India and Southeast Asia: the next decade of compute.
Capital, capacity and AI demand are converging across India and Southeast Asia. Vastra Infra is positioned in two of the region's most strategic clusters.
Three lenses matter: lifecycle stage, facility type and resilience.
By lifecycle stage
Built, commissioned and carrying live IT load. The only capacity truly in the market today.
Funded and building. Typically 12 to 30 months from commissioning, and largely pre-let.
Announced, often with land and an MoU, but not yet building. The largest and least certain category.
A new site on raw land. Power, water, fibre and approvals must all be created. This is the Vastra scope.
Added phases or halls on an existing campus. Fastest to deliver, since the infrastructure exists.
By type of facility
Operator builds and leases space, power and cooling to many tenants. The bulk of Indian supply.
A cloud provider builds and owns for its own use. Google's Visakhapatnam campus is this model.
Built by a bank, telco or corporate for its own workloads, often for regulatory or DR reasons.
Purpose-built for high-density AI compute, with liquid cooling and far higher power per rack.
Small distributed sites close to users, for latency. Measured in single-digit MW.
By Uptime Institute tier
Tier is the classification of redundancy and uptime. It sets capex per MW, and it is what a hyperscale or BFSI tenant underwrites against.
| Tier | Redundancy | Availability | In practice | Where it is used |
|---|---|---|---|---|
| Tier I | Single path, no redundancy | ~99.67% | Any maintenance or fault takes the site down | Rare in institutional builds |
| Tier II | Redundant components, single path | ~99.74% | Some component failure tolerated; path work still disruptive | Small enterprise and edge |
| Tier III | Concurrently maintainable, N+1 | ~99.98% | Any component can be serviced with no downtime | The Indian market standard for colocation and hyperscale |
| Tier IV | Fault tolerant, 2N or 2N+1 | ~99.995% | Any single failure is absorbed with no impact on IT load | BFSI, sovereign and mission-critical; a minority of stock |
Tier III is the working standard across Indian colocation and hyperscale supply; Tier IV is built selectively where a BFSI, sovereign or mission-critical tenant will pay for fault tolerance. Tier speaks to redundancy, not density: AI workloads push rack density far above the conventional 5 to 10 kW, driving liquid cooling and a different building altogether. A site can be Tier III and still be unsuitable for AI. Our developments are designed for AI-ready density from the outset.
Live capacity is in Mumbai and Chennai. The pipeline has moved to Andhra Pradesh and Telangana.
And it is being built in a different form: self-build mega campuses rather than leasable colocation.
| State / hub | Operational | Under construction | Announced pipeline | Type / mix | Lead operators and policy |
|---|---|---|---|---|---|
| Maharashtra Mumbai, Navi Mumbai | ~801 MW | ~448 MW | ~893 MW planned | Colocation-led, multi-tenant. Deepest leasable market, with hyperscale leasing on top. AI added as zones, not core design. | AWS, NTT, Yotta, CtrlS, STT GDC, Sify, AirTrunk, Adani. IT / ITeS Policy 2023; DC as essential service. |
| Tamil Nadu Chennai | ~268 MW | ~153 MW | ~273 MW planned | Colocation and hyperscale, anchored on subsea cable landings. Latency-led demand: OTT, CDN, cloud. | STT GDC, Sify, CtrlS, NTT, AdaniConneX, Yotta, Princeton Digital. DC Policy 2021; expiry reported April 2026. |
| Telangana Hyderabad | ~138 MW | Not disclosed | GPU-heavy AI campuses | Hyperscaler self-build plus AI-native and GPU colocation. Incentives aimed at high density and liquid cooling. | Microsoft, AWS, CtrlS, STT GDC, Sify, CapitaLand, AdaniConneX, AirTrunk. DC Policy 2016. |
| Andhra Pradesh Visakhapatnam | Minimal at scale | Google campus broke ground April 2026 | Largest announced pipeline in India; state ambition reported at 6 GW | Almost entirely hyperscaler self-build, AI-native. Very little leasable colocation is being created. | Google with AdaniConneX and Airtel; Sify; Reliance (reported). DC Policy 4.0; single-desk clearance. |
Announced capacity is not delivered capacity. Industry reporting suggests that across large multi-phase Indian campuses, less than 20 percent of announced capacity is live today, with commissioning phased to 2028 and beyond. In Andhra Pradesh the pipeline is almost entirely self-build: it will anchor the corridor without creating leasable capacity for the operators, global capability centres and enterprises that follow. That is the supply gap, and a Ready-to-Build site is what closes it.
