Artificial intelligence (AI) is no longer a topic for the future – it is a reality in almost all industries. Whether in medicine, industry, finance or telecommunications, AI systems analyse data, make decisions, automate processes and create new business models. But this technological revolution also comes with an ecological challenge: the increasing energy consumption of AI applications.
Generating an AI image requires as much energy as half a mobile phone charge. The computing power required by modern AI models is enormous – especially when training large models. According to estimates, training GPT-3, for example, generated over 550 tonnes of CO₂ – comparable to the annual CO₂ emissions of over 100 cars. A 2024 McKinsey study expects the energy requirements of data centres in Europe running AI applications to rise sharply by 2030: to 150 terawatt hours by 2030. That is about five per cent of total European electricity consumption.
At the same time, AI also offers many opportunities to find sustainable solutions to the challenges of our time: for example, AI can be used to significantly improve weather forecasts or detect natural disasters such as landslides at an early stage. “AI can significantly reduce the energy consumption of factories, make buildings more CO2-efficient, reduce food waste, or minimise the use of fertilisers in agriculture”, emphasises Bitkom CEO Bernhard Rohleder.
AI can therefore be both part of the problem and part of the solution. The challenge is to exploit the great opportunities offered by AI applications while minimising their impact on the environment.
As one of Europe's leading telecommunications providers, Deutsche Telekom and T-Systems are also aware of this challenge and have therefore introduced the „Green AI Principles“. These principles serve as a guide for how AI solutions can be developed and used in a more environmentally sustainable way They show how potential risks – such as a rapidly growing carbon footprint – can be identified and addressed at an early stage. The aim is not only to make the development and use of AI more sustainable within Deutsche Telekom, but also to provide impetus for other players, including companies, public institutions, political decision-makers and the scientific community. The approach aims to embed sustainability as an integral part of AI systems from the outset.
The challenge is to exploit the great opportunities offered by AI applications while minimising their impact on the environment.
Nils Henrik Muthmann, Program Lead Sustainability/ESG at T-Systems International
All of Telekom’s AI applications are powered by electricity from renewable sources. This applies not only to our own data centres, but also to outsourced cloud infrastructures. This ensures that the operation of AI systems does not contribute to an increase in CO₂ emissions.
Models, data, software components and hardware are used multiple times. Reusing already trained models or existing data pipelines not only saves development time, but also significantly reduces energy consumption.
Every AI development is analysed in terms of its energy consumption and carbon emissions. This transparency creates the basis for informed decisions – for example, whether a model should be further optimised or replaced by a more efficient one.
Computing resources are scaled according to demand – i.e. they are only provided when and to the extent that they are actually needed. This prevents oversized systems from consuming unnecessary energy, for example through idle times.
AI models are developed in a modular, efficient manner and tailored to the specific application. Instead of universal “one-size-fits-all” models, specialised, resource-saving architectures are used that only do what is really needed.
Company-wide transparency and the reuse of code and models prevent duplicate developments. This not only saves energy, but also promotes internal collaboration and speeds up innovation.
Energy-efficient programming is promoted – for example, through the economical use of memory, computing cycles and data access. Developers are made aware of the impact of their code on energy consumption and sustainability.
Algorithms and architectures are kept as simple as possible. Complexity is only used where it brings real added value – this not only reduces energy consumption, but also increases the maintainability and robustness of the systems.
Sustainability is considered throughout the entire life cycle of an AI project – from the idea to training to operation and decommissioning. This ensures that ecological aspects are taken into account holistically rather than selectively.
Implementing the Green AI Principles is not purely a technical issue – it requires a holistic rethink of organisation, culture and processes. After all, principles alone do not change the world. To minimise their impact on the environment, Deutsche Telekom and T-Systems are pursuing a multidimensional approach:
Green AI is not a contradiction – it is a necessity. The Green AI Principles are a smart approach to combining technological innovation and environmental responsibility. Companies that invest in sustainable AI today not only ensure regulatory resilience, but also gain a decisive competitive advantage.
