India is emerging as a critical hub for global AI infrastructure, driven by government subsidies, data sovereignty concerns, and geopolitical advantages that are attracting enterprises from Europe and the Middle East to host their AI workloads there. With a population of 1.4 billion and over a billion smartphone users generating vast amounts of data, India is investing heavily in GPU infrastructure and sovereign AI capabilities, fundamentally reshaping how companies think about where to build and run their artificial intelligence systems .

Why Is India Becoming a Major Player in Global AI Infrastructure?

India's push for sovereign AI is driven by legitimate concerns about data privacy and security. Citizens and enterprises increasingly want their data processed and stored within national borders, rather than relying on foreign cloud providers. The Indian government has responded by launching the IndiaAI Mission, which heavily subsidizes computing costs for local model builders, researchers, and academic institutions .

The scale of this effort is substantial. Yotta Data Services, an Indian data center giant, and Gorilla Technology recently announced a landmark partnership to deploy thousands of graphics processing units (GPUs) across India. The agreement calls for deploying about 5,000 GPU cards in the first six months, with plans to eventually scale to 36,000 GPUs. Yotta will operate the infrastructure at its Navi Mumbai data center, offering GPU clusters, bare-metal GPUs, AI lab workstations, and AI model endpoints to enterprises and government customers .

"People want sovereign AI and sovereign models trained on sovereign data. That's a huge wave in India right now, supported fully by the government," said Sunil Gupta, co-founder and chief executive of Yotta Data Services.

Sunil Gupta, Co-founder and CEO, Yotta Data Services

Beyond domestic demand, India is attracting international attention. Due to GPU shortages elsewhere, enterprises from Europe and the Middle East are increasingly looking to India to host their AI training and inference workloads. The country's geopolitical stability compared to other regions makes it an attractive alternative for companies seeking to diversify their infrastructure footprint .

Will India's Growth Cannibalize Southeast Asian Tech Hubs?

Industry leaders are pushing back against the notion that India's infrastructure boom will replace established tech hubs like Singapore and Malaysia. Instead, they argue that India's massive scale will complement, not compete with, existing regional ecosystems .

"India is not here to replace anybody. India is here to help you build scale and velocity. It's here to show you that you can build these large-scale models, and you can be successful with an efficient cost base," stated Jay Chandan, chairman and CEO of Gorilla Technology.

Jay Chandan, Chairman and CEO, Gorilla Technology

The strategy centers on making enterprise AI commercially viable by reducing costs. Many enterprises have developed AI use cases across finance, media, entertainment, and manufacturing, but few have moved into production because the return on investment remains uncertain. By offering GPU infrastructure through a low-cost consumption model, providers aim to help companies cross the threshold into production within three to five years .

How Are Data Centers Managing AI's Massive Energy Demands?

The growth of AI infrastructure is creating unprecedented energy challenges. The International Energy Agency (IEA) predicts that energy demand from data centers will more than double by 2030, with electricity demand from AI-optimized data centers projected to more than quadruple by 2030 . This surge is already affecting residential electricity prices in the United States, where power demands from data centers are being directly blamed for price increases .

The energy intensity of AI workloads is driving infrastructure changes across the industry. Nvidia's roadmap assumes the 1-megawatt rack is not far away, and the company is championing a transition from 48V or 54V direct current at the rack to 800V direct current power for data centers. While this transition may ultimately lead to more efficient power use, it also requires a wider overhaul of data center infrastructure, including more powerful cooling systems, storage, and networking .

• Cooling Systems: Cooling equipment accounts for approximately 38% of an average data center's total energy consumption, making thermal management a critical efficiency challenge.
• Processing Hardware: Processors, chips, and storage hardware alone account for roughly 45% of a data center's total energy use, with AI GPUs becoming increasingly power-hungry.
• Supporting Infrastructure: Security systems, backup power supplies, power conditioning, and lighting make up the remaining portion of data center energy consumption.

Access to power is now as critical as access to GPUs themselves. Nscale, a European neocloud operator, has centered its data center network on Norway, where cold climate and abundant hydroelectric power offer distinct advantages for running power-hungry AI infrastructure. The power footprint of GPUs has driven a transition from air-cooled to liquid-cooled data centers, which has accelerated innovation cycles from two-year release cycles to six-month cycles .

What Energy Mix Is Powering Data Centers Today?

The current energy landscape for data centers reveals a complex picture. From September 2023 to August 2024, renewable energy provided 22% of all data center energy needs, while nuclear energy provided 21%. However, fossil fuel plants continue to account for roughly 56% of data center electricity, highlighting the ongoing reliance on carbon-intensive power sources .

As data center construction has surged over the past three years, developers face a critical choice: they can either plug into existing fossil fuel-based power grids or invest in alternative energy solutions. To reverse this dynamic, data center developers have multiple options available .

• Renewable Energy Siting: Building data centers in locations with abundant solar, wind, or hydroelectric power reduces carbon footprint and grid dependency from the outset.
• On-Site Generation: Installing solar panels and battery storage devices reduces reliance on grid power and carbon-intensive diesel generators used for backup power.
• Energy Efficiency Design: Modernizing infrastructure with software-defined systems and optimized workload management can reduce energy consumption by approximately 50% compared to legacy environments.

The scale of future demand is staggering. Globally, meeting data center electricity demand could require hundreds of gigawatts of additional capacity by the mid-2030s, particularly as large hyperscale facilities continue to expand. This will require a combination of solar, wind, battery storage, next-generation geothermal, and nuclear power to meet demand while maintaining grid reliability .

How Can Enterprises Balance AI Growth With Energy Efficiency?

Enterprises face a fundamental tension: they want to build out AI capabilities while protecting their sustainability reputations and avoiding community backlash against data center developments. The solution lies in treating efficiency as a foundational design principle rather than an afterthought .

One approach involves leveraging specialized hardware and software optimization. Vast Data's Bluefield 4 Smart Network Interface Cards (NICs) can reduce infrastructure power consumption by approximately 75% by eliminating the need for additional physical servers. For every 1,100 GPUs deployed, organizations can avoid deploying an additional 256 physical servers, resulting in substantial cost and power savings .

Another strategy involves running smaller, specialized models locally rather than relying exclusively on hyperscaler infrastructure. Enterprises downloading and running open-source AI internally can switch systems off when not in use, reducing overall energy consumption. Additionally, specialized models are significantly more efficient than general-purpose models like ChatGPT, offering a path to lower energy intensity .

"Software-driven optimisation is critical to ensure compute is fully utilised and energy isn't wasted through idle or over-provisioned infrastructure," explained Karim Abou Zahab, principal for sustainable transformation at HPE.

Karim Abou Zahab, Principal for Sustainable Transformation, HPE

The drive for data sovereignty, particularly in regions like India, will also shape energy demand patterns. As enterprises increasingly require data to remain within national borders, distributed infrastructure and specialized models become more economically viable, potentially reducing reliance on centralized hyperscaler data centers and their associated energy footprints .

It is important to note that enterprise AI workloads currently represent only a fraction of total cloud and data center energy consumption. As of 2024, AI was responsible for approximately 15% of data center energy demand, with most demand still coming from standard compute workloads. However, inferencing energy use is projected to almost double by 2030, reaching 162.5 terawatt-hours, creating both a challenge and an opportunity to prioritize efficiency from design through deployment .