arctic air almost year-round, achieving a PUE of just 1.07.
Direct Liquid Cooling (DLC): Cold water or dielectric fluid piped directly to server CPUs and GPUs via cold plates. Handles 40–100 kW/rack. Already standard in AI GPU clusters.
Immersion Cooling: Servers are fully submerged in non-conductive dielectric fluid. Handles 100+ kW/rack with no fans required. Used by Bitcoin miners and increasingly by AI hyperscalers.
PUE — The Efficiency Metric: Power Usage Effectiveness (PUE) = Total Facility Power / IT Equipment Power. A PUE of 1.0 is perfect. Global average in 2025: ~1.54. Hyperscalers: Google (1.10), Microsoft (1.12). For every rupee of electricity a well-run hyperscale campus uses on computing, it uses just 10 paise on overhead. For the average enterprise data center, that overhead is 54 paise.
Pillar 3: Connectivity
A data center is only as useful as its network connections. Facilities are located near internet exchange points (IXPs) and fiber backbones, with multiple redundant fiber feeds entering from geographically diverse paths. A Tier III colocation facility typically offers carrier-neutral access to dozens of network service providers. Speeds range from 1 Gbps to 400 Gbps and beyond for internal spine-leaf fabrics in hyperscale builds.
Pillar 4: Physical Security & Environmental Monitoring
Multi-factor access (badge + biometric) at every zone boundary
24×7 CCTV surveillance with 90-day retention; man-traps / airlocks at entry points
On-site security personnel, perimeter fencing, anti-vehicle bollards
DCIM (Data Center Infrastructure Management) software: real-time dashboards monitoring power draw at the outlet level, temperature at every rack, humidity, water leak detection, clean-agent fire suppression, and diesel fuel levels
4. What Are Their Power Requirements?
Power is the lifeblood of a data center. Unlike most commercial buildings, data centers require it to be continuous (99.999%+ availability), clean (within tight voltage/frequency tolerances), redundant (multiple independent sources), and scalable.
Consumption by Scale
| Data Center Type | Typical IT Power Draw |
|---|---|
| Small enterprise server room | < 1 MW |
| Mid-size colocation (e.g., Tier III, 500 racks) | 5–20 MW |
| Large colocation campus | 50–100 MW |
| Hyperscale facility (single building) | 100–200 MW |
| Hyperscale campus (multiple buildings) | 400–1,000+ MW |
| Future AI mega-campus (under construction 2025–27) | 1,000–3,000 MW |
100 MW is enough electricity to power approximately 80,000 average Indian homes. A 1 GW AI mega-campus would consume as much electricity as a small Indian city of ~800,000 households.
The AI Power Shock
Training GPT-4 reportedly consumed ~50 GWh — roughly the annual electricity use of 4,600 US homes.
Every ChatGPT query uses roughly 10× more electricity than a Google search.
By 2030, AI workloads are projected to account for 50% of all data center electricity consumption globally, up from ~15% in 2024.
Global electricity consumption attributable to AI: 415 TWh in 2024, projected to reach 945 TWh by 2030.
India-Specific: India’s data centers consumed less than 1% of total national electricity in 2024. By 2030, that share is projected to rise to ~3% — representing ~35–40 TWh/year as installed capacity grows from ~1.4 GW today to 9 GW by 2030.
5. Is Special Power Required?
Yes — in several important ways. Data centers are not just large electricity consumers; they are uniquely demanding customers that require power characteristics unavailable from a standard utility connection.
Quality Requirements
Voltage Regulation: Servers require input voltage within ±5% of nominal (230V / 400V in India). A voltage dip of >10% lasting even a few milliseconds can cause server reboots or hardware damage. UPS systems provide the buffering.
Frequency Stability: Most servers tolerate 49.5–50.5 Hz in India (50 Hz nominal). UPS systems decouple the load from grid frequency fluctuations entirely.
Power Factor Correction: Data centers are large inductive loads. They must actively manage power factor (above 0.95) to avoid penalty tariffs from utilities.
Harmonic Distortion: Switching power supplies in servers introduce current harmonics. Large UPS systems include harmonic filters to prevent these from feeding back into the grid.
