AI’s Growing Energy Demand and Its Environmental Impact

Artificial intelligence is becoming a big part of our everyday lives, changing the way industries work and helping things run smoothly. However, the toll it takes on the environment is often ignored. If we’re serious about building a sustainable future, the U.S. needs to start putting some limits on how AI is used.

The damage caused to the environment from electricity intensive technology is immense. From start to finish, running an AI model requires a lot of energy to keep the systems going. This energy comes from electricity, and a lot of it is used during the three main stages of a machine learning model's life: training, development, and inference.

Training relies on deep learning and runs through data in repeated cycles called epochs. Each of these cycles takes about three times more computing power than the inference stage. Meaning, training a model burns through more energy than just using a model that's already been built.

Even though training doesn’t happen as often as inference, it still happens regularly and uses up even more electricity. Development and tuning are similar to training. They also use deep learning and neural networks to build different versions of models. However, while the development phase is rare, it uses the most energy overall.

Inference is the least energy-demanding step, but it happens constantly. So even though each use doesn’t take much energy, the total can be huge if the model is used billions of times a day. In some cases, inference can actually end up using more energy than training. For example, ChatGPT-3's inference stage uses more power than training.

In 2023, Alphabet’s chairman said using AI in Google searches could use 10 times more power than regular searches. In 2024, Google added an AI overview to the Google search engine.

All this electricity demand is putting major stress on the U.S. power grid. Data centers are becoming more common and they’re overloading systems built for regular community use. The International Energy Agency says data centers could use 1,000 terawatt-hours a year by 2026, about the same as all of Japan. That kind of load can make the grid unstable, causing power surges, blackouts, and even fires, especially in dry areas where wildfires are already a problem.

AI's energy usage becomes a major issue because most of our electricity still comes from burning fossil fuels. These fuels, such as coal, oil, and natural gas, pollute the air and water. They also contribute to serious health problems like respiratory disease and release greenhouse gases such as CO2, methane, and nitrous oxide. These gases trap heat in the atmosphere and lead to extreme weather and global warming. It is because of these environmental hazards that the U.S. should regulate the usage of AI and begin to shift to eco-friendly solutions, such as Green AI.