Operational, under-construction and planned figures for Mumbai, Chennai and Hyderabad are drawn from a single industry dataset reported in May 2026, used together so the markets are comparable. Capacity estimates vary materially between published sources. Items marked reported are press-reported and not independently confirmed. Figures are indicative, as of mid-2026, and subject to verification.
From a tier-two mention to India's largest announced cluster in under two years.
| Project | Announced capacity | Status | Notes |
|---|---|---|---|
| Google AI hub, Visakhapatnam | 1 GW | Under construction; ground broken April 2026 | Approximately US$15B over 2026 to 2030, with AdaniConneX and Airtel. Includes a new international subsea gateway. |
| Reliance cluster (reported) | ~1.5 GW | Reported as approved by the state investment promotion committee, April 2026 | Reported at approximately US$17B, near Bhogapuram, with captive solar and storage. Press-reported. |
| Sify Technologies | ~550 MW | Approved | Also reported as a 50 MW first phase with a cable landing station. Reported figures differ by source. |
| AdaniConneX (earlier allotment) | ~200 MW | Land allotted 2022 | Part of a wider IT park and skill university allotment under the previous state government. |
| Digital Connexion; Meta with Sify | ~1 GW; ~500 MW | Reported, early stage | Press-reported only. Not confirmed, and shown to indicate direction of travel. |
This is an argument for more supply, not less. The pipeline is concentrated in a small number of very large, self-build campuses on multi-year timelines. It anchors the corridor, the subsea gateway and the grid — but it does not create leasable capacity for the operators, global capability centres and enterprises that will follow.
Project Andhra KendramSix constraints sit between an MoU and a commissioned hall.
Announcements are cheap. Megawatts are not. These bind in roughly this order.
Industry consultations rank power ahead of land. A hyperscale campus needs a dedicated 132 kV or 220 kV connection with redundant feeders — and a letter of comfort from a discom is not a bankable connection.
Contiguous, clean-title parcels at 25 to 900 acres are hard to assemble. Agricultural land needing industrial conversion adds an estimated 6 to 12 months on its own.
Facilities above 20 MW can require up to 30 separate clearances. Grid-connectivity approvals, building sanctions and fire NOCs remain live sources of delay, even in single-window states.
Large power transformer lead times moved from roughly 50 weeks in 2021 to about 120 weeks in 2024. Switchgear backlogs extend to 2028. Equipment, not construction, is increasingly the critical path.
Published figures are ultimate campus or IT-load targets. Delivery is phased over years, with commissioning in many cases running to 2028 and beyond. A 1 GW headline is a decade of build.
Operators build to an anchor commitment. Hyperscale blocks now run 25 to 50 MW, and AI deployments 75 to 100 MW. Without a signed anchor, phases are deferred.
Sources: industry consultations and published research on Indian data centre development, 2026. Equipment lead-time figures are global benchmarks. Figures are indicative.
How each state is closing the gap.
The same six constraints, and what each state has actually done about them. Policy is converging; what separates the states now is delivery on the ground.
| Constraint | Maharashtra | Tamil Nadu | Telangana | Andhra Pradesh |
|---|---|---|---|---|
| 1. Power | Power subsidies and electricity duty exemption. Strong grid reliability and competitive industrial tariffs. | Preferential tariff, electricity duty waiver, and priority allotment of renewable power. | Dual grid networks planned. Additional benefits where more than 50 percent of power is green. | Captive and open-access renewable route. Gigawatt-scale intake still to be built. |
| 2. Land | Earmarked land zones for faster allocation. Siting steered to coastal and peri-urban clusters. | State-established data centre parks. Land cost subsidy, and sub-lease permitted without transfer charges. | Subsidised land allocation and a 50 percent rebate on building fees. | Large APIIC allotments and a park-based, plug-and-play model. Reported 480 acres allotted for the Google campus. |
| 3. Approvals | Simplified approvals. Data centres recognised as an essential service. | Single-window clearance, and relaxed zoning and building norms. | Time-bound single window, essential-service status, and specified exemptions. | Single-desk clearance (UPIN), with APEDB as nodal agency. |
| 4. Long-lead equipment | Not addressed by state policy. | Not addressed by state policy. | Not addressed by state policy. | A global supply-chain constraint that no state incentive can shorten. |
| 5. Phasing and delivery | Deep operator base and repeat delivery. The most proven build environment. | Established park model with a track record of commissioned capacity. | Customised packages above investment thresholds, tied to milestones. | Delivery record still being built. The first mega campus broke ground in April 2026. |
| 6. Anchor demand | Deepest tenant market, including a reported US$8.2B AWS investment. | Subsea landings anchor latency-led demand from cloud, OTT and CDN. | Microsoft and AWS commitments, plus a T-Cloud sovereign framework. | The largest anchor in India: Google at approximately US$15B. Anchor secured; leasable supply is not. |
All four states now offer land subsidy, duty exemption and single-window clearance. Incentives are no longer a differentiator.
