The great compute buildout: Can data centres power India’s AI leap?

When OpenAI launched ChatGPT in November 2022, the artificial intelligence (AI) model took the world by storm. Since then a slew of other companies have launched their own AI models, including Anthropic with Claude, Google with Gemini, and xAI with Grok. And that has led users to lean on AI to search, learn and work.

Indians have been at the forefront in adopting AI. ChatGPT alone has more than 100 million users in India, or around 10% of its global user base. Indeed, government thinktank Niti Aayog estimates that AI adoption could unlock approximately $1 trillion in economic value for India by 2035.

But while Indians are among the world’s biggest AI users, much of the computing infrastructure powering this digital activity sits outside the country. India is among the world’s largest consumers of digital services, but a relatively small host of the computing power that sustains them. Specifically, the country generates nearly 20% of the world’s data but accounts for just 3% of global data-centre capacity, according to Boston Consulting Group (BCG).

That is beginning to change gradually. From Visakhapatnam to Mumbai, Hyderabad and Chennai, a construction boom is gathering momentum. More than $100 billion in investments have been committed across hyperscalers, data-centre operators and Indian conglomerates.

In August, Microsoft commissioned a data centre the size of two Eden Garden cricket stadiums in Hyderabad. Around 600 km away in Visakhapatnam, Google is building its AI hub. Another hyperscaler, Amazon Web Services (AWS), is expanding its data centre capacity across Mumbai and Hyderabad.

Local companies are also building campuses designed to answer AI prompts. The expansion is moving beyond traditional technology centres, with projects emerging in places such as Patna and Bodh Gaya.

India’s data-centre capacity, estimated at 1.5GW in 2025, could reach 6–8GW by 2030 and 12–14GW by 2035, according to BCG. The operational data-centre capacity has already expanded from about 375MW in 2020 to nearly 1.8GW by the first half of 2026, according to reality services firm Cushman & Wakefield.

“India’s objective is to move from being the world’s largest user of digital services to owning the compute layer beneath them,” says Sudhanshu Chawla, managing director and partner, BCG. The consultancy expects capacity to grow at an annual rate of 30–40% through the end of the decade and generate $140–200 billion in economic output.

The Reserve Bank of India mandates that all financial data be hosted locally and the Digital Personal Data Protection (DPDP) Act mandates any data related to identity, location tracking, and online user logs be stored in India, laying the ground for the growth of local centres.

Data centres have been granted infrastructure status, improving access to long-term institutional financing. Union Budget 2026-27 also introduced a tax holiday extending to 2047 for eligible foreign cloud providers using notified Indian data-centre infrastructure to serve global customers.

These sops are designed to attract not just capital but the computing workloads that could establish India as an international digital infrastructure hub. Moreover, says Union minister of state for science and technology Jitendra Singh, the growth in data centres could generate “nearly one lakh specialized engineering jobs”.

But will India merely become a vast hosting location for AI technologies developed elsewhere, or can the infrastructure help it build an AI industry of its own? And should it be trying to match the leaders or focus on harnessing AI to tackle the country’s unique and myriad challenges?

Playing catchup

For decades, India’s technology advantage rested largely on software services and a vast pool of engineering talent. The emerging AI economy is changing that equation. Developing and deploying AI increasingly requires access to enormous computing resources, sophisticated chips, reliable power and high-speed networks. AI data centres carry that processing power.

To be sure, data centres themselves are not new. They have been around for decades. Legacy data centres host email, payroll, business software like ERP, websites, banking operations, securities transactions and so on. But these traditional data centres cannot run AI workloads and hence the need for AI data centres.

Explaining the difference, Sridhar Pinnapureddy, founder and chief executive of Hyderabad-based CtrlS Datacenters, says legacy data centres require 5–10 kw of power per rack, and use CPU (central processing unit) chips. AI processing, in comparison, is far more compute-intensive and needs graphics processing units (GPUs) that consume 100–250 kw of power per rack to run GPT models, AI models, autonomous cars, improve drug discovery, financial modelling and so on.