Redundancy Architecture
| Configuration | Description |
|---|---|
| N+1 | One extra unit beyond what is required. If 10 UPS modules needed, install 11. Any one can fail without service impact. |
| 2N (Tier III/IV standard) | Full duplication — two completely independent power paths (A-feed and B-feed) serve every rack. Servers with dual power supplies draw from both simultaneously. |
| 2N+1 | Two full systems plus one extra. Used in Tier IV and ultra-critical environments. |
Example — 2N in Practice: Every rack in an Equinix IBX data center has two PDU feeds: the ‘A’ PDU from one UPS/generator chain, and the ‘B’ PDU from a completely independent chain. Even if an entire generator string fails, the rack remains powered from the other path.
Fuel and Backup Generation
For a 100 MW hyperscale campus:
The diesel generator bank might comprise 40–60 units of 2–3 MVA each
On-site fuel storage: typically 400,000–600,000 litres of diesel (enough for 48–72 hours at full load)
Emergency fuel contracts with multiple suppliers ensure continuous resupply during extended grid outages
Generators undergo weekly block-load testing and monthly full-load testing
High-Voltage Direct Current (HVDC) — The Emerging Standard
Traditional data centers use AC power distribution, which requires multiple conversion steps (AC → DC in UPS → AC out of UPS → DC again inside servers). Each conversion loses 2–5% as heat. HVDC distribution (typically 380 VDC) eliminates one or two conversion stages, improving overall energy efficiency by 8–12%. Google, Meta, and ABB have championed HVDC architectures in hyperscale facilities, and it is now the preferred choice for new large-scale builds.
6. How Do Data Centers Procure Power?
Power procurement strategy is arguably the most complex and strategically consequential decision a data center developer makes. It directly determines operating cost, carbon footprint, and ability to scale. The approach has evolved dramatically over the past decade.
Method 1: Direct Grid Connection via Utility Tariff
The simplest approach: connect to the local electricity utility and pay the applicable commercial/industrial tariff. In India, this means applying to the State Electricity Distribution Company (DISCOM) for a dedicated high-tension (HT) connection at 33 kV, 66 kV, or 220 kV for very large campuses, and paying tariffs set by the State Electricity Regulatory Commission (SERC). In Maharashtra, for example, commercial HT consumers in 2025 pay approximately ₹8–9/kWh (inclusive of fixed demand charges, energy charges, and cross-subsidy surcharges).
Pain Point: Grid capacity constraints are the #1 bottleneck for data center development in India. Mumbai’s Navi Mumbai and Thane corridor, Chennai’s northern belt, and Hyderabad’s Outer Ring Road all have substations running close to saturation, requiring operators to fund new substation construction or transmission lines at their own cost. Application-to-energization timelines of 12–36 months are common.
Method 2: Open Access and Third-Party Power
Under India’s Electricity Act 2003, large consumers (generally above 1 MW) can access the grid to buy power from generators other than their local DISCOM — a right called ‘open access’. This allows data centers to purchase power directly from Independent Power Producers (IPPs), procure through power exchanges (IEX — Indian Energy Exchange), and benefit from potentially lower landed costs compared to DISCOM tariffs, after accounting for wheeling charges, transmission losses, and cross-subsidy surcharges.
Method 3: Power Purchase Agreements (PPAs)
A PPA is a long-term contract between a data center (buyer) and an energy generator (seller) that locks in the price and volume of electricity over a multi-year period. PPAs come in two forms:
Physical PPA (Wheeling PPA): The actual electrons from the power plant are physically delivered to the data center via the grid. The buyer takes title to the electricity and the associated Renewable Energy Certificates (RECs). This requires open access approval, wheeling agreements with the transmission utility, and banking arrangements.
Virtual PPA (VPPA): A purely financial contract. The data center buys power from the grid at market rates, while contracting with a renewable generator for a fixed strike price. If the market price is above the strike, the generator pays the data center the difference; if below, the data center pays the generator. The buyer gets the RECs for sustainability reporting. More common in the US.
Mega-PPA Example: In May 2024, Microsoft signed an agreement with Brookfield Asset Management to deliver 10.5 GW of new renewable energy capacity between 2026 and 2030 — one of the largest clean energy procurement deals in history. In 2025, Meta was the world’s largest corporate clean energy offtaker, with over 10 GW of PPAs signed globally.
