The core assets in India's data center boom are power and land
AI summary card
The core assets in India's data center boom are power and land
Bernstein believes India's data center capacity could rise from about 1.5GW currently to the high end of the 5-8GW range by 2030, with the true advantage belonging to asset owners that control land, grid connection, and dispatchable power.
- The report breaks down the India data center opportunity into equipment, utilities, and data center asset owners, focusing especially on the latter two asset-heavy plays.
- Experience from PJM and ERCOT in the U.S. shows that once data center clusters form, utilities with regional dispatchable generation, land, and grid connection resources benefit significantly.
- India's data center capacity is expected to rise from about 1.5GW to the high end of the 5-8GW range by 2030, while competition for land and substation access in core areas such as Navi Mumbai is intensifying.
- The report expects Indian operators to favor the colocation model, with hyperscalers providing the GPUs, thereby avoiding GPU balance sheet risk and technology obsolescence risk.
- Adani Group is seen as particularly advantaged due to its renewable energy, private thermal power, transmission, land, and grid connection resources; Reliance Industries and L&T are also identified as important beneficiary themes.
Report interpretation
Overview
This report discusses asset-heavy opportunities in India's wave of data center construction, with the core conclusion that competition in data centers ultimately comes down to power access and land resources. The report first reviews how major U.S. data center hubs such as PJM and ERCOT developed, then compares the colocation data center and GPU/cloud hosting business models, and finally maps the framework to India, concluding that Adani Group, Reliance Industries, L&T, and parts of the power utility value chain have beneficiary potential.
Core views
First, India's data center capacity may move toward the high end of the expected 5-8GW range by 2030, making power and land the key bottlenecks. Second, compared with the GPU ownership model, the colocation model has higher capital efficiency and lower technology obsolescence risk, and is therefore expected to be better suited to most Indian asset owners. Third, U.S. experience shows that generators with existing dispatchable power, grid connection positions, and land can capture capacity price upside, wholesale power prices, and long-term PPA premiums. Fourth, the strongest beneficiary framework in India is land × power, with Adani Group standing out most clearly through its combined advantages in renewables, thermal power, transmission, land, and grid connection.
Analysis framework
The report uses a U.S.-to-India mapping approach: first examining data center demand, interconnection queues, generator share prices, and capacity market changes in PJM and ERCOT; then breaking down the business models, capex, and valuations of data center asset owners; and finally screening potential Indian winners based on four variables: land, power, grid connection, and dispatchable capacity.
Methodology notes
Deriving India's data center and power beneficiaries from PJM/ERCOT
U.S. data center clusters depend on supportive regulation, wholesale power markets, fiber, land, and interconnection speed; the report uses these conditions to assess which Indian companies have similar resources.
Differences in asset ownership and capital risk
The colocation model leases data center infrastructure while customers bring their own compute hardware; GPU/cloud hosting generates higher revenue but requires higher capex, shorter contracts, and carries technology obsolescence risk.
Land multiplied by power access is the core moat for data center asset owners
The report believes the key differentiation comes from land reserves in suitable areas, substation and grid access, dispatchable power, and transmission capability.
Valuation ranges for mature DC REITs and private-market transactions
The report uses EQIX and DLR at about 22-23x 12m forward EV/EBITDA and the roughly 21x AirTrunk deal as valuation references for data center assets.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Adani Group / Adani Green / Adani PowerCore potential beneficiary of India's data center and power demand
- Strengths
- It has large land resources, renewable energy, private thermal power, transmission, commercial power capacity, and future expansion capability; the report says its advantages stand out.
- Weaknesses
- Some power supply may need to be delivered through DISCOM or intermediary structures such as Adani Energy Solutions, and the specific contract and regulatory details are not quantified.
- Comparison
- Compared with most Indian companies, Adani Group has the most complete resource combination under the land × power framework.
- Risks
- Group-level financing, regulation, power-source preferences, project execution, and valuation volatility could all affect realization.
- Reliance IndustriesParticipates in India's data center asset opportunity through its land reserves
- Strengths
- The report mentions about 0.5 Mn acres of land in Gujarat and more than 5K acres of large land parcels in Navi Mumbai.
- Weaknesses
- The report notes that it is not under coverage, and its power and transmission advantages are described less fully than Adani Group's.
- Comparison
- Strong on the land dimension, but the report identifies Adani Group as the more prominent land × power combination.
- Risks
- The pace of data center projects, capital allocation priorities, and power access capability still need to be monitored.
- L&TData center, power plant construction, and potential operator
- Strengths
- It can benefit simultaneously from data center construction, power project construction, and potentially participation as a data center owner-operator.
- Weaknesses
- The excerpt does not provide specific order, capacity, or valuation data.
- Comparison
- Compared with pure power asset players, L&T is more of an engineering construction and project execution beneficiary.
- Risks
- Order conversion, margins, project cycles, and capital intensity need ongoing validation.
- EQIX / DLRGlobal benchmarks for mature colocation data center assets
- Strengths
- Mature operators have long-term leases, stable cash flow, and market-recognized high valuations; the report rates EQIX and DLR Outperform.