The government’s IndiaAI Mission has onboarded more than 40,000 GPUs, expanding access to subsidised computing resources. But India’s AI processing capacity remains small compared with the infrastructure available to leading global laboratories. For instance, Harvard University alone has over 1,000 GPUs. The US has around 10 million GPUs, while China’s GPU density is more than 10 times that of India.

India’s data-centre boom could help narrow that gap. For startups, universities and enterprises, local computing infrastructure can reduce dependence on overseas facilities, reduce latency, improve access to computing resources and make some AI applications cheaper and faster to operate.

The new facilities that are coming up are not simply larger versions of traditional server farms; they are being engineered to support a much wider range of computing-intensive applications. “As AI moves from pilots to real world inferencing and production deployments, companies need infrastructure that brings compute, data and users closer together while meeting data residency and compliance requirements,” says Manoj Paul, managing director, India, at Equinix, a California-based data-centre and interconnection services company.

“AI is fundamentally changing how data centres are designed and operated,” says Pinnapureddy. CtrlS is designing facilities capable of supporting rack densities of up to 250 kW. CapitaLand’s Navi Mumbai facility has also been designed for high-density AI computing.

Google expects its Visakhapatnam campus to accommodate GPUs, its proprietary TPUs, conventional CPUs and networking equipment, reflecting the different requirements of AI training, inference and everyday cloud services. “It’s really the foundational layer of the AI economy that is coming up in India. It will help India move from AI pilots to population level impact,” says Alexander Smith, principal, global infrastructure and energy, Asia Pacific, Google.

Who’s doing what

Google’s upcoming AI hub in Visakhapatnam is part of a $15 billion investment announced by the company in 2025. Unlike its existing cloud regions in Mumbai and Delhi, which operate with domestic partners, Visakhapatnam will be Google’s first fully owned and operated data-centre facility in India.

Microsoft has committed $20.5 billion, its largest investment anywhere in Asia to build AI capabilities across India. While AWS plans to invest over $21 billion into AI and cloud facilities between 2026 and 2030.

The projects will bring together subsea connectivity, advanced computing infrastructure and AI capabilities. “What Visakhapatnam is doing in this project is starting to build a third subsea gateway for India, which will offer the country additional resiliency and connectivity,” says Smith.

The new gateway will connect to terrestrial fibre networks reaching metros and smaller cities. It will also strengthen India’s connections with Southeast Asia, Australia and other international markets. Designed primarily to meet Indian demand, Visakhapatnam will become an important regional node in Google’s global network.

Microsoft’s newest cloud region, India South Central, in Hyderabad, will serve domestic customers while functioning as a strategic hub for Asia and the Global South. Its current customers include the Adani Group, Bajaj Finserv, HDFC Bank and PB Pay.

Pointing to the shift from digital public infrastructure to AI, Puneet Chandok, president, Microsoft India and South Asia, says, “India built the rails (UPI, Aadhaar) for a billion people to transact; we are now helping build the rails for them to work and create with AI.”

AWS’ $21 billion India AI investment announced in July is on top of the more than $9 billion invested since establishing its first Indian cloud region in Mumbai in 2016.

“By giving Indian organizations local access to custom AI chips, and managed AI services, our data centres lower the barrier for startups, enterprises and governments to build and scale AI,” says Anurag Khilnani, director, data centre delivery, APAC, AWS. Users include HDFC Securities, Physics Wallah, Care Health Insurance, and DigiLocker.

Indian operators are investing aggressively, too. Hyderabad-based CtrlS Datacenters operates 18 facilities across nine markets and is developing new capacity in Mumbai and Hyderabad. It has also inaugurated a ₹400-crore facility in Patna and announced plans for a ₹10,000-crore, 150 mw AI data centre in Bodh Gaya.