Method 4: On-Site Generation (Captive Power Plants)
Gas-based Captive Power Plants (CPPs): Natural gas or LNG-fired plants co-located with the data center. Provide reliable baseload power but face fuel supply and price risk.
Solar Captive: Rooftop or ground-mounted solar panels. Can offset 10–30% of total consumption for a large campus.
Solar + Battery Storage: Increasingly viable as battery costs decline. A 100 MW data center might pair 50 MW of solar with 200 MWh of battery storage.
Nuclear / SMR (Small Modular Reactors): Microsoft signed a deal to purchase power from the restarted Three Mile Island Unit 1 (835 MW, Pennsylvania). Google signed a PPA for SMR power from Kairos Power (~2030). Nuclear offers 24×7 carbon-free power at high reliability.
India Captive Example: Reliance Industries’ 3 GW AI mega-campus planned for Jamnagar, Gujarat is designed to run entirely on captive renewable energy — a combination of solar, wind, and green hydrogen — making it one of the world’s first truly carbon-free hyperscale AI campuses at commissioning.
7. Do Data Centers Use Renewable Energy?
Renewable energy has moved from a ‘nice-to-have’ to a strategic imperative for data center operators — driven by corporate sustainability commitments, investor pressure, regulatory requirements, and increasingly, pure economics.
Corporate Sustainability Commitments
Google: Committed to operating on 24/7 carbon-free energy by 2030 globally — meaning carbon-free electrons every hour of every day, not just annual matching. Achieved 64% 24/7 CFE coverage in 2024.
Microsoft: Targeting 100% renewable electricity for Azure. Pledged to be carbon negative (removing more CO2 than it emits) by 2030. Achieved 29.9% reduction in Scope 1 & 2 emissions from 2020 baseline.
Amazon Web Services: World’s largest corporate buyer of renewable energy for several consecutive years. Committed to matching 100% of electricity consumption with renewables by 2025.
Meta: The largest corporate clean energy PPA offtaker globally in 2025, with over 10 GW of signed agreements. Targeted net-zero emissions by 2030.
However, a 2025 analysis by NewClimate Institute and Carbon Market Watch raised serious concerns: tech companies’ GHG emissions targets may have ‘lost their meaning’ as AI-driven electricity demand growth causes absolute emissions to rise even as renewable energy percentages stay flat.
The PPA Market for Renewables
In 2024, the tech sector procured over 17 GW of clean energy through direct third-party PPAs — making it the largest corporate clean energy buyer by sector globally. The North American renewable PPA market for data centers alone was valued at \$7.8 billion in 2025, representing 42.1% of global data center renewable PPA revenue.
Nuclear Power: The ‘Dark Horse’ of Carbon-Free Energy
As solar and wind PPAs prove insufficient to achieve 24/7 carbon-free coverage (they only generate when the sun shines or wind blows), nuclear power is emerging as a critical complement:
33% of data center operators reported exploring nuclear power options in 2025, up from just 11% three years earlier
Microsoft-Three Mile Island: Microsoft’s 20-year PPA to purchase all output from the restarted TMI Unit 1 (835 MW) was a watershed moment in 2023–24
Google-Kairos Power: 500 MW of power from molten-salt SMRs, with deliveries starting ~2030
62% of data center operators in a 2025 survey were exploring off-grid solutions including nuclear and natural gas
Renewable Energy in India: Challenges and Opportunities
India added a record 25.1 GW of non-fossil capacity in 2025, the world’s fastest renewable buildout in absolute terms
State policies in Telangana, Karnataka, and Tamil Nadu now mandate 100% renewable sourcing for new large data center campuses
Solar PPAs: ₹2.5–3.5/kWh; Wind PPAs: ₹3.0–4.5/kWh — now below DISCOM retail tariffs for many categories
Challenge: Open access rules, wheeling charges, and cross-subsidy surcharges add ₹1.5–3.0/kWh to the effective cost of third-party renewable power, eroding the cost advantage
Challenge: Grid congestion in renewable-rich states (Rajasthan, Gujarat, Tamil Nadu) limits the ability to physically wheel power to data center-heavy metros
IEEFA Note: India’s data center sector is at a ‘crossroads’: rapid growth is increasing grid stress, while regulatory inconsistencies between states are making renewable procurement unnecessarily complex. Harmonized national-level rules are needed urgently.
