India's Cloud Adoption in 2026: AWS, Azure, GCP, and the Sovereign Cloud Question

Indian cloud spend was estimated at roughly $13B in 2025 and is forecast to cross $20B in 2026. The growth rate — high-teens percentage annually — is among the highest in any major economy. But the headline numbers obscure a more interesting set of structural changes: the relative shares of the hyperscalers are shifting, the sovereign-cloud question has become real, and the regional buildout — particularly in Mumbai, Hyderabad, Pune, and the new Telangana data center cluster — is changing what is operationally possible.

I want to write a buyer-side perspective on what the 2026 Indian cloud landscape actually looks like, based on procurement work we have done for clients in BFSI, healthcare, and enterprise.

Market shares — the rough shape#

Reliable share data is hard to come by because the hyperscalers do not break out India revenue cleanly. Triangulating from public reports, ISV partner reporting, and our own observations across procurement engagements:

AWS is the share leader, with somewhere between 40-48% depending on whose estimate you trust. Strongest in startups, fintech, BFSI’s more cloud-forward segments, and the digital-native enterprise. Three regions — Mumbai and Hyderabad as full regions, plus the GovCloud variant — give them the most flexibility.

Microsoft Azure is the close number two, with 28-34% share. Strongest in large enterprise where the Microsoft enterprise agreement gives them a hook (Office 365, then Azure adoption follows), in the public sector where Microsoft’s government relationships are deep, and increasingly in AI workloads where Azure OpenAI Service has been the easiest path to GPT models. Two regions (Pune, Chennai) plus a third planned (Hyderabad).

Google Cloud Platform is the smaller third, with 12-18% share. Strongest in data and AI-led shops, in startups that are deeply Google-Workspace-anchored, and in advertising and media. Two regions (Mumbai and Delhi NCR).

Oracle Cloud Infrastructure has a meaningful share in BFSI specifically — Oracle’s database installed base translates into OCI for the workloads that want to stay on Oracle. Around 5-8% overall but disproportionately in finance.

Domestic and sovereign clouds — Yotta, CtrlS, ESDS, Tata Communications’ cloud, JioCloud — collectively account for the remainder, with strongest position in regulated sectors (defense, parts of BFSI), in the smaller-than-hyperscaler enterprise tier, and in workloads where the sovereign-cloud narrative matters.

Regional buildout and what it enables#

Cloud is regional infrastructure. The buildout matters because it determines what is operationally possible.

Mumbai (ap-south-1 on AWS, Central India on Azure, asia-south1 on GCP) remains the primary region for all three hyperscalers. Cross-AZ latency is sub-millisecond intra-region. Most production workloads serving Indian customers anchor here.

Hyderabad (ap-south-2 on AWS, planned Azure region) has emerged as the secondary region, with the Telangana state’s aggressive data-center incentives drawing significant build-out. The pair of regions (Mumbai + Hyderabad) enables true multi-region production architectures for the first time at hyperscaler quality.

Delhi NCR has GCP’s secondary Indian region and an Oracle region; AWS does not have a presence here at the time of writing.

Pune has Azure’s Central India region (somewhat confusingly named — it’s actually western-central, not central).

Chennai has an Azure region; AWS does not.

The practical implication: serving customers across the country with low latency requires careful placement. Mumbai gives you Maharashtra, Goa, the western coast, and (with some latency) the rest. Hyderabad gives you the south. Delhi NCR gives you the north. A two-region architecture (Mumbai + one secondary) is now the default for production workloads.

The data residency requirement#

A growing share of cloud architecture decisions in India are driven by data residency requirements, both regulatory and contractual.

RBI has long required certain financial data to be stored within India. The cloud-data-residency clarifications from 2017 and updates since have specified that hyperscaler India regions count, provided the data is not replicated or processed outside.

SEBI has similar requirements for capital markets data.

ABDM-flowed health data must remain in India.

Government data under various procurement contracts must remain in India and often must use an empanelled cloud provider — the MeitY empanelment list.

DPDPA does not currently impose blanket data localization, but the framework allows future country-restriction notifications.

The compounding effect is that any meaningful enterprise architecture in India now has data-residency constraints. Hyperscaler India regions handle this for most cases. For the most-restricted workloads — government, defense, specific BFSI — only the domestic and sovereign clouds may qualify.

