Rising AI Emissions and What to Do About Them

There are reports that Google and Microsoft each consumed more power than some countries in 2023 . That is 24 Terawatt-hours (TWh) each of electricity, which is more than that used by 100-plus countries in the same year. That’s about the same as Azerbaijan’s electricity usage, but higher than the 19 TWh consumed by Iceland, Ghana, the Dominican Republic, and Tunisia. It’s only slightly lower than the 25 TWh that Libya used and the 26 TWh of Slovakia.

Our demand for AI has played a big part in it. Google has reported a 48% increase in its carbon emissions over the past five years, largely attributed to the growing energy demands of its AI operations. This increase highlights the significant challenge tech companies face in balancing technological developments with environmental sustainability.

The Paradox of AI and Energy Efficiency

Despite substantial efforts to reduce the carbon footprint of AI, the rapid growth and deployment of it have led to an overall increase in energy consumption. Here’s how tech companies are striving to mitigate these impacts while continuing to innovate.

Innovations in AI for Sustainability

AI companies are implementing various strategies to enhance the energy efficiency of their operations:

• Specialized AI chips: Google’s Tensor Processing Units (TPUs) are specifically designed for machine learning, making them more efficient than traditional processors. The latest Trillium TPUs, for example, are said to be over 67% more energy-efficient than the previous generation. Similarly, NVIDIA’s A100 GPU is designed to deliver higher performance while consuming less power compared to older models, making it an energy-efficient option for AI tasks .
• Energy-efficient data centres: Google’s data centres use advanced cooling technologies and renewable energy integration. Microsoft has been investing in two phased liquid immersion cooling in its data centres.
• Model compression and efficient algorithms: Techniques like model pruning, quantization, and efficient algorithm design help reduce the complexity and size of AI models, thereby lowering the computational power required to run them .

Key AI Projects for Emissions Reduction

There are ways to offset the increased energy consumption of technology by reducing usage elsewhere. Several AI-driven projects aim to reduce emissions and enhance sustainability in different ways:
• Project Green Light: Google’s initiative uses AI to optimize traffic signals in cities, reducing stop-and-go traffic and emissions .
• Other technology vendors have been working on applications of AI to optimize delivery routes to reduce fuel consumption and emissions by identifying the most efficient paths and avoiding traffic congestion.

Renewable Energy Integration

Tech companies are committed to powering their data centres sustainably. For instance, Google aims to use 100% carbon-free energy 24/7 by 2030, although its recent announcement about the rise in its emissions shows it is a tough target to achieve. The ambition involves substantial investment in renewable energy sources such as wind, solar, and advanced energy storage solutions. Another industry giant, Amazon Web Services (AWS), is on a path to power its global infrastructure with 100% renewable energy by 2030.

The recent reports about Microsoft and Google energy consumption and Google’s emission targets underscore the complex challenge of managing the environmental impact of AI . While significant strides have been made in improving energy efficiency, the rapid expansion of AI technologies continues to drive up energy consumption. It is clear that ongoing innovation and commitment to sustainable practices are crucial as we navigate the future of AI and its role in our world.

Sarah Burnett is the Technology Evangelist at KYP.ai and Industry Analyst