8. Typical Terms of Power Agreements
Whether dealing with a utility, an IPP, or a renewable energy developer, data center operators negotiate power contracts with carefully defined commercial terms:
| Term | Typical Value / Range |
|---|---|
| Contract Duration | 10–25 years; nuclear PPAs averaging ~18 years |
| Pricing Structure | Fixed, fixed with annual escalation (1–3%), or indexed to market hub |
| Solar PPA Price (India) | ₹2.5–3.5/kWh |
| Wind PPA Price (India) | ₹3.0–4.5/kWh |
| Firm & Dispatchable Renewable (India) | ₹5.0–5.5/kWh |
| Nuclear PPA (US, existing large reactor) | \$45–70/MWh (~₹3.8–5.9/kWh) |
| Nuclear SMR forward-start PPA (US) | \$65–100/MWh |
| Minimum Offtake / Take-or-Pay | 70–90% of contracted volume; penalties for shortfall |
| Security Deposit / Performance Bond | Typically 3–6 months of contract value |
| Interconnection Lead Time | 12–48 months; varies by grid congestion and project location |
| Change-in-Law Provisions | Tariff pass-through mechanisms for regulatory changes |
| Early Termination Fee | Net present value of remaining contract cash flows |
| Green Attributes (RECs/I-RECs) | May be included or separately traded; critical for ESG reporting |
Regulatory Trend: US regulators are increasingly requiring standardization. Colorado’s PUC mandated that Xcel Energy use a set of ‘principles’ when contracting with data centers: upfront fees, 15-year contract minimums, minimum bills, security deposits, and early-exit penalties — to protect other ratepayers from effectively subsidizing data center infrastructure upgrades.
9. Major Data Centers in India
India is in the midst of a data center supercycle, driven by rapid cloud adoption, the Digital India programme, smartphone penetration, AI deployment, and mandatory data localisation requirements under the Digital Personal Data Protection (DPDP) Act 2023.
Key Hubs
Mumbai and Navi Mumbai hold ~54% of India’s total installed data center capacity, driven by their position as the financial capital, the landing point for multiple subsea cable systems, and India’s oldest internet exchange (NIXI Mumbai). Hyderabad (6%), Chennai (8%), Pune, Delhi NCR, and Bengaluru account for most of the balance.
Major Operators and Facilities
| Operator / Facility | Key Facts |
|---|---|
| Hiranandani-backed Yotta Infrastructure | Yotta NM1, Navi Mumbai: 30,000 racks, 130 MW, Tier IV certified — one of the largest Tier IV facilities in the world. |
| CtrlS Datacenters | Hyderabad-headquartered, multi-site. Claims Asia’s largest ‘hyperscale-ready’ Tier IV certified facility. Total capacity ~250 MW across India. |
| Nxtra by Airtel | 11 large and 120 edge data centers. Plans to invest ₹10,000 crore to expand to 400 MW by 2025. |
| Sify Technologies | Pioneer Indian colo operator. 30+ data centers. Recently commissioned new campuses in Chennai and Noida. |
| Equinix India (Mumbai, Chennai) | Global colo giant, entered India in 2020. MU1 Mumbai: ~4,000 sq m. Expanding rapidly for multinational cloud/enterprise customers. |
| NTT DATA (formerly Netmagic) | 9 facilities totalling ~300 MW. Strong presence in Mumbai and Bengaluru. |
| AdaniConneX | JV between Adani Enterprises and EdgeConneX. Commissioned a 400 MW campus in Chennai (Dec 2025) with 200 MW of integrated renewables and Tier IV certification. |
| Microsoft Azure India | Operates hyperscale facilities in Pune and Chennai. Jan 2026: expanded Hyderabad region by 50 MW to 150 MW IT load, adding GPU clusters for Azure OpenAI. |
| Amazon AWS India | Operates AWS Asia Pacific (Mumbai) and AWS Asia Pacific (Hyderabad) regions. Significant capacity additions planned for 2026–2030. |
| Google Cloud India | Operates GCP Mumbai region. Announced 8-storey, 381,000 sq ft data center in Navi Mumbai with ₹1,144 crore investment in first 10 years. |
| ST Telemedia Global (STT GDC) | Singapore-headquartered. Significant presence in Mumbai and Chennai. Among top 5 DC operators in India by capacity. |
Investment Momentum: The India data center market attracted investments exceeding ₹27.99 lakh crore (~US\$30 billion) between 2020 and 2025. The pipeline of announced projects through 2030 is even larger, with BlackRock, Brookfield, CPP Investments, GIC, and other global institutional investors all having made multi-billion-dollar commitments.