CapitaLand India Trust India (CLINT) has committed more than ₹5,000 crore to three data centres in Navi Mumbai, Hyderabad and Chennai, with approximately 200 mw of power capacity. The Navi Mumbai tower was handed over to an unidentified global hyperscaler in July, while the second has been pre-leased. “We are in active leasing discussions with both global and domestic customers,” says Gauri Shankar Nagabhushanam, CEO, CLINT.

Equinix has invested more than $365 million in India and operates five facilities across Mumbai and Chennai.

The new data centres are eyeing business from the domestic market. Banks, insurance companies, e-commerce platforms, hospital chains, startups and government agencies are generating vast quantities of data and increasingly deploying AI-driven applications. These uses include AI-powered voice bots, fraud detection, reading medical images and data such as X-rays, ECGs, doing research and sales forecasts. “India will become a global node in Google’s network and the global internet, more broadly,” says Smith.

The opportunity

With a $100 billion investment, the scale of the data centre buildout is unprecedented. Yet the distinction between hosting AI and creating AI remains critical. A data centre owned by a global technology company may serve Indian customers while much of the underlying technology, intellectual property and commercial value remains overseas.

To capture more of that value, India will need to build on its expanding infrastructure with stronger domestic research, affordable access to indigenous models, specialized talent and commercially viable AI applications.

Currently, India lags way behind the US and China in developing frontier AI models. It has not yet produced a globally leading frontier model comparable to those developed by the largest American AI companies, including OpenAI, Anthropic, Google DeepMind, Microsoft Co-Pilot and China’s DeepSeek. Training such models requires vast clusters of specialized chips, billions of dollars in investment and access to advanced research capabilities, all of which India currently lacks.

But the opportunity for India may lie less in trying to build large AI models and more in applying AI to tackle problems on a national scale, in areas such as healthcare, education, agriculture, manufacturing and public services.

Consider healthcare; Google has worked with Aravind Eye Hospital to deploy AI tools that can detect diabetic retinopathy, an eye condition that can cause blindness if left untreated. The technology demonstrates how AI can help doctors identify patients who need treatment.

But deploying such capabilities across hundreds of hospitals requires far more than a pilot. It needs reliable computing infrastructure (data centres), secure access to medical data, affordable processing and the ability to serve large numbers of patients simultaneously. “The breakthrough exists, the technology exists now because of AI. The question is: well, how do you scale that,” says Smith.

There are other challenges to overcome in building models that can work nationwide. For instance, India has 22 scheduled languages, hundreds of regional variations and a population with many polyglots, especially in the South. Building AI systems capable of understanding this diversity and absorbing linguistic nuances is considerably more complicated than translating English text into another language.

Users often combine Hindi and English, or mix regional languages with English, in a single sentence. They also rely heavily on voice commands, images and visual searches. Moreover, some languages, including Sanskrit, Manipuri, Dogri, Konkani, and Sindhi, lack sufficient digitised training material. Developers will need to generate synthetic data and train models to understand local expressions, accents and code-switching.

This is where smaller, specialised AI models could become important. Unlike massive frontier models, small language models (SLMs) require relatively modest computing resources and can be customised for specific languages or industries.

They could power voice assistants for farmers, regional-language educational tools, banking services or government applications. Training and fine-tuning smaller AI models requires less computing power than that needed by large models, while running them close to users can reduce latency from seconds to milliseconds.

The advantage is significant: India may not need to develop the largest foundational model to build a thriving domestic AI industry. It could instead become a developer of specialised AI applications, for agriculture and weather intelligence, diagnosing tuberculosis, micro-lending, among others. These applications, all designed for the local market, can then be exported to other emerging economies.

India has already demonstrated that it can use digital infrastructure to transform the delivery of financial and public services. AI now presents an opportunity to repeat that transformation across a much wider range of industries.

Key Takeaways

~20% | India’s share in the total amount of data generated globally, according to a study by the Boston Consulting Group.

3% | The country’s share in data-centre capacity worldwide, according to BCG. The US is the leader, accounting for more than 60%.

>40,000 | Number of GPUs onboarded by the government’s IndiaAI Mission. The US has around 10 million; China’s density is over 10 times that of India.