Global data centres are becoming hotter than ever before, not only due to heatwaves, but because thermal management is rapidly emerging as the defining constraint on AI growth. For decades, operators built facilities around predictable workloads and stable power densities. Artificial intelligence has shattered those assumptions.
The numbers tell the story plainly. Traditional air-cooling systems typically lose effectiveness beyond approximately 20–50 kilowatts per rack. But AI deployments are already pushing well past that threshold. Some of the latest AI systems are now operating at more than 200 kilowatts per rack, with designs already targeting 400 kilowatts and beyond.
With AI infrastructure reaching a fundamental physical limit, this has forced a complete architectural rethink of how data centres manage both power and heat.
From Air to Liquid
The industry's shift towards liquid cooling is no longer a matter of preference or forward planning. For facilities hosting modern AI workloads, it is a structural necessity. Liquid cooling accounted for roughly 46% of the data centre cooling market as of 2024, per research group Omdia, with air-based systems still dominant in legacy environments. That balance is shifting rapidly.
Data from Markets and Markets show the global data centre cooling market is expected to grow from roughly $11 billion in the mid-2020s to more than $20 billion by the early 2030s. Yet the scale of investment required goes far beyond upgrading a few components, as supply constraints slow down the AI data centres' delivery timelines.
“The next winners in AI infrastructure will not just be companies with GPUs, but those that can deliver liquid-ready AI capacity faster, denser, and more efficiently. This is a big opportunity as well as a challenging time ahead for top contenders like Vertiv Energy, Schneider Electric, Eaton, Johnson Controls, Boyd Corporation, etc,” Shantaram Shinde, Director Partner Business and Cloud Infrastructure at NTT Global Data, tells AIM.
He notes that accelerated compute clusters require liquid cooling for the GPUs, but they still need to retain air cooling for part of the system. “This hybrid reality is one of the most pressing practical complications facing operators today. Facilities cannot simply rip out their air-cooling infrastructure; they must manage the coexistence of two systems with fundamentally different engineering demands, maintenance requirements, and failure modes,” he adds.
The Integration Challenge
The transition to liquid cooling is not merely a matter of installing new equipment. It is a systems-integration problem that many organisations are still learning to navigate.
Divyansh Pal, Head of Strategic Development - Data Center, at Sky IT Infratech, observes that many enterprise data centres are not yet ready to move to large-scale liquid cooling.
“Fragmented component standards, leak risk, retrofit complexity, and installation requirements remain very real execution issues. This complexity is particularly acute for enterprise environments that built their infrastructure during an earlier era of compute,” he explains.
India finds itself at an unusual advantage. Unlike mature markets laden with legacy infrastructure, India's facilities can be built on the present requirements.
"India's data centre market is not hampered by legacy cooling systems, presenting a unique opportunity to build future-ready, high-density facilities from the ground up,” Dev Tyagi, President of UK, India and Asia at the Spanish immersion cooling firm Submer, which has committed to establishing a manufacturing hub in India to serve the entire Asian market.
Power, Heat, and the Unified System
One of the most consequential shifts in how advanced operators now think about data centre design is the collapse of the boundary between power delivery and thermal management. The two systems, historically treated as separate engineering domains, are now understood to be a single, tightly coupled challenge. If power delivery is constrained, compute capacity is limited; if thermal removal is constrained, systems can become unstable, or performance can degrade.
This has a direct bearing on the economics of AI infrastructure at scale, and Indian operators and investors are acutely aware of it. "After the upfront capital investment in servers and facilities, energy costs become the dominant ongoing expense for data centres, making even incremental efficiency gains highly valuable. If you can reduce energy costs by, call it, 10–30%, that's like a huge number,” notes Rajan Anandan, Managing Director of Peak XV Partners.
Peak XV's interest in this space led the firm to back C2i, an Indian startup founded by former Texas Instruments executives, which is targeting the energy losses that occur during power conversion inside data centres, losses that translate directly into heat that cooling systems must then manage. C2i estimates it can cut end-to-end losses by around 10%, roughly 100 kilowatts saved for every megawatt consumed, with knock-on effects for cooling costs, GPU utilisation, and overall data centre economics.
India's Cooling Imperative
India's data centre expansion is not simply a story of growth; it is increasingly a story of constraint. Sunil Gupta, Co-founder, CEO and Managing Director of Yotta Data Services, has argued that building hyperscale AI data centres in India means high-density racks, advanced cooling, and geographically distributed capacity.
“The demand for AI infrastructure in India has already outpaced early estimates, and timely access to infrastructure is now critical to sustain the momentum across the ecosystem. That momentum, however, brings its own complications: India's climate, combined with the sheer heat output of high-density AI racks, creates cooling demands that are qualitatively different from what a conventional enterprise data centre would face,” he told Business Today.
A 1 megawatt data centre using traditional cooling methods requires an estimated 26 million litres of water annually, equivalent to the domestic water needs of approximately 528 people each year. As India pushes towards several gigawatts of new data centre capacity, the water implications alone are becoming a significant policy and planning consideration, particularly in a country that holds 18% of the global population but only 4% of the world's water supply.
This is precisely why the shift to more advanced cooling approaches—liquid cooling, immersion cooling, and direct-to-chip systems—is not merely a technical preference for Indian operators. As Shinde observes, cooling architecture is becoming increasingly important for where projects can be developed and how quickly they can move.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.