10. The Indian Data Center Market
Market Size and Growth
| Year | Installed Capacity (GW) | Market Size (USD Bn) |
|---|---|
| 2024 | ~1.4 GW | ~\$8.2 Bn |
| 2025 | ~4.5 GW | ~\$9.8 Bn |
| 2026 (estimate) | ~5.5 GW | ~\$11.2 Bn |
| 2028 (projected) | ~8.0 GW | ~\$15.5 Bn |
| 2030 (projected) | ~9.0 GW | ~\$19.0 Bn |
| 2031 (projected) | ~15.2 GW | ~\$21.0 Bn |
Demand Drivers in India
Digital Payments Revolution: India processes over 12 billion UPI transactions per month. NPCI’s back-end, banks, and fintechs hosting this require enormous and growing data center capacity.
Cloud Adoption Acceleration: Enterprise cloud spending in India is growing at 30%+ per year. AWS, Azure, and GCP are all expanding their India regions, each requiring hundreds of megawatts.
Data Localisation (DPDP Act 2023): Mandatory in-country storage of personal data of Indian citizens is creating structural, policy-driven demand for domestic data center capacity.
OTT and Content: With 800+ million internet users, India is the world’s second-largest online video market. Netflix, Disney+ Hotstar, Amazon Prime Video, and dozens of Indian OTT platforms require vast streaming infrastructure.
AI and ML: Indian enterprises across BFSI, telecom, retail, and healthcare are deploying AI at scale, driving demand for GPU-equipped colocation and cloud regions.
Government Digital Services: DigiYatra, DigiLocker, Ayushman Bharat Digital Mission, GSTN, and hundreds of Digital India services are hosted on data center infrastructure.
Challenges in India
Power Availability: Reliable 24×7 grid power at required quality remains a challenge outside metros. High dependency on diesel generators increases operating costs and carbon footprint.
Land Acquisition: Industrial land near substations with available power capacity is scarce and expensive in metros.
Regulatory Complexity: Multiple approvals needed across state electricity boards, urban local bodies, environmental regulators, and telecom authorities. No single-window clearance.
Water for Cooling: Evaporative cooling towers consume significant water — a growing concern as Indian cities face water stress. Chennai and Bengaluru facilities are exploring closed-loop and dry cooling alternatives.
Skilled Manpower: Shortage of qualified data center engineers, particularly for critical facilities management and advanced cooling systems.
11. Global Data Center Market — The 10-Year Outlook
The global data center market is in an unprecedented investment supercycle — driven by a confluence of forces that show no sign of abating.
Market Size Projections
| Year | Global DC Market Size (USD Bn) — Best Estimate |
|---|---|
| 2025 | ~\$387 Bn |
| 2026 | ~\$430 Bn |
| 2028 | ~\$535 Bn |
| 2030 | ~\$680 Bn |
| 2033 | ~\$900 Bn |
| 2035 | ~\$1,100 Bn (~\$1.1 Tn) |
At approximately 11% CAGR, the global data center market will roughly triple in size over the coming decade — from ~\$387 billion today to ~\$1.1 trillion by 2035. For reference, this would make it larger than the global aviation industry.
Capacity Growth
2025: ~122 GW global installed IT capacity
2030: ~200 GW (nearly doubling in 5 years — equivalent to commissioning one new 100 MW data center every two days for five years)
2035: Active hyperscaler IT load alone projected to reach ~147 GW, a 6× increase from 2025 hyperscaler capacity
Investment Scale
Amazon, Google, Meta, and Microsoft collectively spent ~\$228 billion on data center capex in 2024 — a 55% year-on-year increase
Projected total hyperscaler capex for 2026: \$600+ billion
Total infrastructure investment needed for data centers through 2030: estimated \$1.2–3.0 trillion globally
Regional Picture
Americas: Largest region (~50% of global capacity), with 17% CAGR through 2030. Northern Virginia, Dallas, Chicago, Phoenix, and Atlanta are dominant US markets.
