The IT industry accounts for about 2% to 4% of global emissions, and public reports from Microsoft and Google show electricity use at both companies continuing to rise because of artificial intelligence. Speaking at GOTO Copenhagen on green AI, Charles Humble argued that sustainability should be a top-level architecture concern. His suggestions include using pre-built and smaller models, running compute where low-carbon electricity is plentiful, and training with renewable energy.
Humble cited International Energy Agency (IEA) figures: combined electricity use by Amazon, Microsoft, Google, and Meta more than doubled between 2017 and 2021, reaching about 72 billion kilowatt-hours in 2021. Microsoft’s 2024 environmental report said its total carbon emissions had risen 30% since 2020, mainly because of new data centers built to meet AI demand. According to Google’s own report, its emissions have surged about 50% since 2019 and rose 13% year over year in 2023 alone, driven primarily by AI. Those reports are mainly about carbon dioxide.
Humble said greenhouse-gas concentrations in the atmosphere have been increasing since the Industrial Revolution. The aim is to limit warming to 1.5 degrees by 2100, but projections indicate about 2.5 to 3 degrees. He described IT as a significant contributor to the problem, partly because of electricity use. The IEA estimates that global data centers used about 240 to 340 terawatt-hours in 2022, or 1% to 1.3% of global electricity demand, while communications networks added another 260 to 360 terawatt-hours. Together that is about 700 terawatt-hours, roughly comparable to Brazil’s 726 terawatt-hours over the same period.
He put IT-sector emissions at about 2% to 4% of the global total and aviation at 3%. Aviation, he said, is actively cutting emissions, while the technology industry is moving in the opposite direction.
Before introducing AI in a business setting, Humble suggested asking:
- Do you really need it?
- Do you need to train a new model from scratch, or can you use a pre-built model?
- If some SLAs can be relaxed, can the workload be deployed in a lower-carbon region?
- Can you use demand shaping?
If a custom model does have to be trained, he said teams should first assess how much data they actually need to collect. In many cases, he argued, a smaller model works better. He also suggested using open-source datasets and, where possible, shifting data collection to times and places where green energy is available. In theory, training entirely on renewable energy could reduce carbon emissions by 45% to 99%.
He said most energy use actually occurs once a model is deployed and kept in operation, and offered four ways to make compute greener:
- Use the smallest hardware configuration that can run the job safely.
- Run workloads where low-carbon power is abundant and the grid has a clear decarbonization plan.
- Choose cloud providers that have data centers in greener regions and provide solid emissions-reduction tools.
- Optimize when jobs run to further cut the carbon footprint.
In summing up, Humble said sustainability should sit alongside cost, performance, security, compliance, and reliability as a top-level design consideration for compute workloads, and that teams should think about how to build systems in a more environmentally sound way.
Why location matters
InfoQ interviewed Humble after the talk. On why geography matters for cutting emissions, he said the total electricity a compute task uses is fixed, but the carbon cost of that electricity varies greatly by place and time, depending on how the data center is powered. If the data center uses grid power, running a workload in a region that relies mainly on hydro, solar, nuclear, and wind—such as Norway, Sweden, or France—can be many times cleaner than in a region that depends on coal or natural gas. Carbon intensity also changes through the day and the year with wind, solar, and demand.
That variation, he said, supports demand shifting in carbon-aware computing, either by time or by location. Time-based shifting is especially suited to latency-insensitive work such as AI training. The simplest location-based approach is to place new latency-tolerant workloads in low-carbon regions by default. A more advanced option is to move workloads between hemispheres with the seasons to make more use of sunlight and solar power.
Demand shaping
Humble described demand shaping as adjusting a software workload, or the type of task it runs, according to how clean power is at the moment and how tight supply is. One approach is to give users a choice: for an AI-driven service, they might talk to a person first instead of an AI system, or switch to a person partway through a conversation.
He also described graceful degradation when carbon intensity is higher, such as lower-quality video or images, turning off autoplay, skipping non-essential animation, lowering the refresh rate, or calling a smaller model. The cost is greater system complexity, and the degraded service level still has to be acceptable.
Greener model training
Beyond demand shifting for training and data collection, Humble said teams should avoid training from scratch when they can. Prefer fine-tuning a pretrained model or, if that meets the need, prompting or retrieval-augmented generation (RAG).
Smaller models can train faster and use energy more efficiently. Distillation, quantization, and pruning can compress models. In some cases, on-device training with federated learning is lower carbon. He noted that for large models in wide use, inference emissions over the full lifecycle are often higher, so the energy efficiency of inference also matters.
He pointed readers who want more detail to Raghavendra Selvan’s book Sustainable AI.
Source: InfoQ. The report links to an English original at https://www.infoq.com/news/2026/10/AI-environmentally-sustainable/.