- Weaknesses
- Valuations are already around 22-23x 12m forward EV/EBITDA, so margin of safety depends on delivery of growth.
- Comparison
- Broadly consistent with the roughly 21x AirTrunk deal and the 20-30x private-market range, making them an important reference for valuing Indian assets.
- Risks
- Interest rates, capital expenditure, lease pricing, power bottlenecks, and intensifying competition could pressure returns.
- CRWV / IREN / NeocloudsRepresentatives of the GPU compute leasing and cloud service model
- Strengths
- Revenue per IT MW can be significantly higher than under the colocation model, allowing direct capture of AI compute demand.
- Weaknesses
- Requires higher upfront capex, shorter contracts, and bears GPU technology obsolescence and balance sheet risk.
- Comparison
- Compared with the colocation model, it has higher revenue upside but greater risk; the report rates CRWV Underperform.
- Risks
- GPU price cycles, customer concentration, financing costs, technological iteration, and utilization falling short of expectations.
- Vistra / Constellation Energy / NRGU.S. examples of power beneficiaries from data centers
- Strengths
- They own dispatchable nuclear or gas assets in PJM and ERCOT and benefit from capacity prices, wholesale power prices, and long-term PPA premiums.
- Weaknesses
- Their earnings are highly dependent on the institutional structure of U.S. regional power markets and the location of existing generation assets.
- Comparison
- They provide a reference for Indian power companies: existing dispatchable capacity and the right regional positioning matter more than simply adding new installed capacity.
- Risks
- Capacity price declines, regulatory intervention, fuel costs, and power demand coming in below expectations.
Key data
- India data center capacityCurrently about 1.5GW, with 2030 expectations at 5-8GW and possibly near the high endFrom the report's industry discussion conclusions.
- India data center capital expenditureAbout Rs 50 Cr/MW excluding computeThe report says this is below U.S. levels.
- Colocation model payback periodAbout 5 yearsUsed by the report as a reference for colocation asset economics.
- Neocloud revenue and capital expenditureRevenue per IT MW is about 8-10x that of colocation, while capex is about $45Mn vs $8-15Mn for colocationHigher revenue comes with higher leverage, shorter contracts, and technology obsolescence risk.
- U.S. AI colocation deal sizeAbout $98Bn, corresponding to 4GW of IT load, with average contract terms of 10-15 yearsReflects the strength of AI colocation demand.
- ERCOT large-load interconnection queueRose from 49GW to 225GW in about 2.5 yearsMainly driven by data center demand.
- U.S. interconnection waiting timeAbout 6 yearsThe report views grid interconnection as the most critical site-selection factor.
- Data center REIT valuationEQIX and DLR at about 22-23x 12m forward EV/EBITDABroadly consistent with AirTrunk at about 21x acquisition multiple and the 20-30x range seen in private transactions.
- U.S. power stock performanceFrom Jan'22 to Dec'25, Vistra and Constellation Energy were about 8x, and NRG about 4xPerformance divergence mainly reflects dispatchable capacity and regional positioning.
- PJM capacity pricesFrom about $29/MW-day in 2024/25 to about $270/MW-day in 2025/26, then to about $329/MW-day in 2026/27Rising capacity market prices directly improve generator earnings.
Impact & implications
The investment implication is that the India data center theme should not focus only on compute or the equipment chain, but more importantly on asset-heavy companies with land, grid connection, transmission, and dispatchable generation capabilities. If the U.S. experience is repeated in India, the availability and speed of power will determine project execution, and companies with combined land and power resources may benefit across data center construction, long-term power supply, colocation operations, and engineering construction.
Risks
- If India's data center capacity does not reach the high end of the 5-8GW range by 2030, the re-rating logic for land and power assets may weaken.
- Grid connection, substation access, land approvals, and local infrastructure could become bottlenecks to project execution.
- The GPU/cloud hosting model faces risks from high capex, short contracts, rising leverage, and technology obsolescence.
- Valuations of data center operators and power beneficiaries already partly reflect AI infrastructure expectations, so they may pull back if growth disappoints.
- Customer acceptance in India of thermal power, renewable energy, and intermediary power supply structures still needs to be validated.
- Changes in policy, PPA, DISCOM, transmission, and power trading rules may affect data center electricity costs and contract structures.
- For some Indian companies, the report discusses only industry beneficiary status and does not provide full stock ratings, target prices, or earnings forecasts.
What to watch
- Whether contracted and under-construction data center capacity in India continues to move toward the high end of the 5-8GW range by 2030.
- Land transaction prices, land acquisition, and substation access progress in key regions such as Navi Mumbai, Panvel, and Gujarat.
- The pace of Adani Group's expansion in commercial power, transmission, thermal equipment lock-in, and renewable installations.
- The share of hyperscalers in India choosing colocation, self-build, or GPU hosting, as well as contract duration and pricing.
- Whether India sees transactions similar to U.S. behind-the-meter colocation and long-term clean dispatchable power PPAs.
- Whether PJM and ERCOT capacity prices, interconnection queues, and U.S. data center power demand continue to reinforce the global mapping logic.