Does AI development threaten climate goals with its power demand?
Modern technology sometimes has unexpected impacts on the climate goals. For example, recent articles in the Financial Times make us aware that power demand growth in the US leads to slower retirement of aging coal power plants. Here is one of the articles focused on the issue AI’s thirst for electricity risks slowing US coal phaseout (ft.com)
The rapid pace of AI development seems to present a dilemma for the US administration, balancing the importance of leading the AI race with the need to meet climate goals. According to the International Energy Agency, the AI application ChatGPT consumes nearly ten times more electricity than Google Search. Meanwhile, an estimated 54 gigawatts of US coal-powered generation assets, about 4 percent of the country’s total electricity capacity, are expected to be retired by the end of the decade. The projected power demand is now putting these retirement plans at risk.
Joe Craft, CEO of Alliance Resource Partners, one of the largest US coal producers, expressed concerns highlighted by the Financial Times: “You can’t replace the fossil plants fast enough to meet the demand. In order to be a first mover on AI, we’re going to need to embrace maintaining what we have.”
Strategies, a consultancy, projects a 4.7 percent growth in US electricity demand over the next five years, nearly double its estimate from a year earlier. This surge is driven by new manufacturing and industrial capacity, as well as data centers powering AI, crypto mining, and cloud services. A study released by the Electric Power Research Institute found that data centers are expected to constitute 9 percent of US power demand by 2030, more than double current levels.
“We need more energy, not less,” Indiana’s Republican governor, Eric Holcomb, told the Financial Times. “We absolutely, as Americans, can’t afford to lose the AI war.”
Indeed, the computing power required for AI is growing exponentially with each new application. The following diagram from The Economist illustrates this issue.
However, the Environmental Protection Agency (EPA) found in its analyses that the power sector can meet demand while reducing pollution and providing reliable, affordable electricity under these rules. Multiple other studies support the feasibility of satisfying demand without compromising climate goals. Here are a few strategies, for example:
- Optimized Data Center Locations: Data centers can be located in areas with lower fossil fuel production, reducing their carbon footprin
- Dedicated AI Hardware: Using specialized AI hardware with improved efficiency for training AI models, the most power-consuming part of AI operations, can significantly cut energy use.
- Photonic AI Chips: These microchips use light instead of electricity for computations, potentially making them faster and more energy-efficient than conventional electronics. Additionally, photonic AI chips are considered more secure against hacking.
This technology is still evolving but promises substantial benefits for both AI development and climate goals.
Keep in touch and sign up to our Newsletter
