Data Centers.
India's data-centre engine doubled in 2025 to ~1.5 GW and is racing to 2 GW — on $126bn of committed capital and an AI-and-localisation demand wall. Capital-rich, power-constrained.
The thesis in one line: demand (AI + hyperscale cloud + DPDP localisation) is near-unbounded; the binding constraints are power, water, land and capital, not customers. Returns are infrastructure-grade where power is secured — and cyclical where it isn't. Overbuild and the hyperscaler capex cycle are the risks to watch.
AI/ML workloads + hyperscale cloud + DPDP/RBI data-localisation mandates. Demand is not the constraint.
1.5 → ~2 GW by 2026 (+30%, ~500 MW fresh supply); 700+ MW under construction, 1-2 GW planned.
₹50-70 cr per MW; $126bn committed, $180bn+ by 2026. Long payback — a capital sink, not a quick flip.
Grid reliability + water are the real bottleneck (CEEW); renewable PPAs now table-stakes, not optional.
Crowded — 39+ operators plus hyperscalers building captive. Margins defended by power deals and land banks.
15-20yr hyperscale leases = infra-grade cash flows; offset by overbuild risk and hyperscaler capex-cycle dependence.
Scores are 0-100 favourability — a Gravitywell judgement read across demand, supply, cost, power/ESG, competition and returns, anchored to the cited 2025-26 data. Not yet a formula-driven score (roadmap); the rating word is the headline, the bar is relative. Trend arrows (▲/▼) are vs an inaugural prior-cycle baseline and are tracked each cycle going forward.
~1.5 GW (2025) → ~4.5 GW by 2030 — IT-load capacity (facility power demand runs ~2× higher)
$126bn committed by end-2025, +45% to $180bn+ in 2026
$10.1bn (2025) → ~$25bn by 2031 (~16% CAGR)
The pivot the rest of the page can't answer alone: is demand absorbing the 1-2 GW pipeline? It is — strongly, and mostly pre-leased — which is both the comfort and the concentration risk.
Demand is cloud/hyperscaler-led and locked in BEFORE commissioning (~77% pre-committed) — but that is the risk, not just the comfort: a hyperscaler capex pause would hit a pipeline already pledged largely to them. BFSI is the fastest-growing buyer (~25% CAGR); colocation is 84% of the market vs hyperscaler self-build (~24% CAGR).
Top 5 ≈ 55-60% of installed capacity. Shares indicative; sources differ.
Three hyperscalers alone pledged ~$67.5bn through ~2030.
Pricing & density — AI racks (100kW+, liquid-cooled) lift yields 20-30% — the margin upgrade behind the build.
Deal structures shifted to stabilised-asset trades + customer/operator/real-estate JVs. Stabilised assets clear at 20-25× EBITDA.
Key gap for public investors: no scaled pure-play yet — Sify Infinit is the first. Most exposure is proxy (Anant Raj) or conglomerate (RIL/Bharti). Valuations (Anant Raj 52× PE) already price aggressive growth.
Where the sector actually lives — most capacity is unlisted, PE- and strategic-backed. Capacity, ownership, and the last marker of value.
Pre-funded 500 MW in Mumbai; not listing
+100 MW Chennai build-to-suit
Tier-4; staged across states for incentives
Renewable-coupled; undercuts grid ~25%
Carlyle entered at $1.2bn (2020); IPO mulled
500 MW AI-rack roadmap
Blackstone's India DC entry
Landmark JV consolidated
Where venture capital enters the theme — mostly “neocloud” GPU-compute, riding the IndiaAI tailwind.
Pivoted model→cloud; ₹3bn rev FY26 (3×), 25+ enterprise clients
Blackwell cluster at L&T's Chennai DC — the VC layer that already reached public markets
20,000+ Blackwell-Ultra GPUs; sovereign-AI positioning
IndiaAI Mission pick; sovereign LLM + compute
Affordable GPU access for developers / smaller teams
Enterprise cloud + edge nodes into tier-2
Colo + AI-cloud; competing with the neocloud pack
VC read: the GPU-cloud / neocloud space is crowding fast (E2E, Yotta Shakti, Krutrim, NeevCloud, NxtGen, Cyfuture, Sarvam) and rides the IndiaAI tailwind (34,000 subsidised GPUs at ₹150/hr). The genuine white-space is the picks-and-shovels — liquid cooling, DCIM/automation, edge and sustainability software are still MNC-dominated (Schneider, Delta) and under-built by Indian startups. That's where the un-crowded VC entry sits.
A screened, exposure-weighted basket — each listed name weighted by its data-centre purity score (not naively equal-weighted), after liquidity and quality screens. Selection is rule-driven and set ex-ante.
Real point-to-point anchors: each name rebased to 100 at −3y; the −1y (295) and now (412) levels from its actual 1Y & 3Y returns, purity-weighted. Intra-period linear (daily shape/drawdowns need a price feed).
Rules-based: include a listed name if its DC purity score ≥ 20/100 AND it clears the eligibility screens. Weight by purity score (exposure-weighted), single-name cap 25%, overflow redistributed pro-rata. Quarterly reconstitution. Selection is rule-driven, set ex-ante — not a curation of past winners.
- ✓ Liquidity & size — investable free-float, adequate ADTV (all current names large/mid-cap)
- ✓ Quality — positive profitability (excludes loss-makers)
- ✓ Purity — DC revenue-exposure / relevance score ≥ 20 of 100
Rules-eligible, pending verified data: HFCL, Tejas Networks, Thermax, Voltas, Siemens, Hitachi Energy. Purity scores are a documented judgement tier (exact DC-revenue % isn't cleanly disclosed; this mirrors how FactSet RBICS / MSCI thematic relevance is applied). Returns are representative point-to-point figures from public sources, not an audited daily-rebalanced backtest.
