Three questions and answers: Is generative AI a climate sow?
Generative AI shows promise, but high energy demands and unclear emissions raise concerns about its sustainability and environmental impact in the long run.
(Image: iX)
Generative AI not only requires huge amounts of data, but also a lot of electricity. Depending on the energy source, there are also COâ‚‚ emissions. Some industry leaders, such as OpenAI boss Sam Altman, are extremely relaxed about this. According to him, humanity will have unlimited energy available from 2030 anyway. However, an AI-using company that already has to account for its environmental footprint today can hardly rely on such prophecies. Boris Ruf, cover author of the new iX 7/2025, explains the situation and what to look out for.
AI is currently regarded as the technology of the future. Are we pinning our hopes on an energy-guzzling climate killer?
Generative AI is actually a very energy-intensive technology. On the one hand, creating AI models requires a lot of computing power, but their operation also consumes large amounts of electricity. Market observers such as the International Energy Agency therefore anticipate a massive increase in energy requirements for this sector.
The environmental impact of this development naturally depends on how the additional electricity required is generated. The major IT companies are supporting the expansion of renewable energies by investing in numerous new energy projects in this area. However, the energy requirements of data centers are already high today and they need security of supply around the clock. However, wind and solar energy are not permanently available, and battery technology is not yet mature enough to bridge any lulls. It will therefore be a major challenge to minimize the impact of the AI boom on the environment.
Which has the bigger power hunger and carbon footprint: inference or model training?
Initially, the focus was primarily on training AI models. This involved processing huge amounts of data over weeks in computing clusters, which requires a considerable amount of resources.
Meanwhile, however, the focus has shifted to the energy requirements of the AI during operation. Billions of parameters are activated with every query to an AI model. Compared to a classic database query, this is a very computationally intensive process. Considering the high volume of queries –, not least due to the increasing integration of AI into a wide variety of processes –, the energy required adds up considerably.
The major providers of proprietary AI models tend to present themselves as black boxes, even when it comes to questions of energy consumption and environmental impact. What should you do as a company if you still need figures for your sustainability reporting? And what tools are available for the open-source models?
As far as the energy consumption and emissions of proprietary models are concerned, the official data situation is unfortunately very thin indeed. Nevertheless, you should at least try to estimate the environmental impact of your own AI projects. We have modeled corresponding COâ‚‚ estimates for various AI models and published an online calculator on this topic. The EcoLogits project also provides tools that can be used to log the estimated carbon footprint of proprietary AI models.
Videos by heise
Those who use open AI models have a better chance of recording their carbon footprint. On the one hand, there are benchmarks such as the AI Energy Score project from Salesforce. By systematically measuring the respective power consumption, a series of freely available AI models can be effectively compared. If the AI is running in your infrastructure, you can even use a tool like CodeCarbon to determine the power consumption yourself.
Martin, thank you very much for your answers! You can find an overview of AI's hunger for energy in the new iX. We also show what the use of large AI models costs and how the carbon footprint of AI projects can be determined. Readers can find all this and many other topics in the July issue, which is available now in the heise Shop or at newsagents.
In the series "Three questions and answers", iX aims to get to the heart of today's IT challenges – regardless of whether it is the user's view in front of the PC, the manager's view or the everyday life of an administrator. Do you have any suggestions from your day-to-day work or that of your users? Whose tips on which topic would you enjoy reading in a nutshell? Then please write to us or leave a comment in the forum.
(axk)
