NVIDIA’s Warm-Water Cooling Vision Meets India’s Extreme Heat

The race to build AI infrastructure has stumbled upon a cooling problem.

All this while, data centres have relied on conventional air-conditioning systems. But AI is changing the rules. Modern AI clusters powered by advanced GPUs are consuming unprecedented amounts of power, generating heat densities that conventional cooling systems struggle to manage.

That’s where liquid cooling comes in, absorbing heat from hot server chips. However, the coolant for such direct-to-chip liquid cooling needs to be ultra-cold, making it costly for data centres in hot climate regions to implement liquid cooling systems.

NVIDIA believes warm-water liquid cooling is the answer. By circulating water directly through cooling plates attached to GPUs and CPUs, the company says operators can dramatically cut energy consumption while supporting increasingly dense AI workloads. For operators preparing for the next generation of AI clusters, liquid cooling is increasingly becoming a requirement rather than a choice.

Yet, as India prepares for a wave of data centre investments, experts say the real question is not whether warm-water cooling works, but whether it can work efficiently in one of the world's hottest and most resource-constrained operating environments.

Why AI is Forcing a Cooling Reset

Traditional enterprise data centres were built around rack densities of roughly 10–20 kW. AI has pushed those limits far beyond their original design assumptions.

"The transition will be gradual," says Thiru Prakassh, Regional Data Centre Lead, APAC, at French cloud computing company OVHcloud. Existing facilities will continue serving enterprise applications and cloud workloads, but AI infrastructure is creating new thermal challenges.

"Rack power densities that once sat at around 20 kW are now reaching 135 kW in hyperscale AI environments—thermal loads that air cooling was simply not designed to handle," he says.

Instead of cooling entire server rooms, NVIDIA’s approach enables warm water to flow through cold plates attached to processors and accelerators. Because water transfers heat far more efficiently than air, the chipmaker says operators can reduce their dependence on energy-intensive chillers, which traditionally account for a significant portion of a facility's electricity consumption.

The company has demonstrated architectures capable of operating with coolant temperatures approaching 45°C, arguing that higher coolant temperatures can substantially improve overall energy efficiency.

Can Warm-Water Cooling Survive India?

While the technology offers compelling efficiency gains, India presents challenges that many Western markets do not.

"The real question isn't whether warm-water cooling can work in India, it's whether we're designing it for Indian conditions," says Vijay Sampathkumar, Chief Business Officer at cooling solutions company Refroid Technologies.

According to him, cooling can account for up to 40% of a data centre's electricity consumption, while even small increases in coolant temperatures can reduce cooling energy requirements.

However, Sampathkumar argues that India's climate demands a different approach.

"High ambient temperatures, varying humidity levels, and regional water constraints mean that cooling strategies developed for North America or Europe cannot simply be replicated here," he says. "Instead, deployments need to combine warm-water cooling with intelligent thermal controls, efficient heat rejection systems, and climate-specific design principles."

That challenge becomes particularly visible during peak summer months. Cities such as Delhi, Hyderabad, and Ahmedabad routinely experience temperatures above 45°C. As ambient temperatures rise, efficiently rejecting heat from cooling systems becomes increasingly difficult.

Sampathkumar believes the opportunity extends beyond merely importing foreign technology.

"India doesn't need to reinvent liquid cooling, but we need to reinvent how liquid cooling is deployed," he says. "India's opportunity is not just to adopt global technologies, but to develop cooling solutions that can serve emerging markets facing similar climatic and infrastructure challenges."

Water: Risk or Red Herring?

As liquid cooling gains traction, there are fears of water stress in areas where data centres are located. According to the World Resources Institute, over 50% of India’s data centres are built in high-water-stress regions. 

According to the Council on Energy, Environment and Water, water consumption in data centres was 150 billion litres in 2025. This is set to double by 2030.

Industry experts, however, say the reality is more nuanced.

"The industry's conversation has largely focused on power efficiency, but water efficiency deserves equal attention as AI infrastructure scales," says Sampathkumar.

He argues that modern direct-to-chip cooling systems largely operate through closed-loop systems where coolant is continuously recirculated.

"There is a common misconception that liquid cooling automatically leads to higher water consumption," he says. "The bigger question is how efficiently the overall cooling ecosystem manages heat rejection while optimising water usage."

Anwesha Sen, Assistant Programme Manager at Takshashila Institution, shares a similar assessment.

"Based on what we're hearing from various data centre operators in India, a significant number of data centres use closed-loop systems, which implies a one-time water consumption (on a significant scale)," she says. "If that is the case, it seems unlikely that water would be a bigger constraint than electricity for Indian data centres."

However, Sen believes a critical problem remains unresolved.

"What is crucial is transparency around water use for individual data centres," she says. "There are a bunch of numbers on company-wide water consumption, but they don't do much when trying to gauge water consumption by data centres in specific areas, particularly in water-stressed regions."

Sampathkumar argues that operators must begin treating water efficiency with the same importance as energy efficiency.

"As AI deployments become denser, water efficiency will become a boardroom metric, not just an engineering metric," he says. "The operators that succeed will be those that design for both PUE (power usage effectiveness) and WUE (water usage effectiveness) from the outset."

The Economics is Shifting

Today, most organisations remain focused on securing access to scarce GPU resources. But once AI workloads move into production, operational efficiency quickly becomes part of the discussion.

"Power and cooling become a significant part of the total cost of ownership," Sampathkumar says. "Customers want to understand what their infrastructure will cost to run over the next three to five years."

Sustainability is also becoming a procurement consideration. "Enterprises are increasingly evaluating cloud providers not only on performance and price, but also on their energy and water footprint," says Prakassh. 

In India, he adds, "cooling efficiency is becoming a key differentiator."

A Policy Vacuum 

The debate around cooling comes as India prepares for an unprecedented expansion of digital infrastructure. Estimates suggest that between 7 GW and 10 GW of additional data centre capacity could be developed over the next five years. 

For Sen, technology is only part of the solution. The bigger challenge is regulation. She argues that policy intervention is essential for managing resource allocation and guiding future development.

"This is essential, and in the absence of a national policy that sets benchmarks, we see states competing to provide huge incentives to hyperscalers, which is a risk to maintaining environmental safeguards," Sen adds.

Warm-water cooling may ultimately become the foundation of AI infrastructure worldwide. But in India, its success will depend on more than engineering. It will require climate-adaptive design, transparency around resource use and policy frameworks that balance rapid digital growth with environmental sustainability.