Rob Campbell, President of Communications, Enterprise & Cloud at Flex talks emerging tech and keeping your cool in a scaling environment
As computing demands continue to rise, particularly with the exponential growth of artificial intelligence (AI) and high-performance computing (HPC), data centres and industrial systems are facing new challenges around heat management. The energy-intensive nature of modern computing infrastructure generates vast amounts of heat, much of which is traditionally treated as waste. However, advances in technologies such as liquid cooling, intelligent thermal design and integrated energy systems are creating new opportunities, helping businesses turn what was once an operational burden into a valuable resource.
Rethinking heat management in high-performance computing
The escalating thermal demands of AI and cloud computing are pushing conventional cooling strategies to their limits. Traditional air cooling, which has served the industry for decades, is no longer sufficient as rack power densities routinely exceed 50kW, a threshold that once defined the upper end of typical data centre design.
AI workloads are driving this shift, with rack densities expected to rise dramatically in the coming years. NVIDIA forecasts that AI racks could reach 600kW by 2027 and early pilots of 1MW racks are already underway. Hyperscalers and chipmakers are collaborating on new architectures and standards to support this evolution, anticipating broader deployment of 1MW racks before the end of the decade. These ultra-dense systems, designed for high-performance computing and AI, will consume enormous amounts of power, generating heat at levels that traditional cooling methods simply cannot manage.
Liquid cooling: the foundation for heat reuse
Liquid cooling has emerged as a cornerstone technology to address this challenge. Unlike air cooling, which disperses heat into the atmosphere, liquid cooling systems absorb heat directly at the source, often at the chip or board level, and carry it through a closed-loop system.
While not a new approach, the needs of AI data centres and recent technology advancements have driven a resurgence in liquid cooling, transforming it into a high-priority, fast-growing market. It offers several key advantages over traditional air-based systems, but the largest benefit is in efficiency. Liquids, particularly engineered or dielectric fluids, have a much higher heat capacity than air, enabling more efficient thermal transfer directly at the source. This reduces reliance on large-scale HVAC systems, lowering operational energy costs and allowing for more compact, efficient facility designs.
Although liquid cooling is becoming more prevalent, most data centres are still designed to support a broad range of workloads – not just AI. These facilities typically serve a mix of traditional IT, cloud services and high-performance computing workloads. As a result, there is increasing interest in hybrid cooling solutions, which combine the efficiency of liquid cooling where needed with lower-cost air cooling for less demanding systems. These hybrid approaches strike a practical balance between sustainability, performance, and capital expenditure, allowing data centre operators to adapt to changing workloads.
Integrating cooling, power, and compute design
To support next-generation workloads, particularly those driven by AI, data centres must integrate compute density, power delivery and thermal management into a cohesive system. Emerging technologies are enabling this shift by facilitating smarter, more responsive infrastructure designs.
One key development is power-aware thermal design, which allows data centres to better manage heat generated by high-performance components. AI workloads, especially those involving GPUs, can create unpredictable power surges and localised hot spots. Intelligent co-design of thermal and power systems enables engineers to anticipate these fluctuations, minimising inefficiencies and reducing stress on infrastructure equipment in and around the data centre.
The rise of modular, high-density rack systems is also transforming scalability. These racks are built to support power loads well beyond 125kW, with designs already in development to handle up to 1MW. Their modular nature allows operators to add capacity without a complete overhaul, while also enabling more effective heat capture and reuse across multiple stages of operation.
Direct-to-chip cooling and advanced power delivery technologies are pushing thermal efficiency to new levels. Precision systems, such as micro-convective cooling, remove heat directly from critical silicon components. When combined with innovations like vertical power delivery (VPD), these solutions reduce energy loss and maintain higher-quality waste heat.
The role of AI and machine learning
One of the more transformative forces in thermal and energy optimisation is machine learning. Predictive analytics can monitor energy usage and thermal loads at a granular level, forecast peak demand periods and shift loads to optimise thermal capture, as well as dynamically adjust cooling systems for efficiency and performance.
AI can also be used to simulate the long-term impact of infrastructure upgrades, giving operators confidence to make capital investments in complex integrated systems.
Meeting the scalability and efficiency challenges of the future requires a system-level approach, often called a ‘grid-to-chip’ strategy. This holistic model starts at the power grid interface, flows through power distribution units (PDUs), into the rack-level architecture, down to the board and chip level and loops back through thermal recovery systems. Importantly, this approach enables resiliency and flexibility, helping operators manage not only heat but also power quality, redundancy and grid disturbances.
Closing the loop on energy waste
Beyond technical efficiencies, the shift toward intelligent heat management brings significant sustainability advantages for businesses. Traditionally, the heat generated by data centres and HPC environments is viewed as a costly by-product. But with the advances in liquid cooling and heat reuse technologies, this waste can be recovered to improve energy efficiency and reduce overall carbon emissions.
Liquid cooling systems, with their closed-loop design, also require less water than evaporative air cooling, which can support water conservation goals in regions facing resource constraints. Additionally, by enabling higher computing density with lower spatial and energy footprints, these systems help data centres do more with less, reducing the need for new infrastructures.
By aligning cooling strategies with real-world deployment scenarios and focusing on efficiency in both water and energy use, these improvements support sustainability initiatives. This not only helps businesses meet net-zero targets, but also demonstrate leadership in sustainable innovation.
Looking to the future
As more data centres and industries explore heat reuse, governments and regulatory bodies are beginning to take notice. Policies incentivising heat reuse are emerging, particularly in Europe, where the EU has placed heat reuse on its sustainability agenda. Compliance may soon shift from voluntary to mandatory, particularly for large-scale facilities.
Standardised heat interface units, interoperable cooling loops and metrics for waste heat quality are also needed to enable broader adoption. Cross-industry collaboration, between tech providers, energy utilities, regulators and facility operators, will also be critical to establishing these frameworks.
Overall, emerging technologies are fundamentally reshaping how heat is managed and reused across digital and industrial infrastructure. Liquid cooling, intelligent rack systems and grid-to-chip integration offer a pathway for businesses in the future. By embracing this new paradigm, data centres and industrial facilities alike can reduce their carbon footprints, improve energy resilience and unlock new forms of value in an increasingly digital and energy-conscious world.
