Artificial intelligence is forcing a fundamental rethink of how data centres are designed, built and operated, not because of software complexity but because of heat. As AI and high performance computing workloads drive rack densities far beyond historic norms, thermal management has emerged as one of the defining engineering challenges of the digital economy, a trend examined in recent industry analysis published by infrastructure specialist Legrand.
Modern facilities are now confronting what operators describe as an unprecedented cooling challenge. Cabinets that once consumed 3 to 5 kilowatts of power routinely operate at 10 to 30 kilowatts today, while specialised AI and high performance computing environments exceed 100 kilowatts per rack. Some projections place future systems at 300 kilowatts or more, levels that traditional air cooling systems were never designed to handle.
The shift reflects the physical consequences of artificial intelligence itself. GPU acceleration, higher processor core counts, increased memory density, faster storage technologies and high bandwidth networking equipment all generate additional thermal load. Each improvement in computational capability translates directly into more heat that must be removed reliably and efficiently.
Air cooling reaches its physical limits
For decades, conventional cooling architectures based on computer room air conditioning units and raised floor airflow defined data centre design. Those systems now face fundamental physical constraints. Air can carry only a limited amount of thermal energy, making it increasingly difficult to prevent hot spots as densities rise.
Operators report challenges ranging from uneven cooling distribution and bypass airflow to growing energy consumption that undermines sustainability targets. Even improvements such as hot and cold aisle containment struggle under modern workloads, where higher airflow requirements create pressure limitations and reduce overall efficiency.
These limitations have wider implications for artificial intelligence deployment. Thermal instability does not simply affect equipment lifespan, it constrains how much compute can be installed within a facility. Cooling capacity is increasingly determining AI capacity itself.
At the same time, environmental expectations are tightening. Cooling systems account for between 30 and 40 per cent of a facility’s energy consumption, placing pressure on operators to achieve lower power usage effectiveness targets. Industry goals that once accepted PUE levels of 1.5 now aim for 1.2 or lower, with leading operators targeting 1.1.
Water consumption presents an additional constraint. Traditional evaporative cooling approaches can consume millions of gallons annually, creating challenges in regions facing water scarcity or regulatory restrictions. As sustainability commitments and investor expectations grow, cooling strategies must balance performance with environmental responsibility.
Hybrid cooling becomes the new architecture
The response emerging across the industry is not the replacement of air cooling but its reinvention through hybrid approaches that combine established methods with liquid cooling technologies.
Rear door heat exchangers, direct liquid cooling and immersion cooling systems are increasingly deployed alongside conventional airflow systems, each addressing different density requirements. Liquid cooling allows heat to be removed closer to the source, supporting racks exceeding 100 kilowatts while reducing energy consumption associated with large-scale air handling.
No single technology meets all requirements. Operators are instead adopting zone-based strategies, applying different cooling methods across facilities depending on workload density. Lower density enterprise systems may continue using air cooling, while AI clusters transition to liquid-based solutions capable of managing extreme thermal loads.
Infrastructure flexibility has become central to these decisions. Cooling systems must adapt to evolving hardware generations and unpredictable workload patterns, while minimising space consumption and allowing incremental expansion as computing demand grows.
Operational complexity meets infrastructure reality
Advanced cooling introduces new operational demands. Liquid systems require chilled water distribution, leak detection procedures and specialised maintenance skills. Monitoring platforms must integrate multiple cooling technologies into a unified thermal management strategy, while staff training becomes essential to maintain safe operations.
Yet this complexity reflects a broader transition in the AI era. Data centres are no longer passive facilities supporting digital services, they are active engineering environments where energy, thermodynamics and computation intersect.
The industry’s conclusion is increasingly clear. Artificial intelligence is not only transforming software and business models but redefining the physical infrastructure required to sustain them.
As computational demand accelerates and environmental pressures intensify, cooling strategy will increasingly determine how far AI can scale. The organisations that successfully combine efficiency, flexibility and sustainability in thermal design are likely to define the next generation of digital infrastructure, where managing heat becomes as critical as processing data itself.
