Data centres are experiencing a significant transformation, driven by the rise of AI and the pressing need to combat climate change. Whilst the sustainability challenges posed by AI are often highlighted, the technology’s potential to enhance the sustainability of data centres is frequently overlooked.
On the one hand, the data-intensive workloads generated by AI will see power consumption soar to unprecedented levels in high-capacity data centres that demand more energy. However, the technology itself can also unlock transformative efficiencies within these facilities, paving the way for next-generation data centres that are both high-capacity and sustainable. The projected growth of AI in the energy market, from $4 billion in 2021 to $20 billion by 2031, highlights the increasing significance and application of AI technologies in optimising energy use.[1]
AI boom fuelling demand for high-capacity data centres
Earlier this year, NVIDIA CEO Jensen Huang noted that “accelerated computing and generative AI (GenAI) have reached a tipping point,” highlighting the global surge in demand across various sectors. His remarks came after NVIDIA’s stellar financial performance in the three months leading up to January 28th, during which the company’s revenue skyrocketed by 265%, reaching $22 billion. This impressive growth was primarily driven by NVIDIA’s data centre business, which experienced a staggering five-fold increase, contributing $18.4 billion and constituting most of the company’s revenue for the quarter.[2]
This surge in demand is not isolated to NVIDIA but reflects a broader trend driven by the proliferation of AI technologies. Over the next five years, the amount of data generated will surpass the total produced in the past decade, necessitating a significant expansion of storage capacity in data centres worldwide and a corresponding rise in the need for electricity to power these facilities. According to Bain, the demand for electricity to power data centres will rise by as much as 5% per year in the next few years.[3]
A key factor contributing to this rising energy demand is the escalating computational power required for AI training which is doubling every six months, a rate far more rapid than the 20-month intervals observed 50 years ago. A simple query to AI models like ChatGPT requires nearly ten times the electricity of standard Google searches. [4] The International Energy Agency (IEA) projects that by 2026, data centre electricity consumption could reach 1,000 terawatt-hours (TWh), double that of 2022 and roughly equal to Japan’s total annual consumption. By 2030, data centres could account for 13% of global electricity consumption and 6% of the world’s carbon footprint.[5]
Tech giants, recognising the scale of the problem and their significant contribution to it, are racing to mitigate the environmental impact of their operations.[6] Whilst hyperscale cloud data centres, which use specialised GPUs, are more energy-efficient for AI-related tasks, their massive scale has become a growing concern. Google’s data centres consumed approximately 5 billion gallons of water in 2022, a 20% increase from the previous year, primarily for cooling purposes. Microsoft saw a 34% increase in water usage over the same period.[7]
These companies face mounting pressure to reduce their carbon footprint and meet neutrality targets. Google aims to reach net-zero emissions across all operations by 2030, while Microsoft targets becoming carbon-negative by 2030, with a long-term goal of removing all the carbon it has emitted since its founding by 2050. Achieving these targets won’t be easy because increasing demand for AI applications could negate any gains in energy efficiency made by using AI, potentially leading to a net increase in energy use.
AI-driven energy efficiency in data centres
Most data centres aim to operate in a “steady state,” striving to maintain consistent and predictable energy consumption over time to manage costs and ensure reliable performance. This operational model is crucial because data centres depend on the local electricity grid, which provides a daily mix of natural gas, nuclear, and renewable energy sources. The limited transmission lines between regions mean data centres cannot easily switch to alternative grids to access more stable or cleaner energy sources.
As a result, they must rely on whatever energy is available locally, which can fluctuate significantly. Maintaining a steady state helps mitigate the risks associated with these fluctuations, allowing data centres to manage costs and reliability despite the variability in the energy mix they receive from the grid.
However, AI-driven solutions offer enormous potential to address these challenges by optimising energy usage and predicting and managing demand more effectively.
Optimising power usage in data centres
As data centres continue to expand and evolve, the need for efficient power management becomes increasingly crucial. AI offers innovative solutions to optimise power usage, reduce operational costs, and minimise environmental impact. By leveraging AI-driven technologies, data centres can achieve significant improvements in energy efficiency across various aspects of their operations. The following sections explore key areas where AI is revolutionising power optimisation in data centres.
Dynamic load balancing and workload optimisation
Fluctuating workloads in data centres cause some servers to become overworked while others remain underutilised, resulting in energy waste. AI algorithms can analyse usage patterns and distribute workloads in real time, ensuring optimal utilisation of computing resources. This approach reduces energy consumption, improves performance, and lowers operational costs by minimising the need for excess capacity.
Adaptive cooling system management
Traditional cooling systems in data centres rely on static settings with fixed settings that cannot adapt to real-time conditions, often operating at full capacity, regardless of actual needs. AI-driven thermal modelling offers transformational improvements by enabling real-time, dynamic adjustments to cooling systems.
By analysing sensor data, AI creates digital twins—virtual models of the data centre environment. These models account for upcoming high-intensity computing tasks, expected external temperature fluctuations, and planned maintenance. They simulate various scenarios, predicting temperature changes and potential hotspots within the facility before they occur, allowing for precise and efficient cooling management.
Power usage effectiveness (PUE) optimisation
PUE is a crucial metric for assessing data centre energy efficiency, calculated as the ratio of total facility energy consumption to the energy consumption by IT equipment. While a PUE of 1.0 indicates perfect efficiency, values typically exceed this due to additional energy needs for cooling, lighting, and other infrastructure requirements.
