The digital transformation is changing, and is destined to change more and more, many aspects of our lives: the way we work, communicate, connect with each other in the most diverse areas. And although this may bring enormous benefits in the years to come, it must be kept in mind that this great – and necessary – transformation does have an impact.
We often tend to underestimate the fact that the digital economy rests on a physical backbone made up of data centres: infrastructures that are increasingly strategic – for generative AI as well as for streaming – but also highly energy-consuming. Understanding where we are today, and where we are going, is crucial to gaining a true awareness of the change taking place: only in this way will it be possible to prevent digital progress from burdening the climate more than we can afford.
The environmental impact of data centres
In 2022 alone, the global electricity consumption of data centres has been estimated at between 240 and 340 TWh, or about1-1.3% of global electricity demand. The corresponding greenhouse gas emissions are significant, but so far small: about 330 MtCO2eqworldwide, or about 0.9% of energy-related greenhouse gas emissions.
On the other hand, the growth in demand for digital services has been staggering. Indeed, between 2015 and 2022, internet traffic is estimated to have increased by 600% and workloads in data centres by 340%. However, data centre consumption has not grown proportionally due to major efficiency improvements: the increase in energy required, on a global basis, has been 20-70%, and this is due to more efficient servers and cooling systems, as well as the shift from small – and less efficient – corporate data centres to large cloud and hyperscale data centres optimised for efficiency.
In spite of this, particularly in some geographical areas, data centres weigh significantly on the electricity grid. In Ireland, for example, they were already absorbing 18% of all national electricity in 2022 – a threefold increase since 2015 – while in Denmark consumption is expected to increase sixfold by 2030, bringing them to around 15% of national electricity use. In the United States, on the other hand, where there are thousands of data centres, the estimated consumption in 2023 was 176 TWh, equivalent to 4.4% of the entire US electricity consumption and corresponding to more than 100 MtCO2 of annual emissions. These figures show that despite unquestionable progress in efficiency, the carbon footprint of data centres is already very significant, and particularly concentrated in countries with high digital density.
The perspectives of AI
The issue becomes even more important considering that the accelerated digitisation underway suggests a further increase in demand for computing, storage and data traffic over the next few years. With a consequent exponential growth in energy requirements if no significant innovations are made. Estimates show, in fact, that the total energy consumption of data centres will rise from 352 TWh in 2023 to around 857 TWh in 2028, with a compound annual growth rate close to 20 per cent: more than double in just five years. These are estimates, of course, but the message is clear: if digital demand continues to grow at the current rate, the pressure on energy networks and climate targets will be enormous.
Driving this growth are a number of specific digital technologies and trends such asartificial intelligence, which is seen as one of the main drivers of increased loads in data centres. Indeed, training deep learning models on huge datasets requires complex high-performance computational infrastructures and, once deployed, these AI models run for millions of users, generating a continuous energy load. In particular, specific AI models, such as Large Language Models (LLM), trained on gigantic amounts of textual data, represent a clear case in point, especially due to the popularity of chatbots such as ChatGPT, which have led to an enormous increase in the computational load in AI data centres: in addition to the energy required for training, in fact, the inference infrastructure must support billions of user queries and interactions, while maintaining fast response times. All this certainly has a non-negligible impact on the carbon footprint of the ICT sector, and thus on the more general achievement of environmental sustainability goals.
However, it is crucial to make it clear that AI is not only a cause of consumption, but can – and must – be considered at the same time as a strategic tool for efficiency. There are already many areas in which this technology can make a huge contribution in terms of sustainability: in data centres themselves, moreover, AI can help a great deal in terms of optimising cooling or even server load.
This is why looking at the issue of data centre sustainability with the necessary attention means pursuing a path based, primarily, on three elements: technological innovation, paradigm shifts – such as the realisation of less energy-intensive AI models – and, last but not least, decarbonisation of the energy that powers these infrastructures.
Renewables between challenges and opportunities
Decarbonising the power supply of these infrastructures means, in a nutshell, switching from electricity from fossil fuels to electricity from renewable sources – or nuclear power – to cover the energy needs of servers, cooling and auxiliary equipment. A transition that presents great opportunities, but also non-trivial challenges.
For some time now, the major industry players have been proactive in this area and, through long-term Power Purchase Agreements (PPAs), are now among the world’s largest purchasers of renewable energy: companies such as Amazon, Microsoft, Meta and Google alone have already contracted nearly 50 GW of new renewable capacity to power their data centres and operations. These initiatives are important because in perspective, a data centre powered entirely by renewable energy – directly or via the grid – would have almost zero operational emissions, breaking the link between increased digital services and increasedCO2 emissions. In this direction, in addition to off-site purchases, there are also opportunities for self-production and innovative solutions: some data centres integrate solar or micro wind power plants, and consider using batteries and storage systems to maximise self-consumption of clean energy.
Achieving the goal of sustainably powering these infrastructures, however, also presents important critical issues, including those related to the variability of sources. In other words, the energy from sun and wind often does not coincide with the load profile of data centres – which require continuous and stable power. This implies that, at times when wind and sun are not available locally, the data centre continues to use electricity from the grid, which may include fossil fuels. The path is therefore far from simple, but with the right and necessary steps – for example, developing grids and storage systems to align renewable production and 24/7 consumption – renewable energy can be the key to combining the development of these infrastructures with the fight against climate change.
For a digital climate ally
In short, digital transformation needs increasingly powerful data centres, and the planet needs them to be increasingly sustainable. The course is as clear as it is complex: drive efficiency, contain the ‘energy burden’ per service rendered, and decouple computing growth from emissions, with clean energy available when and where it is needed. And this is not just a task for the IT giants: regulators, utilities and companies must align incentives, standards and investments. The development of digital must not be a limitation to the achievement of climate goals but, on the contrary, must be an essential enabler. In this way, the AI and digital services of today and tomorrow can play a key role in building a sustainable future, without blowing the climate balance.