Now multiply that by thousands of racks, stack them across new AI campuses coming up in Mumbai, Hyderabad, Pune, and Visakhapatnam - and you start to see the real bottleneck standing between India and its AI superpower dreams. It isn't chips. It isn't talent. It's heat.

India is in the middle of its largest data center build-out in history. The colocation market alone is forecast to add $7.31 billion in 2026, with operators targeting up to 15 GW of capacity within five years, up from under 2 GW today. But here's the catch nobody talks about enough: most of that legacy capacity was built for CPUs and air conditioners - not for GPUs that run hot enough to melt the old playbook.

The Heat Problem Is Bigger Than Most People Realize

Traditional data centers were designed around 5-15 kW per rack. A modern AI rack running NVIDIA's GB200 NVL72 hits 120 kW+ per rack, and the upcoming Vera Rubin and GB300 generations are targeting 150-600 kW per rack. That's an 8x to 40x jump in heat density in just a few years.

Individual GPUs are part of the story too. A single NVIDIA H100 draws 700W, the B200/GB200 hits around 1,000-1,200W, and the upcoming GB300 is projected at roughly 1,400W per module. The industry consensus is blunt: once GPU power crosses 700W, air cooling is no longer technically viable. Above 35 kW per rack, direct-to-chip liquid cooling becomes mandatory; above 100 kW, immersion cooling is the only way to keep PUE under control.

This isn't a future problem - it's already here. IIT Bombay and Vertiv recently launched a joint R&D project specifically because GPU power consumption jumped from 700W in 2022 to an expected 1,200W in 2025, with 54% of Indian businesses already adopting AI in some form .

Why This Is Make-or-Break for India Specifically

1. The power math doesn't work without efficient cooling.
Cooling alone can consume 30-40% of a data center's total electricity in air-cooled facilities. India's IT load capacity, currently around 1.4 GW, is projected to surge to 6.5-17 GW by 2030. On an already power-constrained grid, every percentage point saved on cooling overhead translates directly into more compute capacity for actual AI workloads.

2. The economics of cooling are now a board-level decision.
PUE (Power Usage Effectiveness - basically "how much extra power you burn just to cool your servers") tells the story:

• Traditional air cooling: PUE 1.5-1.8
• Air with economizers: PUE 1.3-1.5
• Direct-to-chip liquid cooling: PUE 1.1-1.25
• Immersion cooling: PUE as low as 1.02-1.05

For an operator running thousands of GPUs, the gap between 1.5 and 1.1 isn't a rounding error - it's the difference between profitable AI infrastructure and an electricity bill that eats the margin alive.

3. Liquid cooling is now its own investment category.
India's data center capacity has quadrupled since 2020 to about 1.5 GW, with cooling investments pegged at roughly $1.3 billion per GW of new capacity, part of an overall $2-2.5 billion thermal management opportunity. Global players like Vertiv and Schneider Electric, alongside Indian innovators like Refroid Technologies and Uravu Labs, are racing to fill this gap.

4. AI workload growth is structurally tied to cooling capacity.
Analysts expect AI workloads to drive at least 75% of all future data center growth in India. Globally, inference already accounts for 80-90% of total AI computing and is expected to drive 75% of total AI energy demand by 2030. Inference runs 24/7, generating sustained heat loads air conditioning simply can't sustain economically.

What "Doing It Right" Looks Like

The good news: this is a solved engineering problem, just not yet solved at India's scale. Globally, facilities running Direct-to-Chip (D2C) liquid cooling - where coolant runs through cold plates attached directly to GPU dies - achieve PUEs as low as 1.09, even in challenging climates. For India, where ambient temperatures in cities like Hyderabad and Visakhapatnam regularly exceed 35°C, the case for liquid cooling isn't theoretical - it's existential. Free-air cooling, which works well in cooler climates, simply isn't an option for most of the year across Indian geographies.

This is exactly why new AI-ready campuses are being designed from the ground up around liquid cooling, high-density rack architectures (100kW+ per rack), and renewable-power-paired infrastructure - rather than retrofitting old air-cooled shells.

The Bottom Line

India's AI story is being written right now - in new GPU clusters, sovereign AI models, and enterprise adoption already at 54% and climbing. But none of that compute matters if racks have to be throttled from overheating, or operating costs balloon because cooling eats 40% of the power bill. A Liquid Cooled Data Center is no longer just an advanced infrastructure option; it is becoming essential for supporting high-density AI workloads efficiently. Liquid cooling isn't a sustainability checkbox anymore - it's the foundational layer determining whether India's data centers can run the GPUs that power its AI ambitions, economically.

Data Center Cooling System

That the contents of third-party articles/blogs published here on the website, and the interpretation of all information in the article/blogs such as data, maps, numbers, opinions etc. displayed in the article/blogs and views or the opinions expressed within the content are solely of the author's; and do not reflect the opinions and beliefs of NASSCOM or its affiliates in any manner. NASSCOM does not take any liability w.r.t. content in any manner and will not be liable in any manner whatsoever for any kind of liability arising out of any act, error or omission. The contents of third-party article/blogs published, are provided solely as convenience; and the presence of these articles/blogs should not, under any circumstances, be considered as an endorsement of the contents by NASSCOM in any manner; and if you chose to access these articles/blogs , you do so at your own risk.

CTO

Cyfuture Cloud is a cutting-edge cloud infrastructure and service platform delivering next-gen computing solutions for businesses, researchers, and developers. Specializing in Cloud Hosting, we offer highly scalable, secure, and performance-optimized environments tailored for modern workloads. Our platform empowers innovation with a comprehensive suite of services, including AI as a Service, GPU as a Service, Inferencing as a Service, and Fine-Tuning capabilities, enabling faster AI model development, training, and deployment. Whether you're building intelligent applications or running complex simulations, our robust infrastructure backed by NVIDIA-powered clusters ensures seamless scalability and performance. With our IDE Lab Service, users can access pre-configured development environments in the cloud to streamline coding, testing, and deployment, all within a collaborative, secure setup.