Selection-bias caution: this basket is built from names already public over the window, which the AI/DC theme has already re-rated. Past returns are therefore upward-biased and NOT a forward estimate; the rules (not the hindsight) are what's intended to repeat.
INFORMATIONAL / RESEARCH ONLY — not investment advice, not a stock recommendation, and not a SEBI Research-Analyst model portfolio. Past performance does not indicate future returns. Figures are representative; precise CAGR/XIRR and a real NAV require audited price series. Do your own research / consult a SEBI-registered adviser.
The externalities a government must price in — the sector's demand is digital, but its constraints are physical.
Power & water are the licence to operate. The Central Ground Water Authority has already blocked/delayed Hyderabad & Chennai sites over groundwater; BEE PUE norms + the Draft National DC Policy 2025 reward green-certified builds (20-yr tax breaks). India's ~50% non-fossil grid helps the renewable path. India PUE ~1.4 already beats the 1.5-1.8 global average.
AI capex sustains; power secured via renewable PPAs; pre-leasing stays >90% (≈10 GW facility power)
Strong demand, but grid/water/land throttle the pace; winners = secured-power platforms
Global AI-capex pause hits the hyperscaler-concentrated pipeline; overbuild → vacancy + multiple compression
Funding is shifting cash → debt; RBI is easing bank lending against REIT/InvIT units. The open question behind the $126-180bn 'committed' headline: how much is actually financeable — it's 86% foreign-capital-dependent and rate-sensitive.
Risks quantified, not just listed — the levers that swing the underwriting. Directional, illustrative.
- ↑ AI/ML training + inference — power-dense racks (liquid-cooled) are the new demand curve.
- ↑ Hyperscale cloud — AWS, Microsoft, Google building captive + leasing colo at scale.
- ↑ Data localisation — DPDP Act + RBI financial-data rules force domestic processing.
- ↑ OTT, 5G and edge — pushing nodes into tier-2 cities.
- ↑ Submarine cables + cheap renewable PPAs — landing capacity where power is securable.
- ! Power & water — grid reliability and cooling water are the true ceiling (CEEW).
- ! Capital intensity — ₹50-70 cr/MW, long payback; rate-sensitive, balance-sheet heavy.
- ! Overbuild — 1-2 GW planned against demand that, while strong, is hyperscaler-concentrated.
- ! Hyperscaler capex cycle — a global AI-capex pause would hit India's pre-leased pipeline.
- ! Land, cooling & talent for AI density; tariff / tax / DPDP-rule changes.
Core-plus infrastructure: 15-20yr hyperscale leases, but ₹50-70 cr/MW capex and power risk. Favour operating platforms and secondaries over greenfield at peak valuations.
Buy the picks-and-shovels, not the concrete — liquid cooling, DCIM/automation, edge, and AI-infra software where the multiples (and exits) live.
Long secured-power operators / REITs vs the power-and-water constraint; watch overbuild and the global hyperscaler capex cycle as the turn signal.
DPDP localisation is the demand tailwind; grid, water and land are the bottleneck. Treat as strategic AI-sovereignty infrastructure — incentivise, but plan the power first.
Data vintage June 2026. Anchored to 2025-2026 industry and official prints; figures across sources differ and are reconciled to the cited ranges. Sources: Market $10bn → $25bn; drivers (IBEF)P · Capacity ~1.5 GW, +34% (Cushman/CBRE)S · Private backers/valuations — Nxtra $3.1bn, Blackstone/Lumina (Tracxn/BS)S · Listed-beneficiary returns (Tickertape / Equitymaster / Simply Wall St)S · Unit economics (JM Financial / CareEdge)S · Deals + 20-25× EBITDA (Houlihan Lokey)S · Sify Infinit IPO / listed proxies (Outlook Business)S · Power ~3% of grid (IEEFA / CEA)S · Water 358bn L (Reccessary)S · 1 lakh jobs by 2030 (MoS S&T)S · India #2 APAC / global benchmark (Newkerala)S · Neocloud / AI-cloud startups (Inc42 / TechCrunch)S · $126bn committed → $180bn 2026; hyperscaler pledges (Fiscalzenith)S · Operator shares + geography (Mordor / Outlook / Cushman)S · Power & water constraint (CEEW)S · Scorecard scores, unit-economics ranges & basket purity weights — Gravitywell estimateE
Data confidence. Confidence tiers — HARD (official/exchange/operator prints: capacity, IPOs, deals, returns, state policy, RBI/CEA data); FIRM (tier-1 industry: absorption, pre-lease, vacancy, PUE, market size); SOFT/JUDGEMENT (scorecard scores, purity scores, pricing ranges, unit economics — representative, sourced where possible, labelled). Figures across analysts differ; we reconcile to cited ranges and flag the metric basis (e.g. IT-load vs facility power).
Sourcing. Every figure is sourced and dated. We tier provenance — Primary (official, regulatory, exchange or company filings), Secondary (tier-1 industry research and reputable media), and GW estimate (our own reconstruction or opinion, labelled, never presented as external fact). We prefer primary where it exists, reconcile divergent prints to cited ranges, and hold every number point-in-time — dated, and never silently restated; revisions publish as dated changes.
Fact vs opinion. Facts vs opinion — market sizes, official prints, prices, named deals and agency ratings are sourced facts (Primary/Secondary). Scores, grades, purity weights, scenario paths and indicative sparkline points are Gravitywell's analytical opinion (GW estimate) — labelled, not presented as external data.
The sector, each cycle.
Data Centers refreshed every cycle, with the scorecard, dashboard, and capital read. More sectors rolling out.
← All sectors