EMEA: Frankfurt, Amsterdam, London, and Paris (the ‘FLAP-D’ cluster) dominate. Growth shifting to secondary European markets (Madrid, Warsaw, Milan) as power constraints bite in primary hubs.
Asia-Pacific: 12% CAGR to 2030; capacity expanding from 32 GW to 57 GW. Tokyo, Singapore, Sydney, and Seoul lead, with India growing fastest in absolute terms.
Emerging Markets: Malaysia (Johor Bahru), Indonesia (Jakarta), Saudi Arabia, UAE, Nigeria, and South Africa attracting significant new investment.
The AI Mega-Campus Era
Meta: \$10 billion, 1 GW campus in Lebanon, Indiana (under construction 2025)
Microsoft ‘Project Stargate’: \$100 billion AI infrastructure programme across the US, with multi-GW campuses in Texas, Wisconsin, and Pennsylvania
Amazon: Multi-GW AI campuses planned in Ohio, Georgia, and Virginia
Reliance Jio / Jamnagar: 3 GW AI campus in Gujarat powered entirely by captive renewables
NEOM (Saudi Arabia): 300+ MW AI infrastructure campus as part of the \$500 billion smart city project
12. What Does This Market Depend On?
1. AI Adoption Curve
The single biggest uncertainty is how fast generative AI and AI inference scale. If AI adoption follows the high-growth scenario (ChatGPT-type ubiquity across every business process), data center power demand will be at the top end of forecasts. If compute efficiency improves faster — through algorithmic breakthroughs or new chip architectures — growth could moderate. Nvidia’s roadmap (H100 → B200 Blackwell → Rubin) suggests power per rack will keep rising even as compute efficiency improves.
2. Grid Infrastructure and Power Availability
Transmission network capacity: US grid operators like PJM have a 3+ year backlog of interconnection requests totalling over 2,600 GW. Building new 500 kV transmission lines takes 10–15 years.
Distribution transformer shortages: Global shortage of large power transformers is causing 2–4 year lead times.
India-specific: State-level substation capacity expansion, right-of-way for new transmission corridors, and DISCOM financial health all act as bottlenecks.
3. Renewable Energy and Carbon Policy
Corporate sustainability mandates and emerging carbon regulations (US SEC climate disclosure rules, EU Corporate Sustainability Reporting Directive, India’s forthcoming BRSR requirements) are pushing operators toward 100% renewable energy procurement. The speed of renewable buildout — and specifically the availability of 24/7 firm, carbon-free power — will determine whether sustainability pledges are achievable.
4. Water Availability
Traditional cooling towers in a 100 MW data center can consume 1–3 million litres of water per day. As climate change intensifies droughts and cities impose water restrictions, this is becoming a material operational risk. Meta’s Lulea campus uses no water for cooling (pure outside-air free cooling). Operators are investing in closed-loop cooling, dry-air cooling, and water recycling systems.
5. Semiconductor Technology Trajectory
The compute density and power efficiency of chips directly determine data center power requirements. Nvidia’s Blackwell B200 GPU delivers 2.5× more training performance per watt than the H100. Custom silicon from Google (TPU v5), Amazon (Trainium 2), and Microsoft (Maia 2) are also pushing efficiency forward. If Moore’s Law continues to deliver, AI model serving could become dramatically cheaper per unit of compute — moderating power growth.
6. Regulatory and Policy Environment
Data localisation laws (India DPDP, EU GDPR, China PIPL): Force operators to build in-country rather than serving markets from offshore
Environmental permitting: Singapore paused new data center approvals from 2019–2022; Ireland imposes strict energy caps
Incentive regimes: India’s Data Center Policy (2020) and state-level policies in Telangana, Maharashtra, and Tamil Nadu provide capital subsidies and fast-track approvals
Grid access rules: Open access regulations, wheeling charges, and REC frameworks determine the economics of clean power procurement
7. Security and Geopolitics
Geopolitical tensions (US-China tech decoupling, Taiwan Strait risk) are reshaping data center geography. Companies are building ‘sovereign cloud’ infrastructure in politically stable jurisdictions and avoiding concentration risk. India — with its democratic institutions, large domestic market, and improving digital infrastructure — is a major beneficiary of this trend.
