Infinite growth

Christiane Weihe

Artificial intelligence has immense capabilities – but it also poses significant demands. The large amounts of data and multiple computing processes associated with AI require substantial energy inputs. For example, when ChatGPT answers a query, the AI consumes between three and 10 times as much electricity as a traditional search engine. And the development phase of this popular large language model has a large environmental footprint – the training of ChatGPT in Version 3 alone is estimated to have produced 500 tonnes of CO2. Due to the immense popularity of these systems, their power consumption will increase significantly in the coming years. What can be done to address this issue, not least for the sake of the energy transition? The Oeko-Institut is working to answer that question in many of its projects.

According to a forecast by the International Energy Agency (IEA), data centres’ global energy consumption will double to more than 1,000 terawatt-hours annually between 2022 and 2026 due to the AI trend – equivalent to an increase of 21% per year. An end to the trend is not in sight. It is mainly driven by so-called “hyperscalers” – vast data centres operated by corporations like Google, Microsoft and Amazon. “This growth is highly problematical, as the example of Ireland clearly shows. Data centres already account for a fifth of Ireland’s power consumption. This is putting national climate targets at risk and driving up electricity prices for consumers. That’s why Ireland now has an emerging citizens’ movement that wants to prevent the establishment of more data centres,” says Jens Gröger from the Oeko-Institut. In Germany, the share of data centres’ consumption in total electricity demand is currently around 3.5% – on a clear upward trajectory. “We are seeing this development in Frankfurt, for example. It has one of Europe’s largest Internet exchanges, and 20% of the city’s electricity is channelled into data centres.”

All this is happening against the backdrop of a society that is becoming increasingly reliant on electricity. Heat pumps, e-mobility, a hydrogen economy: they need large amounts of electricity as well. “However, in contrast to these examples, where a shift from fossil fuels to renewable energies is under way, we are seeing artificial intelligence generating additional energy demand linked to new types of services. This growth is jeopardising the energy transition,” says Jens Gröger, Research Coordinator for Sustainable Digital Infrastructures at the Oeko-Institut.

The success of their business models is creating a dilemma for the major tech companies as well: they have pledged to be carbon-neutral by 2030. “But it’s increasingly clear that they can’t meet their electricity demand from renewables. That’s why they are bringing nuclear energy into play – an act of megalomania, in my view.” Microsoft, for example, has initiated plans to restart the nuclear power plant at Three Mile Island in Pennsylvania, which was shuttered after a partial core meltdown in 1979. Google and Amazon are turning to small-scale reactors, known as Small Modular Reactors. “But these supposed solutions come with the risks that we’re already familiar with from our use of nuclear power, such as unresolved safety and disposal issues,” says Jens Gröger. “It was at that point, in my view, that digitalisation shed any pretence of innocence. So the question we should be asking ourselves right now, and no later, is this: are we genuinely willing to accept nuclear risks simply to generate a few humorous memes?”

The efficiency option

There are ways to reduce digital infrastructures’ electricity demand through efficiency measures, however. With the Blue Angel for Data Centres ecolabel, researchers at the Oeko-Institut, on behalf of the German Environment Agency (UBA), have identified key areas where leverage can be applied. And with the Blue Angel for Software Products, they have ensured that the energy consumption of digital services can be measured and optimised. In practice, however, voluntary instruments such as these are rarely used. So it’s good news that Germany’s Energy Efficiency Act has established the basis for a national energy efficiency register for data centres. “At present, there’s a lot we don’t know about the actual energy and resource consumption of data centres. We are hoping that mandatory publication of key environmental data will spark a competition for efficiency within the industry.” In the Public Energy Efficiency Register of Data Centres (Peer-DC) project on behalf of the German Federal Ministry for Economic Affairs and Climate Action (BMWK), the Institute’s experts collaborated with the University of Stuttgart (IER) and other project partners on laying the foundations for the register. “We have developed an evaluation system and evaluation software that take various criteria into account, including the efficiency of building technology and the performance of the information technology,” Jens Gröger explains. “We have also analysed whether and how an energy efficiency label for data centres could be rolled out Europe-wide.” In future, this would enable customers to identify which data centre scores best for environmental performance.

The environmental footprint

Making an informed choice: this should also be possible for anyone who uses a computer or a smartphone. In the eco:digit – Enabling Green Computing and Digital Transformation project, the Oeko-Institut is working with the Open Source Business Alliance (OSBA) and software developer Adesso, among others, to develop a methodology for determining the key environmental impacts of digital services. The project is coordinated by the German Informatics Society and funded by the BMWK. “The aim is to calculate an environmental footprint for specific software applications that includes not only greenhouse gas emissions but also resource and water consumption and other environmental impacts.” The project team is therefore creating a simulation environment, known as a “test bench”, which is able to evaluate mobile apps and desktop and cloud-native applications. “The Oeko-Institut has developed a methodology for this purpose, which covers the manufacturing, operation and disposal of the hardware and links it to the related software.” A further objective is to introduce the methodology into the international standardisation process with a view to its broad-scale rollout.

Saving the best for last

When it comes to efficiency, too, artificial intelligence has its upsides: it can optimise technical processes by making the best possible use of renewable energies, for example. “At present, however, there is a complete lack of clarity around the scale of these positive effects and whether they can balance out the increased electricity demand,” says Jens Gröger. “A technology impact assessment is essential to ensure that AI does us more good than harm. Had this been done for nuclear power, for example, we might have dispensed with nuclear altogether because of the associated risks. And that might make sense with some AI applications as well.”

---

Sustainable information and communication technology is the main focus of Jens Gröger’s work. A graduate in energy, chemical and process engineering and the Oeko-Institut’s Research Coordinator for Sustainable Digital Infrastructures, he investigates energy-efficient data centres and IT infrastructures and the environmental impacts of software and cloud services.