Publications

Green Clean Software Sustainability

Authors: Luís Cruz, Patricia Lago, Henry Muccini, Eoin Woods

Published in: IEEE Software (pp. 25–31).

Abstract: The energy footprint of software-intensive systems poses a significant concern. Energy-hungry software, such as blockchain applications and cryptocurrencies, the pervasive integration and usage of central cloud and edge services and applications, along with AI-enabled systems, contribute to this issue. To better understand the scale of the problem and the fact that this is not getting better, let us consider a few examples: GPT-4 training required approximately 5 gigawatt hour (GWh) (including computing-only estimates), and up to 50 GWh when considering the full data center overhead and all GPUs used for the training with an energy inference cost close to ∼ 0.34 Wh per query. GPT-5, comparatively, required an estimate of 3.5 GWh for training (including computing-only estimates) and can consume 18–40 Wh per query (for a medium-length GPT-5 response of ∼ 1,000 tokens), which is 50–100× higher than GPT-4. As another example, Bitcoin's total annual energy consumption reached 173 TWh in 2025 with a crypto energy usage per transaction of approximately 1,135 kilowatt hour (kWh) compared to the 84 kWh per transaction used by Ethereum. In addition, the global digital transformation of all industry sectors is accelerating the steep increase in software energy demands. All this has a direct impact on the energy consumption of data centers, estimated to be 415 TWh in 2024 and expected to grow globally to 945 TWh per year by 2030. Green clean software pertains to the minimization of the energy needed to execute and use software-intensive systems. Adopting renewable energy resources to "feed" software execution is simply not enough, so reducing the carbon footprint must go hand in hand with minimizing the energy footprint. On the other hand, software-intensive systems may be used to support green processes that aim to reduce the environmental impact of the sector, and, in fact, of any industry sector, of society, and planet Earth. Examples include software supporting the production and consumption of renewable energy resources, smart software for green-oriented behavioral change (e.g., adopting green public transportation and sustainable work practices), and the combination of energy optimization and digitalization (so-called twin transition). In addition, software sustainability from an environmental perspective may also concern software engineering and its processes: the energy used to develop, evolve, and maintain software-intensive systems is nonnegligible and also needs to be addressed.

@article{cruz2026greenclean,
author={Lu\'{i}s Cruz and Patricia Lago and Henry Muccini and Eoin Woods},
title={{Green Clean Software Sustainability}},
journal={IEEE Software},
year={2026},
volume={43},
number={2},
pages={25--31},
month=mar,
doi={10.1109/MS.2025.3647131},
issn={1937-4194},
publisher={IEEE Computer Society},
address={Los Alamitos, CA, USA}}

Read me: DOI: 10.1109/MS.2025.3647131.