Attracting AI Data Centre Infrastructure Investment in India

ELI5/TLDR

India generates 20 percent of the world’s data but hosts only 3 percent of global data centre capacity. That gap is the whole story. Deloitte’s report lays out six pillars — real estate, power, connectivity, compute, talent, and policy — where India needs to fix things fast if it wants to be an AI hub instead of just an AI consumer. The numbers are real (45-50 million additional sq ft needed by 2030, $60 billion already invested 2019-2024, GPU supply stuck at ~29,000 units), but the recommendations read like a consulting firm’s wish list for a government client.

The Full Story

The Gap

India’s AI market is valued at $7-8 billion today, expected to hit $20-22 billion by 2027 at a 30 percent CAGR. But the infrastructure underneath is thin. The country has roughly 150 data centres with 1,200-1,300 MW of IT load capacity — mostly crammed into four cities: Mumbai (48 percent of total capacity), Chennai, Bengaluru, and Delhi-NCR, which together account for 87 percent of inventory. Compare that with the US, which has 5,381 data centres and about 16,000 MW.

India attracted $1.2 billion in private AI investment in 2024. The US attracted $109.1 billion. That ratio tells you where things stand.

The Money Pouring In

The investment pipeline is impressive on paper. Over $60 billion flowed into Indian data centres between 2019 and 2024, with $19 billion in 2024 alone. The big plays:

Jio plans the world’s largest data centre in Jamnagar — 3 GW capacity, estimated $20-30 billion cost, in partnership with NVIDIA
Microsoft is building 660 MW capacity in Telangana
AdaniConneX committed $5 billion over five years, targeting 1 GW by 2030
ST Telemedia is investing $3.2 billion to add 550 MW
Princeton Digital Group announced $1 billion for Mumbai and Chennai expansion
Yotta placed a $1 billion cumulative order with NVIDIA for AI chips

Capacity is projected to reach 2,100-2,200 MW by 2026, with an assured 5,000-6,000 MW by 2030 — and a potential additional 5,000-6,000 MW if policy gets it right.

The Six Pillars (and Why Each One is Hard)

Real Estate. India needs 45-50 million additional sq ft by 2030, up from 13 million in 2023. Construction cost is low — $7.5 million per MW, 40-45 percent below the global average — and land runs $100-130 per sq ft. But acquiring that land takes over a year, you need 40-plus approvals from multiple government levels, and there are no standardised clearance processes across states. Fiscal incentives are inconsistent — some states offer stamp duty relief, others offer nothing.

Power. A traditional rack draws 5-10 kW. An AI rack draws at least 10x that. India’s data centres consumed 9-10 TWh in 2023 and will need an additional 40-45 TWh. The country still runs 55 percent on non-renewable energy. Average PUE sits at 1.9 versus 1.3 for sustainability-optimised designs. Electricity is cheap ($0.08-0.14/unit) and 45 percent of the grid is less than 10 years old — real advantages. But renewable energy banking policies vary wildly between states, and the Karnataka High Court struck down the national Green Energy Open Access Rules in January 2025.

Water is the hidden crisis. A 20 MW data centre needs 1.37 million litres of water daily for cooling. The very cities hosting most data centres — Bengaluru, Mumbai, Chennai — face severe water scarcity. Bengaluru alone has a 500-million-litre daily deficit.

Connectivity. India’s mobile data costs are the world’s cheapest at $0.16 per GB (versus $6.00 in the US). The country has 3.9 million km of fibre optic network and 270 million 5G users. Mumbai and Chennai have strong submarine cable access. But leased high-speed lines (10 Gbps+) cost INR 10-20 lakh per month, regulations around dark fibre are ambiguous, and rural fibre coverage remains patchy.

Compute. India had about 4,000 GPUs in 2023, rising to 10,000 by end of 2024 through IndiaAI tenders, with 29,000 announced under the IndiaAI Mission for 2025. Target is 50,000 by 2027-28. Individual H100 GPUs cost $30,000-50,000 each. The US AI Diffusion Framework caps India’s access to high-end chips at 50,000 NVIDIA H100-equivalent units until 2027. India has only 7 supercomputers in the global TOP500, versus 172 for the US.

Talent. India has the second-largest AI workforce globally — 600,000-650,000 professionals — with the highest AI skill penetration rate at 2.8 (US: 2.2). But demand is growing to 1.25 million+ by 2027 at 15 percent CAGR while the AI market grows at 30-35 percent, signalling a widening gap. None of the world’s top 25 elite AI research institutions are in India.

Policy. Foreign companies using Indian data centres face permanent establishment risk under the Income Tax Act, potentially triggering double taxation even if they do nothing beyond hosting data. Data centres get no sector-specific tax breaks comparable to SEZs or IFSCs. A proposed retrospective GST amendment threatens to reverse a Supreme Court ruling that allowed input tax credit for data centre construction. Data localisation requirements are scattered across multiple laws. IP frameworks for AI-generated content are unclear.

What Deloitte Recommends

Across 24 identified challenges, the report proposes interventions including:

Single-window clearance systems and Data Centre Facilitation Units
Data Centre Economic Zones operating on a plug-and-play model
Recognising data centres under the Essential Services Maintenance Act
Capital subsidies for MEP costs and 5-year interest subvention on borrowed capital
Grid modernisation with weighted tax deductions for captive renewable plants and hi-tech cooling
Standardised national banking policies for renewable energy
Clarifying dark fibre regulations in line with Singapore and Australia models
Government-to-government negotiations to soften US chip export restrictions
Exempting data hosting from PE classification under Section 9(1)(i) of the Income Tax Act
Extending IFSC-like tax incentives to data centres
Amending the Copyright and Patent Acts for AI-generated content
Visa incentives for global AI experts

Claude’s Take

This is a competent industry report doing exactly what you’d expect a Big Four firm to do: map a landscape, identify gaps, and hand the government a menu of asks that would also benefit the firm’s clients. The data is solid and well-sourced — the 20-percent-data-but-3-percent-capacity framing is genuinely striking, and the city-level breakdowns and cross-country benchmarking tables are useful reference material.

But the framing is relentlessly optimistic in that consulting-firm way. Every challenge is “an opportunity.” The Jio Jamnagar data centre ($20-30 billion, 3 GW) gets mentioned without any scepticism about whether it will actually be built at that scale. The report treats government policy as the primary bottleneck, which conveniently means the answer is always “more incentives, more exemptions, more subsidies” — things Deloitte’s clients would directly benefit from.

The genuinely interesting tension buried in the report: India’s water crisis vs. its data centre ambitions. The cities with the best connectivity and talent are the ones running out of water. Liquid cooling and zero-water designs exist but are expensive. This is a real constraint that policy tweaks alone won’t solve.

The US AI Diffusion Framework capping India’s GPU access at 50,000 H100-equivalents is also underplayed. If the US treats India as a Tier 2 country for chip access, no amount of domestic policy reform changes that math.

claude_score: 5. Useful as a reference document for anyone tracking India’s data centre buildout. The numbers are well-aggregated. But it reads like what it is — an industry advocacy document dressed up as analysis. No surprises, no contrarian takes, no hard trade-offs acknowledged. The recommendations are a wish list, not a prioritised strategy.