How AI boom rides on data centres keeping their 'cool'
Cooling system technologies have emerged as a critical determinant of how quickly the next generation of data centres can expand
According to the International Energy Agency (IEA), global electricity consumption by data centres is projected to more than double to around 945 terawatt-hours (TWh) by 2030, driven largely by AI workloads.
Most of the attention has focused on advanced chips from companies such as NVIDIA and the powerful servers required to train and run large language models. Yet a less visible constraint is emerging behind the scenes. The ability to remove heat efficiently is becoming just as important as access to computing power itself.
As AI workloads grow more complex and energy-intensive, cooling systems are rapidly evolving from an operational necessity into a strategic component of digital infrastructure. In many regions, cooling capacity is expanding faster than server deployments because without effective thermal management, the next generation of AI simply cannot operate at scale.
A new era of high-density computing
Generative AI has transformed the economics of computing. Training and deploying large language models require vast numbers of graphics processing units (GPUs), which consume significantly more power than traditional central processing unit (CPU)-based servers.
According to estimates from Schneider Electric and NVIDIA, conventional enterprise server racks typically operate at between 5 and 15 kilowatts while AI racks increasingly require 50 kilowatts or more. Some next-generation deployments are expected to exceed 100 kilowatts per rack.
Major cloud providers, including Microsoft, Google and Amazon Web Services, are investing heavily in AI data centres, alongside operators such as Equinix, Digital Realty and India’s CtrlS, all of which are redesigning facilities to accommodate higher power densities and more sophisticated thermal management requirements. NVIDIA’s latest GPU platforms are designed for significantly higher power envelopes than previous generations, forcing operators to rethink data centre layouts, power delivery and thermal management.
Every watt consumed by a server eventually becomes heat, creating a fundamental infrastructure challenge that is reshaping how AI data centres are designed. As GPU power consumption rises, so does the difficulty and cost of keeping systems within safe operating temperatures.
As Pankaj Sharma, executive vice-president of Secure Power at Schneider Electric, observed: “AI applications, especially training clusters, are highly compute-intensive and require large amounts of processing power provided by GPUs or specialised AI accelerators. This puts a significant strain on the power and cooling infrastructure of data centres.”
Meeting these demands requires more than installing additional servers. AI hardware cannot simply be deployed in existing facilities without first upgrading cooling and power infrastructure. In many cases, operators invest in liquid cooling and power distribution before additional servers arrive.
Today’s AI data centres must be designed as integrated systems where power, cooling and computing capacity evolve together. Every infrastructure decision now affects performance, efficiency and long-term scalability. The priority is no longer to build larger data centres but rather create smarter infrastructure capable of sustaining dense AI workloads without compromising efficiency or reliability.
Evolving cooling technologies
Traditional air-cooling systems, designed for lower-density enterprise workloads, are increasingly unable to dissipate the heat generated by modern AI racks, driving the shift towards liquid-based cooling technologies. Liquid cooling technologies offer a step-change in thermal management. Direct-to-chip systems circulate coolant through plates attached directly to processors, removing heat far more efficiently than air. Immersion cooling goes further, submerging servers entirely in specialised dielectric fluids.
Poor cooling directly impacts server performance through thermal throttling, where processors automatically slow down to avoid overheating. Better cooling allows operators to maximise performance, extend hardware lifespan and reduce operational risk.
Energy efficiency remains a key performance metric for data centres. Power usage effectiveness (PUE) measures how much total facility energy is required relative to the energy used directly by computing equipment, with a score closer to 1.0 indicating greater efficiency.
According to the Uptime Institute’s Global Data Center Survey 2025, the industry’s weighted average annual PUE has remained just above 1.5 for the sixth consecutive year, reflecting the continued influence of older facilities even as newer hyperscale data centres achieve much lower values. Google’s 2025 Environmental Report states that its global data centre fleet recorded an average annual PUE of 1.09 in 2024, highlighting how energy efficiency has become a key competitive differentiator for hyperscale operators.
The growing demand for cooling also raises sustainability concerns. The US-based Environmental and Energy Study Institute estimates that a medium-sized data centre can consume around 110 million gallons of water annually for cooling while large facilities may use up to 5 million gallons each day.
Towards sustainable AI growth
Closed-loop liquid cooling systems can significantly reduce water consumption while improving thermal performance. Microsoft has tested designs that minimise or eliminate water use for cooling during significant parts of the year.
Long-term renewable power agreements, already being signed by companies including Google, Amazon Web Services and Microsoft, are essential but insufficient on their own. Investment in transmission networks, energy storage and new grid capacity will also be required to accommodate rising demand from AI infrastructure.
Waste heat recovery deserves greater attention. In some regions, excess heat from data centres is already being reused to support district heating networks, industrial processes and nearby buildings.
AI itself may help solve part of the problem. AI-driven thermal management systems can monitor workloads, predict cooling requirements and optimise equipment performance in real time, reducing both energy use and operating costs.
Cooling as a strategic advantage
The AI boom is redefining what matters most inside a data centre. Advanced chips and powerful servers remain essential, but cooling infrastructure is emerging as the decisive factor in determining where AI capacity can be built and how efficiently it can operate.
The next phase of AI development will depend not only on access to semiconductors and electricity but also on the ability to manage heat efficiently, reliably and sustainably. In the race to scale AI, the organisations that solve the cooling challenge may gain the greatest competitive advantage.