13. Future Trends: What Comes Next
Nuclear Renaissance
Nuclear power — specifically Small Modular Reactors (SMRs) — could be the ‘killer app’ for data center power. SMRs promise compact footprint, factory-built construction, and 24/7 carbon-free output. Key milestones:
Kairos Power (Google): First US SMR for a commercial data center, ~2030
X-energy / Dow Chemical: First industrial-site SMR in the US — if successful, could trigger widespread adoption
India: Department of Atomic Energy developing an SMR policy framework that may include dedicated nuclear power for industrial consumers including data centers
Liquid Cooling Becomes the Default
By 2027, liquid cooling is expected to be the standard for new AI data center construction. Nvidia’s Rubin generation (2027) will require 180–360 kW/rack; next-generation potentially 720 kW+; and ‘1 MW rack’ architectures are expected by the early 2030s. No air-based cooling system can handle this. Two-phase immersion cooling — in which a refrigerant boils off the chips and condenses on a water-cooled heat exchanger — offers extremely high heat transfer density and is entering commercial deployment.
AI-Optimised Design
Disaggregated architecture: Compute, memory, and storage separated and connected by ultra-high-speed optical interconnects — optimised for AI inference, where memory bandwidth is the bottleneck
Edge AI: Inference workloads pushed to the edge; a distributed network of small ‘AI inference pods’ in telco towers, office buildings, and smart city infrastructure
AI-managed infrastructure: DCIM systems themselves run by AI, enabling predictive maintenance, dynamic power allocation, and real-time thermal optimisation
The ‘Digital Infrastructure Real Estate’ Revolution
Data centers have become one of the fastest-growing alternative asset classes in global real estate. Listed Data Center REITs — Equinix (EQIX), Digital Realty (DLR), Iron Mountain (IRM) — have delivered exceptional total returns over the past decade. Private equity firms, sovereign wealth funds (GIC, ADIA, CPP), and infrastructure funds are pouring capital into the sector. In India, InvIT and REIT structures are being explored for monetizing data center assets.
Conclusion: The Power Behind the Curtain
Data centers are no longer a niche IT concern — they are critical national infrastructure, on par with highways, ports, and power plants. The electricity they consume is already reshaping global energy markets, driving multi-gigawatt renewable energy procurement, reviving nuclear power programmes, and straining electricity grids from Virginia to Maharashtra.
For India, the opportunity is immense: a young, tech-savvy population; a government committed to digital infrastructure; record-low renewable energy costs; and a surging demand for cloud, AI, and digital services. But capturing that opportunity requires solving the power puzzle — ensuring reliable, affordable, and increasingly clean electricity at the right locations, at the right time.
The next decade will see the data center industry consume as much electricity as entire nations, spend trillions of dollars in capital, and become one of the defining infrastructure challenges of the 21st century. Understanding how these facilities are built, powered, and operated is no longer optional for anyone in business, energy, technology, real estate, or public policy.
What aspect of data center infrastructure are you most focused on — power procurement, sustainability, AI infrastructure, or the India opportunity? Drop your thoughts in the comments.
#DataCenters #AI #EnergyTransition #DigitalInfrastructure #India #CloudComputing #RenewableEnergy #PowerSector #DigitalIndia #Infrastructure
Key Sources
Uptime Institute Tier Classification System — uptimeinstitute.com
JLL 2026 Global Data Center Market Outlook — jll.com
S&P Global: Global data center power demand expected to almost double by 2030
IEA / IEEFA / Mordor Intelligence — India Data Center Market Reports 2025–2031
Precedence Research: Data Center Market Size, Share, and Trends 2026–2035
Pillsbury Law: Power Purchase and Interconnection Agreements for Data Centers
Data Center Dynamics: Everything data center operators need to know about PPAs
GlobeNewswire: Hyperscaler Data Center Capacity to Surge More Than 6× by 2035
IBEF: India’s data centre market projected to reach US\$22 billion by 2030
Programs.com: Measuring the Data Center Boom — Facts and Statistics 2026
AI Tool Discovery: AI Data Center Power — 415 TWh in 2024, 945 TWh by 2030