Software-related CO2 emissions account for 4-5% of global emissions. This is equivalent to the emissions of all aviation, shipping, and rail combined.

The current gold standard paper on the climate and transformative impact of ICT indicates that precise estimates of software-related CO2 emissions vary according to methodology, but generally agree as to the main sources: digital devices, data centers, and networks. Software development itself can be energy intensive, such as training LLMs (large language models) and other AI models. This is why Hugging Face calculates the carbon cost of AI using a life cycle analysis, taking into account both training and use over time along with the larger computing infrastructure.

According to Harvard Business Review, by 2040, software is expected to account for 14% of the world’s carbon footprint. It is clear that reducing emissions from software will be a crucial part of meeting the targets outlined by the 2022 Intergovernmental Panel on Climate Change (IPCC) report, “Mitigation of Climate Change.”