The sovereign-cloud push#

The “sovereign cloud” narrative has strengthened materially in 2024-2026, driven by several factors: the geopolitical anxiety about US-hosted infrastructure, the desire to support domestic industry, and specific procurement preferences in government and regulated sectors.

Yotta’s Shakti GPU Cloud has been the most visible domestic infrastructure story, with thousands of NVIDIA GPUs and a credible cloud service offering. Their pitch to enterprises has been India-hosted, India-sovereign, India-compliant — a real value proposition for the right buyers.

CtrlS, NxtGen, ESDS are well-established domestic IaaS providers with reasonable enterprise traction.

JioCloud (Reliance’s cloud) is a wildcard. Reliance has been investing heavily, has the customer base (Jio’s enterprise SIM business), and has signaled intent to compete with the hyperscalers in select workloads.

The MeitY empanelled cloud list — the official list of cloud service providers approved for government procurement — has slowly expanded to include the hyperscalers’ India regions alongside domestic players, but the empanelment process still favors domestic providers for the most-sensitive workloads.

The honest assessment: for most enterprise workloads, the hyperscalers’ India regions are operationally and economically superior, even after accounting for the sovereign-cloud premium some buyers place. For the workloads where sovereignty matters — defense, parts of government, certain BFSI sub-segments — the domestic providers have a real niche.

The cost shape#

A few patterns worth knowing.

Reserved instance / savings plan discounts are particularly generous in India for the hyperscalers — usually 25-50% off list for one-year commitments, more for three-year. The marketplace credit programs (especially AWS Activate and Microsoft for Startups) are aggressive for early-stage companies.

Egress costs remain the hidden cost driver. Cross-region replication, multi-cloud architectures, and CDN strategies all need to be designed with the egress meter in mind. We have seen workloads where the egress bill exceeded the compute bill.

GPU availability has been variable. The H100 and H200 SKUs have been periodically constrained on all three hyperscalers in Mumbai region. Hyderabad has somewhat better availability. For sustained GPU workloads, Yotta’s GPU cloud has been the better economics in 2025-2026, though the operational tooling is less mature than the hyperscalers’.

ISV billing through marketplace has matured significantly. Buying Databricks, Snowflake, Confluent, or MongoDB through your cloud marketplace and counting it against your commitment is now standard practice and produces meaningful effective discounts on those tools.

Procurement patterns we recommend#

For an Indian enterprise doing a cloud-procurement program in 2026:

1. Pick a primary hyperscaler based on workload fit, not on theoretical multi-cloud. Real multi-cloud is operationally expensive and produces inferior outcomes for most teams. Pick one and be excellent on it; if a specific workload needs a second cloud, do that for that workload.

2. Lock the commitment carefully. Three-year commitments unlock the deepest discounts but commit you to a forecast you may not hit. We usually recommend a one-year primary commit with a savings-plan layer rather than a three-year reserved-instance commit.

3. Plan for data residency from the architecture stage, not at audit time. Tag every data store with its residency class. Use IAM policies and Terraform constraints to enforce.

4. Treat the sovereign-cloud question as workload-specific, not architecture-wide. A handful of workloads may justify Yotta or a domestic provider; the rest belong on a hyperscaler.

5. Use the marketplace. Buying Databricks, MongoDB Atlas, Confluent Cloud through AWS Marketplace and counting against your AWS commit is a free 5-15% effective discount on those tools.

6. Monitor egress aggressively. Set up daily cost-anomaly alerts. We have caught misconfigured backup replications, runaway log shipping, and oversized CDN cache misses on multiple client engagements.

What’s coming in 2026 and 2027#

Three things to watch:

The third hyperscaler region announcements have been hinted by both AWS and Azure. If they land, multi-region architectures become more flexible.

The sovereign-AI mission’s expansion may compel certain AI workloads to use domestic compute. The boundary of “AI workload that must run on Indian sovereign cloud” is being negotiated.

The Reliance/JioCloud trajectory is the wildcard. If they ship at scale, the market dynamics shift.

Where pdpspectra fits#

Our cloud engineering team runs cloud architecture and migration programs for clients in India and internationally. We are hyperscaler-pragmatic, not partisan — we design for the workload, not for the vendor relationship — and we do the regulatory architecture as part of the technical design.

Related reading: the Snowflake vs Databricks vs BigQuery post, the AWS Bedrock vs OpenAI vs Anthropic post, and the multi-cloud strategy post.

---

Cloud in India is no longer “pick one and forget it.” Talk to our team about your architecture.