The AI Paradox: Balancing Data Center Energy Costs with Climate Innovation at Climate Week NYC

• Energy Consumption Reality: AI currently accounts for 1.9% of global electricity usage and 0.9% of total global energy consumption, significantly lower than the commonly cited estimate of 60 times that figure.
• Future Climate Gains: The International Energy Agency (IEA) estimates that by 2035, AI applications could save 1.4 gigatons of CO2 emissions, a "handprint" three to five times larger than its current carbon footprint.
• Scientific Breakthroughs: Machine learning has successfully improved the efficiency of the photosynthesis enzyme rubisco by 50%, demonstrating AI’s role in accelerating biological discovery.
• Computing Power Surge: According to the London School of Economics, AI computing power usage increased tenfold between 2018 and 2022, raising concerns about greenhouse gas emissions given that fossil fuels still generate over 60% of global electricity.
• Industrial Efficiency: Siemens technology can reduce building emissions by 30%, while AI integration can manage 30% of electrical loads with 99.9% availability for public transportation systems.

The discussion at Climate Week NYC highlighted a paradox: while the demand for computing power surges, often reliant on fossil fuels, AI offers potential solutions to environmental crises. Angela Barranco, Executive Director of the Climate Group’s North America division, opened the panel by noting the enormous investment flowing into AI infrastructure alongside valid concerns about its environmental impact. She challenged the audience to consider whether the benefits are emerging quickly enough to justify such a massive build-out.

Dr. Eva Riesenhuber, Global Head of Sustainability at Siemens, urged attendees to look at both sides of the ledger. She clarified that while public perception often exaggerates AI's energy use, the actual figures are more manageable. Currently, AI utilizes approximately 0.9% of global energy and 1.9% of global electricity. Riesenhuber noted that many people mistakenly believe the figure is sixty times higher than reality.

Looking toward the future, Riesenhuber cited IEA estimates predicting that AI could save 1.4 gigatons of CO2 by 2035. This potential environmental benefit, or "handprint," is projected to be three to five times greater than its current footprint. She emphasized that Siemens aims to leverage this capability further. By utilizing their technologies, the company can cut building emissions by 30%. Furthermore, AI allows for managing 30% of electrical loads while maintaining 99.9% availability for public transportation, thereby leveraging electrification to reduce the overall footprint of AI infrastructure itself.

Dr. Evelyn Wang, Vice President for Energy and Climate at MIT, argued that the most promising climate breakthroughs from AI lie in scientific discovery. She explained that in fields with expansive design spaces—such as catalyst discovery, advanced materials, battery technologies, and biological systems—traditional trial-and-error experiments can take years or decades without yielding optimal solutions.

Wang highlighted a specific example involving photosynthesis. The enzyme responsible for this process, rubisco, is crucial yet notoriously inefficient and slow. Wang’s colleague, Matt Shoulders, and his team utilized machine learning to identify new mutations in the enzyme. Through experimental validation, they demonstrated that engineering this enzyme could make it 50% more efficient. This achievement illustrates how AI serves as a powerful tool for discovery, enabling breakthroughs in scientific underpinnings that are essential for addressing climate challenges.

The panel also addressed the physical infrastructure supporting AI. The London School of Economics reports that computing power usage for AI grew tenfold from 2018 to 2022. With fossil fuels still providing over 60% of global electricity, this rising demand poses a risk of increased greenhouse gas emissions.

Wang suggested distinguishing between AI as an analytical tool and the data centers that house it. She described AI as a powerful instrument for enabling discovery and optimizing energy grids, whereas data centers represent the physical infrastructure required to support it. While acknowledging the resources necessary for these facilities, she argued against conflating the two concepts.

Riesenhuber reinforced this perspective from an industrial standpoint. She acknowledged that while the environmental costs of supercomputing are significant, the intelligence generated is indispensable for navigating global sustainability issues. "We are racing to mitigate them as fast as possible because we are hooked on the intelligence," she stated. "We do need the intelligence to design the future that this planet needs." She concluded that providing peace and prosperity requires ensuring everyone can develop, and AI is necessary to help with these challenges.