AI has great potential for sustainability, but its energy requirements cause emissions equivalent to Sweden's annual energy consumption. By developing energy-efficient AI, we can not only reduce the technology's footprint, but also create innovative solutions that support the climate transition.
Artificial Intelligence (AI) has great potential to contribute to sustainable solutions and energy efficiency in businesses. At the same time, its energy consumption is a key challenge for the technology as it seeks to drive a low-carbon future. Today, AI systems account for 0.1 percent of global greenhouse gas emissions, equivalent to the annual emissions of Sweden. By developing energy-efficient AI systems, society can reduce the climate impact of technology and pave the way for innovative solutions that support a sustainable transition.
"ChatGPT uses less energy to generate a page of text than a human working on a regular laptop," says Olof Mogren, research leader in AI at RISE. "However, every source of emissions is a source to be considered, and the use of AI can reward behaviours that increase energy consumption and greenhouse gas emissions."
AI requires not only hardware investment, but also energy-efficient programming. Many systems and software in use today carry technical 'baggage' from previous generations of code, resulting in high energy consumption.
Efficient AI algorithms can reduce energy consumption
"We see that small improvements in design and programming can lead to significant energy savings," continues Olof Mogren. By developing new algorithms and optimising existing code, the energy consumption of AI can be dramatically reduced.
RISE is working to develop more efficient AI algorithms that use less computing power and energy, but still perform as well or better than their predecessors. RISE is also actively researching how data centres can be built and operated in a sustainable way.
"Despite the energy and emissions generated by AI systems, the real potential of AI is far greater," says Olof. "By using AI to optimise energy consumption and reduce carbon emissions in areas such as material production, energy management and transport, we can achieve significant climate benefits that far exceed the emissions generated by the AI technology itself. This is especially true with the increased use of AI systems with energy-efficient code.
Effective AI is about combining intelligent algorithms with optimal use of computing power and learning data
Efficient AI: intelligent algorithms and optimised learning processes
RISE is working on several initiatives and research projects in the field of AI and climate change, with a focus on streamlining learning processes for AI models.
"Effective AI is about combining intelligent algorithms with optimal use of computing power and learning data," says Olof Mogren. The goal is to make the model learn as efficiently as possible within a reasonable data and computational budget. One example is graph neural networks for flow computations relevant to process industries, transport and the environment. Traditional methods, based on physical equations, require large amounts of computing power. By training graph neural networks with techniques that cleverly exploit equivariance for rotations and translations, deep learning neural networks can learn faster with less data and computational power.
Nordic Workshop on AI and Climate Change
RISE is a driving force in networks such as CLIMES and Climate Change AI Nordics. Together with RISE, the organisations are organising the 2025 Nordic Workshop on AI for Tackling Climate Change on 13 May in Gothenburg, where researchers from the Nordic countries will gather to discuss the impact of AI on climate change.
