Greater efficiency in data centers prevents a massive increase in energy consumption

In early May, the Swiss Federal Office of Energy (SFOE) published new statistics on data centers in Switzerland titled “Data Centers in Switzerland—Electricity Consumption and Efficiency Potential.” The full report is now available.To mark this occasion, energeiaplus spoke with the project manager, Martin Jakob of the Zurich-based company TEP Energy, to learn about the key findings of his research. 

Energeiaplus: The report shows that energy consumption by data centers has risen only moderately. How can you explain the fact that the expected sharp increase did not materialize?  

Martin Jakob: As trite as it sounds, it also has to do with simple arithmetic. If only one segment out of three or four is growing while the others stagnate, the growth of that one segment is significantly put into perspective. Public perception has focused primarily on the new large data centers in the greater Zurich area. In the process, the internal data centers of companies and government agencies, the traditional smalland medium-sized colocation centers,and the rest of the data centers in Switzerlandhave beenoverlooked. The outsourcing thatleadsto consolidations, closures, orresizing of internal data centers—and the associated efficiency gains—are far less visible. Regarding expectations: Many were likely guided by global news reports and high-profile large-scale projects. I myself was skeptical a few years ago about simply adding up all the projects on the table to generate a forecast or expectation. Our expert discussions have now revealed that only a fraction of the requests that power grid operators receive for connecting data centers are actually implemented. 

Does it matter that many data centers today aren’t even operating at capacity and therefore use less electricity than people think? 

Yes, it plays a major role. Many of today’s existing data centers are not yet operating at full capacity, which means their electricity consumption is lower than the maximum connection capacity on file with the power grid operators or the publicly available performance data from data centers would suggest. This is shown by our surveys as well as those conducted by the Canton of Zurich.

Martin Jakob, TEP Energy

With this baseline study, we can demonstrate that the increase was only moderate for various other reasons. Importantly, training of large AI models hardly takes place in Switzerland. Equally important is efficiency, which manifests itself in several ways.A major factor is the shift in recent years from smaller, internal server rooms or data centers to colocation and cloud data centers. These can be operated more efficiently due to better utilization—a phenomenon known as higher virtualization. This reduces the need for IT equipment, cooling infrastructure, and thus electricity.Today’s servers are also much more powerful and energy-efficient. Today, a server usessignificantly less electricity to deliver the same computing power. If we were to use the servers and storage from five years ago to meet today’s demand for IT performance,electricity consumptionwouldbe 20–40% higher. On the IT side, current trends are moving away from hard drives toward SSDs and toward greater virtualization ofservers and storage. For data center infrastructure—from hyperscalers down to small internal data centers—measures such as enclosing server racks and separating cold and hot aisles have been implemented. This has led to a general improvement in the PUE efficiency metric (Power Usage Effectiveness).  

What significant changes have been observed since the publication of the last report?  

The aforementioned shift from in-house data centers to colocation and cloud data centers is certainly a major change in recent years. This also applies to companies that previously did not have an in-house data center or a server room, as more and more applications are no longer running on individual PCs but online—that is, in a data center. This includes traditional office applications such as accounting, employee and customer management, as well as various industry-specific software. This is reflected in the fact that the floor space of commercial data centers in Switzerland increased by slightly more than a quarter between 2020 and 2024.  

From a global perspective, artificial intelligence (AI) is,of course,currentlyahotly debatedtopic. However, our study found that AI plays a minor role in Switzerland from an energy perspective.  

You say that artificial intelligence currently plays only a minor role in Switzerland—how do you explain that?  

AI certainly plays a certain role in Switzerland, just not in terms of electricity consumption, but rather in terms of its application. AI has become part of both people’s everyday lives and business operations in Switzerland.  There is also still potential for growth. However, Switzerland is less affected by the highly energy-intensive training of AI models, which leads to a sharp increase in installed capacity and electricity consumption in certain countries and regions. The reason for this is that electricity and land prices are high compared to other regions, and Switzerland does not pursue an active policy of promoting data centers. Large open spaces for very large data centers are also rather scarce. And even when potential sites are available, they often lack the necessary electricity supply capacity. Where Switzerland does have locational advantages, however, is in its political and social stability and security. This could play a role as the concept of a “sovereign cloud” (securecloud infrastructure)becomes more concrete. Businesses and the research community are certainly interested in making greater use of AI, provided that the framework conditions regarding data protection and confidentiality are guaranteed.  

What surprised you while preparing this report?  

I was surprised by the high concentration of large data centers in the greater Zurich area and the neighboring regions of Aargau and Schaffhausen. Small and medium-sized data centers, on the other hand, are much more widely distributed across Switzerland. I was also surprised by the scale of the new construction projects and the fact that they are being built even though some of the small and medium-sized data centers are not yet operating at full capacity. I was also surprised by the still very low capacity utilization rates at the large data centers; according to a survey by the Canton of Zurich, this figure stands at only 30%. It’s clear that certain reserves are factored in when building a new data center, but some of the respondents are also somewhat skeptical as to whether these reserves might be too large. I was also surprised that even a renowned organization like the International Energy Agency (IEA) initially overestimated the electricity consumption of data centers in Europe in its report “Electricity 2024 Analysis and Forecast to 2026.” This shows how important sound foundational data is.   

Can the efficiency of data centers be improved even further?

Absolutely. The energy efficiency potential remains high at 38% of electricity consumption. Efficiency gains can still be achieved on both the infrastructure side and the IT side. For the infrastructure, this would be particularly beneficial in the areas of cooling, air and refrigerant management, and uninterruptible power supply. On the IT side, improvements can be made particularly through optimized utilization, consolidation, and efficient server and storage technologies. The overall efficiency of the energy system can also be increased by making greater use of the heat generated by data centers. Stronger regional and local coordination is very important here.  

What can we expect in the coming years? The report looks ahead to 2030—that’s practically tomorrow—but what comes after that? 

Yes, 2030 is coming soon. From an IT perspective, however, it’s a considerable period of time—an entire generation of devices. Even though the pace of change is currently rapid, particularly in data center construction and AI, certain considerations can still be made. We’re doing just that in the BFE’s ongoing project on energy outlooks. Here, we’re working with scenarios. The challenge is to outline a certain range of possible developments without becoming too speculative. When it comes to energy demand in general—and IT in particular—estimates are often too high and rarely too low. We address this to some extent in the study, so it’s worth taking a look…  

Fabien Lüthi, Communications, Swiss Federal Office of Energy
Image: iStock and Evi Fragkolia