Subscribe to India Today Magazine
According to the International Energy Agency (IEA), global electricity consumption by data centres is projected to more than double to around 945 terawatt-hours (TWh) by 2030, driven largely by AI workloads.
Most of the attention has focused on advanced chips from companies such as NVIDIA and the powerful servers required to train and run large language models. Yet a less visible constraint is emerging behind the scenes. The ability to remove heat efficiently is becoming just as important as access to computing power itself.
As AI workloads grow more complex and energy-intensive, cooling systems are rapidly evolving from an operational necessity into a strategic component of digital infrastructure. In many regions, cooling capacity is expanding faster than server deployments because without effective thermal management, the next generation of AI simply cannot operate at scale.
A new era of high-density computing
Generative AI has transformed the economics of computing. Training and deploying large language models require vast numbers of graphics processing units (GPUs), which consume significantly more power than traditional central processing unit (CPU)-based servers.
According to estimates from Schneider Electric and NVIDIA, conventional enterprise server racks typically operate at between 5 and 15 kilowatts while AI racks increasingly require 50 kilowatts or more. Some next-generation deployments are expected to exceed 100 kilowatts per rack.
Major cloud providers, including Microsoft, Google and Amazon Web Services, are investing heavily in AI data centres, alongside operators such as Equinix, Digital Realty and India’s CtrlS, all of which are redesigning facilities to accommodate higher power densities and more sophisticated thermal management requirements. NVIDIA’s latest GPU platforms are designed for significantly higher power envelopes than previous generations, forcing operators to rethink data centre layouts, power delivery and thermal management.
Every watt consumed by a server eventually becomes heat, creating a fundamental infrastructure challenge that is reshaping how AI data centres are designed. As GPU power consumption rises, so does the difficulty and cost of keeping systems within safe operating temperatures.
As Pankaj Sharma, executive vice-president of Secure Power at Schneider Electric, observed: “AI applications, especially training clusters, are highly compute-intensive and require large amounts of processing power provided by GPUs or specialised AI accelerators. This puts a significant strain on the power and cooling infrastructure of data centres.”
Meeting these demands requires more than installing additional servers. AI hardware cannot simply be deployed in existing facilities without first upgrading cooling and power infrastructure. In many cases, operators invest in liquid cooling and power distribution before additional servers arrive.
Today’s AI data centres must be designed as integrated systems where power, cooling and computing capacity evolve together. Every infrastructure decision now affects performance, efficiency and long-term scalability. The priority is no longer to build larger data centres but rather create smarter infrastructure capable of sustaining dense AI workloads without compromising efficiency or reliability.
Evolving cooling technologies
Traditional air-cooling systems, designed for lower-density enterprise workloads, are increasingly unable to dissipate the heat generated by modern AI racks, driving the shift towards liquid-based cooling technologies. Liquid cooling technologies offer a step-change in thermal management. Direct-to-chip systems circulate coolant through plates attached directly to processors, removing heat far more efficiently than air. Immersion cooling goes further, submerging servers entirely in specialised dielectric fluids.
Poor cooling directly impacts server performance through thermal throttling, where processors automatically slow down to avoid overheating. Better cooling allows operators to maximise performance, extend hardware lifespan and reduce operational risk.
Energy efficiency remains a key performance metric for data centres. Power usage effectiveness (PUE) measures how much total facility energy is required relative to the energy used directly by computing equipment, with a score closer to 1.0 indicating greater efficiency.
According to the Uptime Institute’s Global Data Center Survey 2025, the industry’s weighted average annual PUE has remained just above 1.5 for the sixth consecutive year, reflecting the continued influence of older facilities even as newer hyperscale data centres achieve much lower values. Google’s 2025 Environmental Report states that its global data centre fleet recorded an average annual PUE of 1.09 in 2024, highlighting how energy efficiency has become a key competitive differentiator for hyperscale operators.
The growing demand for cooling also raises sustainability concerns. The US-based Environmental and Energy Study Institute estimates that a medium-sized data centre can consume around 110 million gallons of water annually for cooling while large facilities may use up to 5 million gallons each day.
Towards sustainable AI growth
Closed-loop liquid cooling systems can significantly reduce water consumption while improving thermal performance. Microsoft has tested designs that minimise or eliminate water use for cooling during significant parts of the year.
Long-term renewable power agreements, already being signed by companies including Google, Amazon Web Services and Microsoft, are essential but insufficient on their own. Investment in transmission networks, energy storage and new grid capacity will also be required to accommodate rising demand from AI infrastructure.
Waste heat recovery deserves greater attention. In some regions, excess heat from data centres is already being reused to support district heating networks, industrial processes and nearby buildings.
AI itself may help solve part of the problem. AI-driven thermal management systems can monitor workloads, predict cooling requirements and optimise equipment performance in real time, reducing both energy use and operating costs.
Cooling as a strategic advantage
The AI boom is redefining what matters most inside a data centre. Advanced chips and powerful servers remain essential, but cooling infrastructure is emerging as the decisive factor in determining where AI capacity can be built and how efficiently it can operate.
The next phase of AI development will depend not only on access to semiconductors and electricity but also on the ability to manage heat efficiently, reliably and sustainably. In the race to scale AI, the organisations that solve the cooling challenge may gain the greatest competitive advantage.
Subscribe to India Today Magazine
