AI-driven workloads are pushing data centres to handle unprecedented levels of computing power, while rising GPU densities create new challenges around heat, power consumption and infrastructure efficiency. The rapid growth of AI is forcing organisations to rethink conventional cooling systems and explore infrastructure that is capable of supporting higher-density computing requirements.
As data centres become more energy-intensive, businesses are also focusing on sustainability, power management, PUE, WUE, water utilisation and the efficient use of expensive data centre space. Liquid cooling is increasingly emerging as an infrastructure solution that can address these requirements while supporting higher rack densities and improving overall resource utilisation.
Priya Dialani from Analytics Insight Podcast spoke with Vijay Sampathkumar, who serves as Chief Business Officer at Refroid Technologies, to examine the growing adoption of liquid cooling in India, the impact of rising GPU heat and power requirements, and the role of energy efficiency, sustainability and ROI in shaping the country’s AI data centre ecosystem. Here are the excerpts of the interview:
I believe liquid cooling is no longer a futuristic technology. It is already becoming important for today’s data centres. Over the past few years, the conversation has changed significantly. Earlier, I had to approach data centres and explain why liquid cooling mattered. Today, operators are asking about its benefits and return on investment.
I am seeing data centres consider fully liquid-cooled infrastructure as well as specific liquid-cooled data halls, depending on the GPUs and CPUs they are deploying. In my view, liquid cooling is becoming an important part of data centre infrastructure in India.
One of the strongest reasons behind liquid cooling adoption is the increasing heat generated by modern GPUs. If you take NVIDIA chips as an example, their power levels have moved from around 300 watts towards 2,800 watts per chip.
At the same time, chip sizes are becoming smaller, which means the concentration of heat is becoming much higher. From a physics perspective, conventional air cooling cannot keep addressing these thermal loads indefinitely. This is why direct-to-chip liquid cooling is becoming a fundamental requirement for high-performance infrastructure.
I would not say GPU heat is the only driver. Power availability and power costs are equally important concerns for data centre operators.
Even if an organisation has the capital to build a large data centre, it still needs sufficient power. I often ask customers: you may have the money to buy advanced NVIDIA servers, but where will the power come from? How will you optimise it? How will you reduce its consumption and cost?
This is where liquid cooling can play an important role. Carbon credits, heat reuse, ESG goals, and PUE reduction are becoming increasingly relevant to customers evaluating the technology.
I believe liquid cooling can support both energy and water management when it is deployed correctly. At Refroid, we use specialised cooling fluids, and the ability to manage inlet and outlet temperatures can help optimise water utilisation.
There is significant potential in heat reuse. When interacting with customers in Europe and North America, one question that comes up is how the heat generated by data centres can be reused. In certain applications, that recovered heat can potentially be used within facilities.
For me, the conversation should go beyond the size of the cooling distribution unit, heat exchangers and redundancy. Enterprises should also ask what impact the infrastructure can have on ESG, PUE and overall resource efficiency.
I believe organisations need to stop looking at liquid cooling simply as another technology upgrade. They should treat it as infrastructure.
AI is changing what data centres require. Enterprises are not only looking for more computing capacity. They also want resilience, efficiency, lower power consumption and measurable ESG outcomes. Liquid cooling needs to be evaluated as part of that larger infrastructure strategy rather than as an isolated cooling technology.
There is a clear change in the ROI conversation. Earlier, organisations were asking why they needed liquid cooling. Now, more customers are asking what they will get from the investment.
I believe ROI should not be measured only through electricity savings. Liquid cooling can increase rack density and help organisations use expensive data centre real estate more efficiently.
With air cooling, racks require space for airflow. With liquid cooling, multiple racks can be placed much closer together while managing heat more efficiently. This can give organisations more usable real estate and additional computing capacity within the same facility.
The business case for liquid cooling extends far beyond power savings. Rack density, space utilisation, computing capacity and infrastructure longevity can all contribute to the overall ROI.
Energy savings are only one part of the benefit. Ease of use, longevity, available space and the ability to support more computing capacity also matter when organisations evaluate liquid cooling.
I expect India's data centre ecosystem to expand rapidly as AI deployments scale. I see government support, available talent, and India's technology expertise as important factors that can help the country attract more data centre and hyperscaler activity.
At the same time, I believe we need to continue developing our talent base through training and knowledge transfer. Based on my projection in the discussion, India's data centre power requirement could grow from around one gigawatt to 5-6 gigawatts over the next three to four years.
India's AI infrastructure growth will depend on more than physical data centre capacity. Talent development will be equally important. We need continuous training and knowledge transfer to build the skills required for this rapidly changing industry.
Collaboration across the ecosystem will be critical. Government, industry, academia and researchers need to work together. When the ecosystem becomes more synchronised, India will be better positioned to strengthen its data centre capabilities and move closer to becoming a major global AI hub.
Liquid cooling is already moving beyond the experimental stage. The rise of high-density GPUs, growing power requirements, sustainability goals and the need to optimise expensive data centre space are all driving adoption.
My central message is simple: organisations should treat liquid cooling as infrastructure rather than just another technology purchase. As AI workloads continue to grow, data centres will need cooling systems that can support higher compute density while improving efficiency and resource utilisation.
The future of India's AI infrastructure will not depend only on how much computing power we can deploy. It will also depend on how efficiently we can power, cool and manage that infrastructure.
Listen to the full discussion on the Analytics Insight Podcast.