Equipment lead times, and a site that is actually ready to build. Andhra Pradesh's Data Centre Policy 4.0 targets 200 MW of additional capacity, against a pipeline now reported at 6 GW.
State incentive positions are drawn from published state data centre policies and industry summaries, 2026. Items described as reported are press-reported and not independently confirmed. Positions are indicative and subject to verification against the current notified policy in each state.
Capacity is only worth building if it is leased.
These are the demand segments that take Indian data centre capacity, and what each one needs.
Google, Microsoft, AWS, Oracle, Meta. Take capacity in very large blocks on long leases, or self-build. Anchor the market and set the standard.
Needs: scale, power certainty, low latency
GPU-cloud providers and AI platforms. The fastest-growing segment, and the reason density and cooling now drive design.
Needs: high density, power, liquid cooling
The captive technology and back-office centres global corporates run from India. Follow the hyperscalers into a cluster.
Needs: proximity to talent, reliable colocation
Banks, insurers and large corporates, driven by regulation, data localisation and disaster recovery. The core of Tier IV demand.
Needs: resilience, compliance, DR pairing
Content, streaming and delivery networks. Latency-sensitive, and the natural user of subsea cable landings.
Needs: subsea landings, peering, low latency
State and central workloads, sovereign cloud, and public digital infrastructure. A growing, policy-driven segment.
Needs: data residency, security, sovereignty
For Visakhapatnam: the hyperscaler is already committed and the subsea gateway is being built. The follow-on demand — GCCs, enterprise, BFSI and GPU-cloud — needs leasable, high-density capacity that the self-build campuses will not supply.
From ~375 MW in 2020 to ~1.5 GW in 2025 — and doubling again by 2031.
- National capacity 2025
- ~1.5 GW
- FII share of DC equity
- 80–86%
- Chennai live capacity
- ~200 MW
- Chennai CAGR
- >22%
Up from ~375 MW in 2020
Of $14.7B inflows 2020–2025
Targeted ~550 MW by 2030
2025–2030 forecast
Chennai — second-tier behind Mumbai, with 25%+ of new supply.
Chennai is institutionally validated by STT GDC, Sify, NTT, CtrlS, Equinix CN1, CapitaLand India Trust, AdaniConneX, Colt DCS, Nxtra, Digital Connexion and Yotta. The market is dominated by colocation scale-ups — leaving room for purpose-built, hyperscale AI platforms.
Where Vastra fits
Project Surya is a greenfield, phased 100 MW development positioned for hyperscale and AI workloads, with on-site battery storage and a high renewable profile — offering the State a new, purpose-built platform rather than incremental expansion of an existing one.
Johor: from 10 MW in 2021 to 1,500+ MW today.
Johor has moved from approximately 10 MW of capacity in 2021 to over 1,500 MW operational by 2024, with another 4 GW-plus under construction or planned. Sub-5 ms latency to Singapore makes it the natural cross-border availability zone for hyperscalers facing Singapore's land and power constraints.
- Total supply growth
- 10 → 5,800 MW
- Market vacancy
- 1.1%
- Wholesale rate
- US$110–140
- Latency to Singapore
- <5 ms
2021 to H1 2025
H1 2025
per kW per month
Capital is abundant. Differentiated platforms win.
Foreign institutional investors contributed ~86% of India DC equity inflows 2020–2025.
Platforms attract $200M–$5B+ tickets. Single-project plays — Surya's natural framing — go via co-development JV equity and anchor-led structures.
Major peers anchor to renewable PPAs. Vastra's BESS + renewable architecture is structured to fit sustainability-linked finance from day one.
Sources: Vestian, IMARC, Mordor Intelligence, JLL, ResearchAndMarkets and public company announcements. Figures indicative.