AI Is Fueling Emissions—But It's Also Our Best Bet to Cut Them
While the journey to becoming Net-Zero cannot be complete without the contribution of Artificial intelligence (AI), this transformative force for sustainability also comes with its own rapidly-expanding carbon footprint. AI is at the heart of conversations regarding climate change. While it is a technology that has the power to help bring about climate conversations, it can also exacerbate environmental challenges.
The Carbon Cost of Intelligence
Training large AI models demands substantial energy. For instance, a University of Massachusetts research states that several common large AI models can emit more than 626,000 pounds of carbon dioxide. This monumental figure is nearly five times the lifetime emissions of the average American car, including the manufacture of the car as well.
Why does this happen? This is because the infrastructure and mechanism that AI functions on is what increases its footprint. Data centers that are integral in AI operations, accounted for about 2% of global electricity demand in 2023. This number is projected to rise to 4.5% by 2030. Companies like Google have experienced a 13% year-over-year increase in emissions, reaching 14.3 million metric tons in 2023, primarily due to AI-related energy consumption .
AI as a Tool for Emission Reduction
Despite its energy demands, AI holds promise for mitigating emissions across various sectors. AI also offers tools to mitigate its environmental impact. In order to come up with effective ways to cut emissions, vast datasets need to be analyzed, and this is what AI is known for. Climate-induced risks can be mitigated better with the help of algorithms. This is why scaling up AI operations can cut more emissions than what data centres generate. A few examples of this includes the fact that in building management, AI can help to optimize heating, ventilation as well as air conditioning (HVAC) systems. A Manhattan-based case study concluded that the integration of AI tools helped bring about a 15.8% reduction in HVAC energy consumption, saving $42,000 annually and cutting 37 metric tons of CO₂. Talking about the shipping sector, AI-powered underwater robots like Hullbot have reduced ferry fuel consumption by 13% by preventing biofouling, thus lowering resistance and emissions .
Photo by Alexander Sinn on Unsplash
Navigating the Dual Role of AI
AI, at this point seems like a double-edged sword. A research, published in the journal Nature, also states that AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers. Therefore, while it is believed that GPT‑3 (2020) emitted ~588 tons of CO₂, GPT‑4 (2023) ~5,184 tons, and Llama 3.1 405B (2024) ~8,930 tons, it also non-negotiable when it comes to moving towards sustainability.
What is the way out? A balanced approach. It becomes imperative at this point to harness AI's potential for sustainability. This can be done in the following ways:
Energy-Efficient AI Development: The creation of AI models that require less computational power without compromising performance should be prioritized.
Renewable Energy Integration: The adopting of renewable energy sources when it comes to powering data centers can go a long way to offset the carbon footprint of AI operations.
Lifecycle Emission Tracking: Implementing comprehensive tracking of emissions throughout the AI lifecycle, from development to deployment.
Policy and Regulation: Establishing guidelines that encourage sustainable AI practices and hold organizations accountable for their environmental impact.
Therefore, we see that AI is at a pivotal juncture in the climate action journey. While it holds immense potential to revolutionize efficiency and accelerate decarbonization, its growing environmental footprint demands urgent attention. The key lies in harnessing AI responsibly: through energy-efficient development, renewable-powered infrastructure, and robust lifecycle emission tracking. Fitsol, a number one decarbonization partner empowers businesses to measure and optimize the climate impact of their AI-driven operations, ensuring technology becomes a powerful enabler—not a barrier—on the path to Net Zero.
Citations
University of Massachusetts research: https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/
Source Url: https://fitsol.green/resources