How AI can slash data centre costs

The generative AI boom has seen demand for data centre power soar. Capacity demands had already been increasing rapidly due to the near-universal adoption of cloud-based apps and storage across businesses and for private use cases. Now, advanced AI models have burst into the scene in a similar way, bringing with them a need for unprecedented computing power. At its heart, AI is based on the rapid analysis of vast amounts of data—which means data centre capacity is having to grow exponentially to keep pace.

And because of all this, power needs are jumping, too. Driven in part by AI adoption, data centres across the globe are consuming ever-increasing amounts of electricity. Two years ago, data centres consumed 460 terawatt hours, and the International Energy Agency predicted this figure was set to double within four years. That means that by 2026, data centres could be using 1,000 terawatt hours annually. What does that mean? Well, it’s roughly equivalent to the electricity consumption of Japan, which has a population of 125 million people. Our technology needs are effectively threatening to add a whole new developed nation’s worth of energy usage to the global total.

There are clear issues with this—most pressingly, the ecological and environmental impact this energy use could have. The electricity that powers data centre racks has to come from somewhere, and the amount that’s generated via clean or renewable sources varies wildly across the world. As AI usage grows, the IT infrastructure industry needs to play its part in delivering solutions that work for the planet as well as end users. At the same time, cost is the other significant consideration; if generative AI is to continue flourishing, it’s important that the associated costs, including energy use, are carefully managed.

AI: Cause or cure?

But the cause of the problem could be the solution. As well as drawing on resources, AI can transform data centre usage. Using AI in data centres could lessen energy consumption, improve efficiency, and provide financial savings. For example, AI can help keep data centres cool by analysing historical data and predicting when systems will face increased workloads and, inevitably, the high temperatures that come with them.

Additionally, AI requires a lot of electricity to operate. Naturally, AI processing in data centres can result in higher heat output, which seems to be a counterproductive approach when implementing AI for sustainability purposes, especially if it is to be used for servers and model training. This explains why the majority of AI developments in data centres are concentrated on simple functions, such as enhancing rack-level cooling and power efficiency. 

Indeed, many of the existing problems that companies are facing with traditional data centres, such as thermal management and data breach and security issues, can be eliminated through AI use—allowing the industry both to respond to the emerging challenge of growing energy use and to streamline operations, ultimately enabling new growth. 

Let’s look in more detail at two key ways AI can contribute to a more efficient data centre estate: workload management and predictive analytics.

Workload management

AI is highly effective at optimising processes, so why not use it to optimise data centres? With the help of AI, data centre operations can be optimised by putting operational data to work. With the right intelligent systems in place, data centre resources can be allocated in a way that reduces downtime and minimises energy wastage. AI can analyse energy-intensive and routine tasks, dynamically utilising energy based on the processes.

At the same time, the development of GPU-based AI data processing and LLMs means fundamental performance can be significantly enhanced, achieving up to 50 times the processing speed of CPU-based systems. This advancement means that a few hundred GPU-based systems could replace tens of thousands of CPU-based servers, leading to reduced server space, lower energy costs, decreased core-based software licensing fees, and simplified management.

There are many other processes that contribute to the effective functioning of a data centre, like maintaining the temperature of the entire system. Data centres generate a lot of heat due to the workloads they handle. If the machines in the data centre aren’t kept cool, they’re liable to overheat, break down, and bring operations to a halt. So you’re not just using energy to run the racks—you’re also using (a lot more) energy to cool them. 

Here, AI can lend a hand by analysing the system and more effectively utilising the power available to keep the data centre temperature controlled and cool.

Predictive analysis

Another domain where AI greatly helps is in the analysis of data. As data centres become more complex, it is increasingly important to analyse historical data to predict factors like workload and temperature. These systems improve over time, and as AI advances, they can analyse more data and predict these factors with greater accuracy.

Data centre companies are already using machine learning to create intelligent thermal management solutions, aiming to reduce energy consumption in data centre infrastructure. These kinds of predictive systems can achieve a significant reduction in power usage, leading to significant annual power cost savings of many hundreds of thousands of kilowatt hours—which translates to millions of kilograms of carbon dioxide.

In essence, AI is at its best when it’s automating tasks that would take a human hours or days of laborious, repetitive work. By using its formidable capabilities to improve the basic functionality of data centres, businesses can ensure their infrastructure is fit to serve the next great tech boom—and reduce their carbon footprint.