Where is economic value accumulating in India's data-centre buildout?
Operational capacity rose from ~375 MW in 2020 to ~1.5 MW in 2025 (MeitY), 1,530 MW at Sep-2025 (CBRE), 1.6 GW mid-2026 with ~6 GW pipeline to 2029 (JLL). H1 2026 absorption was 101 MW, 82% pre-committed hyperscale. Demand shifted from enterprise cloud to AI-led high-density workloads. Power availability is now as decisive as fibre — grid, substations, renewables, and backup set timelines.
Announced commitments (~$94B, 2019–2025) must be separated from powered, utilised MW — only commissioned capacity generates cash flows.
1. Executive Summary Scale crossed from niche to infrastructure cycle. AI raises rack power, cooling, connectivity intensity. Hyperscalers (Microsoft $17.5B 2026–29, Google $15B 5-yr Vizag hub, AWS $12.7B to 2030) anchor demand. Geography: Mumbai/Chennai/NCR/Bengaluru ~90% of stock, Hyderabad/Pune/Vizag rising for AI campuses. Track lifecycle: land → power approval → construction → grid connection → commissioning → utilisation.
2. Scale: From Niche to Multi-GW Pipeline 375 MW (2020) → 1,500 MW (2025) + 38,231 GPUs via 14 providers; 260 MW new supply 9M-2025 across 23M sq ft; block sizes 20 MW → 50 MW+. At gigawatts, each campus drives power, construction, cooling, fibre, financing demand. Distinguish verifiable commissioned MW from forecast pipeline.
3. AI Workloads Redefining Design Enterprise phase → AI training/inference: CtrlS up to 45 kW/rack with liquid cooling; AdaniConneX 50–500 MW AI factories; direct-to-chip + immersion cooling. Three effects: higher substation/transformer intensity, higher MEP/thermal content per MW, denser networking complexity. Rising kW/rack is the early supplier-demand signal.
4. Hyperscaler Commitments as Demand Evidence Microsoft: $17.5B + Hyderabad South Central live Aug 2026 (4th India region). Google: $15B Vizag gigawatt hub with AdaniConneX/Nxtra + green power + cable landing. AWS: $12.7B to 2030, Mumbai + Hyderabad regions. Shift from leasing to multi-year campus commitments; 82% pre-commitment corroborates. Still separate commitments from powered MW.
5. Power and Grid as Binding Constraint Ministry of Power (Feb 2025): ~854 MW IT load; reliability/renewables/efficiency under review. Large campuses need major connections, substations, transmission, firm renewables — often slower than buildings. Monitor: Generation/PPAs (signed PPAs, storage pairing). Transmission (approval queues). Equipment (order books tied to named projects). Energy management (contracted vs hourly clean delivery). Test: named campus + scope + quantified order.
6. Geography: Mumbai/Chennai Anchor, Hyderabad/Vizag Expand Mumbai: cables, fibre, financial demand, operator ecosystem. Chennai: routes, enterprise, land — both JLL hyperscale hubs. Hyderabad: Microsoft + AWS regions, reinforcing loop. NCR: government/enterprise. Pune: emerging AI-compute hub. Vizag: integrated gigawatt + green power + fibre + landing model. Weight power, fibre, land, water, regulation, renewables, customer proximity jointly.
7. Policy Toward Infrastructure Economics TRAI 2022: land, power, connectivity, approvals, standards. Budget 2026–27: tax holiday to 2047 for foreign buyers of Indian DC services + 15% safe harbour — route global workloads via India. UP 2026 Policy: Rs 2 lakh cr target, subsidies, duty exemptions, AI-compute cover — but MoUs > construction. Treat policy as leading only with land + power + ground-breaking + orders + connection.
8. Lifecycle: Announcements to Cash Flows Funnel: announcement → land → power approval → construction → procurement → grid connection → commissioning → onboarding → utilisation. Leading: approvals/orders. Coincident: connections/commissioned MW. Lagging: revenue/EBITDA. Track announced/committed/under-construction/operational separately; require power milestones; monitor pre-commitment/absorption/pricing; map campus → operator/customer/power/cooling/supplier; distinguish contracted vs firm clean power.
9. Five Documented Weaknesses 1. Headlines overstate supply — $94B vs 1.5–1.6 GW operational. 2. Power timelines under-disclosed — grid can strand complete buildings. 3. Renewable contracting ≠ 24/7 firm supply. 4. Supplier exposure asserted without project linkage. 5. Utilisation/pricing sparse — 101 MW absorption known, project vacancy/ramp patchy.
10. Five Concrete Improvements 1. Quarterly commissioned-MW tracker reconciling MeitY/JLL/CBRE at campus level. 2. Power-milestone disclosure per campus (approval, substation, connection date, firm vs contracted). 3. Standard AI-readiness metrics (rack density, cooling, contracted hyperscale). 4. Project-level supplier attribution with contract values before exposure claims. 5. Joint pre-commitment + absorption + pricing monitoring.
11. Conclusion Scarcity shifted from floor space to reliable power, delivered cooling, contracted MW. Hyperscalers confirm direction; Mumbai/Chennai retain networks; Hyderabad/Pune/NCR/Vizag absorb AI campuses. Returns hinge on approvals → connection → utilisation, not announcements.
Where Brane Fits: campus-to-power/customer/supplier/utilisation monitoring; evidence grounding via financial AI methodology; reliability via benchmark; leads vs lags via signal index.
FAQ Q1: Current scale? 1.5–1.6 GW (2025–mid-2026) from 375 MW (2020); ~6 GW pipeline to 2029; ~$94B committed — execution tracking critical. Q2: Why is power central? Grid + substations + firm renewables often slower than buildings; determines timelines, location, bankability, supplier demand. Q3: Which cities? Mumbai/Chennai/NCR/Bengaluru ~90%; Hyderabad/Pune/Vizag rising for large AI campuses. Q4: Separate announcements from supply? Track 4 MW buckets separately; require land/power/construction/connection/onboarding evidence. Only powered + utilised MW support cash flows. Q5: Policy incentives? Budget 2026–27 holiday to 2047 + UP subsidies — help only with construction + connection + commissioning.
Sources: MeitY/PIB Mar 2026; JLL mid-year Aug 2026; Microsoft $17.5B Dec 2025 + Hyderabad Aug 2026; Google Cloud Feb 2026 + Adani Apr 2026; CBRE Nov 2025; CtrlS; AdaniConneX; Budget speech 2026–27; Invest UP 2026; Ministry of Power RS Q798 Feb 2025.
This article is educational and does not make a recommendation about any security. Evidence as of September 2026.