Environmental Sustainability and Architecture of Artificial Intelligence Technologies

Delivering a speech on the environmental sustainability of artificial intelligence at the GOTO Copenhagen conference, Charles Humble drew attention to the IT sector's share in global emissions and argued that sustainability must be one of the top-level architectural concerns. The report cited data from the International Energy Agency while noting that electricity consumption and carbon emissions of tech giants such as Microsoft and Google have increased in line with AI demands.

While the IT sector is held responsible for 2 to 4 percent of global emissions, data from the International Energy Agency (IEA) shows that the total electricity consumption of Amazon, Microsoft, Google, and Meta more than doubled between 2017 and 2021, reaching approximately 72 terawatt-hours in 2021. According to Microsoft's 2024 environmental report, total carbon emissions have increased by 30 percent since 2020 due to data center construction. On Google's side, emissions increased by about 50 percent compared to 2019, and by 13 percent year-over-year in 2023 alone. The IEA estimates that global data center electricity consumption in 2022 was between 240 and 340 terawatt-hours, and network consumption was between 260 and 360 terawatt-hours.

Charles Humble suggested using pre-built smaller models when integrating artificial intelligence into business processes, performing operations in regions where low-carbon electricity is abundant, and conducting training with renewable energy. Stating that training conducted entirely with renewable energy can reduce carbon emissions by 45 to 99 percent, Humble emphasized that sustainability must become a top-tier evaluation criterion alongside cost, performance, security, and reliability in IT workloads.