AI data centres are hungry for power — here’s how to manage their appetite

The International Energy Agency (IEA) predicted in 2024 that global data centre electricity consumption would double to reach 1,000 terawatt hours (TWh) by 2026, equivalent to Japan’s annual electricity consumption. The world’s biggest tech companies plan to invest $1 trillion in new data centres over the next five years, primarily because of the sustained growth of AI and cloud services. 

However, AI and machine learning models require far more computing power than traditional cloud computing functions, putting enormous pressure on local utility grids. For example, processing the average ChatGPT query demands 15 times more energy than a traditional web search.

Data centres are also expensive to operate, but downtime is even more costly. Every minute a data centre is offline can result in losses of tens of thousands of pounds. Consequently, maintaining efficiency and round-the-clock reliability is crucial to remain competitive, profitable and scalable.

So, data centre professionals face several challenges when looking to meet these demands cost-effectively and sustainably. Here are a few strategies that will enable you to overcome them. 

Take advantage of AI

While AI is driving the increase in power demand, it might also help solve it via optimised cooling, predictive condition-based maintenance, data handling, and effective demand balancing. 

Connectivity plays a crucial role here. An AI-driven automation system can monitor the vast array of data points present in a medium to large-scale data centre. This capability provides operators with a comprehensive view of the data centre's performance, energy consumption, and asset health at any moment.

Operators can then use these insights to make targeted efficiency improvements. Consider, for instance, the cooling system. Typically, the upstream chiller and distribution system are viewed as separate entities, which means that efforts to improve the efficiency of one component might inadvertently reduce the efficiency of the other. 

However, an automation system allows operators to understand their interdependencies. This holistic perspective empowers them to make informed decisions to improve the entire system’s efficiency. 

Plan modular

When constructing or expanding a data centre, consider adopting a modular approach. Modularity allows for sustainable scaling, meeting power and availability demands while simplifying specification and installation processes. Instead of building the entire facility simultaneously, you can develop it in stages. For example, a planned 200-megawatt (MW) data centre can be built in 20MW sections, onboarding customers as demand grows. This method is more cost-effective, avoiding excess capacity. 

It also minimises downtime and risk, especially when using prefabricated solutions like skids and eHouses, which are assembled off-site and factory-tested before delivery. Purchasing a single prefabricated product, rather than multiple components requiring assembly, saves time and money. Local utilities and governments often prefer this approach, as the gradual growth of a data centre is more likely to gain approval than a full-scale new construction.

Extend modularity to your UPS

A modular approach is also highly effective in a UPS setup. Many data centres still use traditional monolithic UPS systems, where all components are built into single blocks. This design limits capacity adjustments without replacing the entire system or adding another UPS. In contrast, modular UPS systems comprise multiple smaller, interchangeable, autonomous modules working together in one frame. For instance, a one-megawatt (MW) modular UPS can be constructed from four parallel 250-kilowatt (kW) modules.

While monolithic UPS systems often have lower initial costs, modular setups offer better TCO and availability due to their inherent redundancy. Operators can specify a single module rather than an entire monolithic UPS block. To illustrate, achieving N+2 redundancy for a 1 MW data centre with a 1.5 MW modular UPS frame can save 30% compared to N+N redundancy using two 1 MW monolithic blocks.

Modular systems are often lighter, facilitating installation on raised floors or rooftops. They also reduce the need for additional power equipment like cables and switchgear. In the scenario above, two parallel monolithic UPS would require two input and two output switches, while a modular setup needs only one of each.

Moreover, modular UPS systems are more energy efficient. During periods of low demand, modern modular setups can automatically switch modules to sleep or economy mode, reducing energy consumption.

Consider shifting to medium voltage 

Data centres are becoming bigger and will continue to do so. This rise in power demand is making medium-voltage (MV) equipment increasingly more viable and cost-effective. 

An MV Uninterruptible Power Supply (UPS) can deliver power protection to the entire data centre, not just to the racks like a low-voltage (LV) UPS would. They also bring notable energy-efficiency gains, resulting in long-term cost savings and a more attractive total cost of ownership (TCO).

MV UPSs also have a modular installation capability. This avoids relying on numerous LV UPSs that require regular service and maintenance, making MV UPSs more cost-effective. 

For instance, you can parallel wire ten 2.5 MW UPS blocks to form a collective 25 MW system. Such configurations bolster the rapid deployment of additional units, increasing overall system capacity without introducing added complexity. Additionally, the lower currents at MV levels require cables with smaller cross-sections, reducing initial costs and improving TCO. 

A positive contribution

Power availability will always be a pain point for data centre operators. By considering these strategic approaches, data centre managers can meet increasing power demands while upholding sustainability commitments and managing costs. As AI and digital services expand, these practices will be crucial in ensuring data centres positively impact the global energy landscape.