Software architecture has a greater impact on energy consumption and CO₂ emissions than many people realize. In this interview, Peter Kutschera explains how to design energy-efficient systems – and why this not only benefits the environment but also makes economic sense.
IT is often not directly associated with climate protection. What role does software architecture play in the context of energy consumption and CO₂ emissions?
Currently, around 4 % of global electricity consumption is used by IT. With an appropriate software architecture tailored to user needs, significant savings can be achieved.
There are many architectural patterns – such as caching, event-driven architectures, or batch-oriented data processing – that can help reduce the resource consumption of applications.
In addition, newer approaches such as peak shaving (balancing peak loads over time) or time and location shifting (running applications when and where green energy is available) can further improve energy efficiency.
In your session, you demonstrate how energy consumption and CO₂ emissions can be measured. How can teams use this knowledge to make better architectural decisions?
There is a well-known quote by Peter Drucker: “You can’t manage what you don’t measure.” Project teams should therefore define business scenarios, analyze their energy consumption, and optimize them accordingly.
I’ve been in the industry for quite some time – and in the past, when resources were more limited, software efficiency was taken much more seriously. Today, performance requirements are often addressed by simply adding more hardware instead of optimizing the software or its operation.
Many companies are under cost pressure. How can an energy-efficient IT architecture deliver both ecological and economic benefits?
This is something many organizations overlook: efficient software requires fewer resources and therefore reduces operational costs, as less hardware is needed.
In addition, efficient software typically leads to improved response times, which in turn increases user satisfaction.
Cloud platforms now offer opportunities for more CO₂-efficient operation. What should architects consider to fully leverage this potential?
It starts with selecting the right cloud platform. Providers such as Scaleway or OVHcloud often place a stronger emphasis on sustainability than some hyperscalers.
Where regulations allow, it can also make sense to choose regions with a higher share of renewable energy (e.g. Norway) and run applications there.
Dynamic scaling can help avoid overprovisioning, while unused resources can be automatically shut down (often referred to as scale-to-zero or “LightSwitchOps”).
Using ARM instead of x86 processors or leveraging spot instances in the cloud can further reduce energy consumption.
With the growing use of AI, energy consumption is increasing significantly. What approaches do you see for making AI systems more energy-efficient without sacrificing performance?
Currently, large language models (LLMs) are widely used when discussing AI. The first question should always be whether an LLM is actually the right solution for the problem at hand – and this should be reflected in an explicit architectural decision.
The next step is to evaluate which model best fits the use case: do you really need the largest and most powerful model, or would a smaller, more specialized model be sufficient?
Another important factor is training. Do you need to train a model yourself, or can you use pre-trained models, such as those provided by Hugging Face?
Hugging Face also provides an AI Energy Score for many models, making it easier to identify energy-efficient options: https://huggingface.co/spaces/AIEnergyScore/Leaderboard
What are common misconceptions or mistakes you see in practice when it comes to Green Software?
A common misconception is that measuring the CO₂ emissions of software is too complex, which often leads teams to skip measurement altogether. However, proxy metrics such as CPU, memory, and network usage can already provide valuable insights into energy efficiency.
The less energy a system consumes, the lower its CO₂ emissions – assuming that the energy is not entirely sourced from renewables.
Another frequent misconception is that Green Software requires highly efficient programming languages such as Rust or C. This is not the case. Efficient software can be developed in any modern programming language. The key is to identify inefficiencies in the code and address them – and to consider the efficiency of libraries and frameworks being used.
More on this topic at the Software Architecture Forum 2026: There, Peter Kutschera will show in his session how to design and implement energy-efficient software architectures in practice.