AI optimises PUE by monitoring and adjusting operational parameters in real time. By analysing sensor data throughout the facility, AI can identify inefficiencies and adjust CPU frequencies, fan speeds, and power supply voltages, reducing unnecessary energy consumption, without compromising performance.
Predictive maintenance and energy anomaly detection
AI enables data centres to conduct predictive maintenance by analysing sensor data to identify patterns that signal potential equipment failures. This early recognition enables proactive maintenance when and where malfunctions will likely occur, preventing the activation of energy-intensive emergency cooling systems and reducing reliance on power hungry backup systems.
Furthermore, the technology enhances energy anomaly detection in data centres by monitoring real-time data from various sensors and comparing it to established baselines of energy consumption patterns. When deviations are detected, indicating potential issues such as malfunctioning equipment or irregular cooling patterns, AI systems alert operators for swift resolution. This proactive approach prevents prolonged periods of inefficient energy use and ensures optimal equipment operation.
Smart grid integration and renewable energy
Integrating renewable energy sources like solar and wind into the grid presents challenges due to their variable availability. AI addresses this by forecasting renewable energy availability using weather data and predictive analytics. This enables data centres to shift non-critical workloads to peak renewable energy production periods, maximising the use of clean energy and reducing reliance on fossil fuels.
Key and emerging players
Driven by the dual pressures of escalating computational demands and urgent sustainability imperatives, the data centre industry is experiencing a renaissance. giving rise to a new breed of technology firms. These companies, ranging from innovative start-ups to established players, are leveraging cutting-edge AI and machine learning technologies to redefine the boundaries of data centre efficiency. Their solutions are not merely incremental improvements, but transformative approaches that promise to reshape the very foundations of how data centres operate.
Across the pond in the US, StormForge (formerly Carbon Relay) has made significant strides in optimising resource allocation and efficiency within data centres through their advanced AI-powered solutions. Their two main products, Optimise Live and Optimize Pro, focus on real-time data analysis and resource management, as well as historical data examination and future predictions, respectively. They autonomously right size resources usage for Kubernetes workloads. They aim in particular at reducing “cloud waste” that stems from over-provisioned infrastructure that ends up being unused or under-used, heightening operating costs and energy consumption without adding value.
In the field of demand response Google is a leader, pioneering the use of AI for demand response through its in-house carbon-intelligent computing platform. The platform, which functions like an intelligent task manager for its data centres, leverages real-time data on renewable energy availability and forecasts to shift computing workloads and prioritise using clean energy sources when carbon-free energy is available on the grid.[8]
Meanwhile, in Europe, several companies are making their mark. Dutch innovator and recognised name in the demand response space, Sympower, is enhancing energy flexibility in data centres. Their advanced AI-based flexibility platform analyses real-time energy market data and grid conditions to identify optimal times for buying and selling energy. This approach not only enables data centres to utilise cheaper renewable energy sources but also helps stabilise the overall electricity grid. Sympower has built the technical solution that enables data centres to participate in frequency containment reserve (FCR). They help for instance the Bikupa Datacenter in Sweden to provide up to 10MW FCR available to the grid.
Amsterdam-based Coolgradient is pioneering the use of AI software to ensure data centres operate at peak efficiency. Their advanced machine-learning models analyse real-time sensor data throughout facilities, identifying cooling and power usage inefficiencies. This ‘energy autopilot’ approach not only reduces energy costs but also minimises the environmental impact of data centre operations. Their clients typically get a payback within a year.
Based in the UK, QiO technologies offers innovative software solutions to enhance energy efficiency and sustainability in data centres. Their flagship product, DC+, uses advanced AI and machine learning to analyse real-time data from servers, power supplies, and cooling systems, making dynamic adjustments to optimise power settings and workload distribution, achieving up to a 45% reduction in energy consumption. This significant energy saving reduces electricity bills and extends the lifespan of data centre equipment, providing a higher return on investment for data centre operators.
Further afield, Israeli firm Evolution Innovation Group is at the forefront of energy management technology with its Smart Energy Solutions (SES) software platform. Their adaptive systems, specifically designed for data centres and mobile networks, use AI and deep learning to dynamically adjust energy usage based on real-time data and environmental conditions.
By monitoring and responding to fluctuations in data centre operations and external environmental factors, SES helps reduce energy consumption and operational costs while maintaining optimal performance. This approach leads to significant cost savings and enhanced sustainability, contributing to a reduced carbon footprint. It also ensures data centres can efficiently handle varying workloads and changing conditions.
Lower carbon, high-capacity data centres with AI and green technologies
The transition to low-carbon, high-capacity data centres is critical as the demand for data storage and processing continues to surge. AI combined with innovative green technologies will spearhead this transition.
Green technologies such as advanced cooling systems and energy-efficient hardware are already complementing AI to further diminish energy usage and environmental impact. French company Qarnot goes one step further to reduce the environmental impact of servers. Its QBx computing clusters, which are water-cooled or free-cooled, are distributed throughout the city in buildings where waste heat is recovered (housing, offices, schools, logistics warehouses, etc.). Through its Q.ware software Qarnot can then offer high-performance computing capabilities with high environmental value by avoiding the costs associated with infrastructure, maintenance, and cooling of a traditional data centre. Thanks to Qarnot, cloud computing footprint is reduced by 80%.
Together, AI and green technologies are set to revolutionise data centre operations by enabling them to manage larger capacities while reducing their carbon footprint. This not only supports sustainability objectives but also meets the growing data demands brought about by the rise of AI in the future.
Learn more about us.
View our transactions.
