AI’s dirty secret: How data centers are polluting the planet
Researchers warn that AI’s environmental impact is larger than previously thought.
The rapid expansion of Generative Artificial Intelligence (GenAI) is projected to cause $7.6 billion in public health damage in the United States by 2023, due to the reliance on fossil fuels to power energy-intensive data centers, according to a recent study by researchers at the University of California. The study warns that, as demand for AI-powered services surges, healthcare costs linked to emissions from AI infrastructure could exceed $20 billion per year by 2030.
GenAI models, such as OpenAI’s GPT, require vast computing resources to train and operate. Even after the DeepSeek moment, which demonstrated that some advanced models can be trained with significantly fewer computing resources, the prevailing consensus is that AI-related energy consumption will continue to rise sharply in the coming years.
Despite efforts by major technology companies to transition to renewable energy, the unprecedented demand for AI computing power is outpacing green energy adoption, forcing many data centers to rely on fossil fuels. Most U.S. data centers today are connected to power grids dependent on coal and natural gas and also use diesel generators as backup power sources. Additionally, the production of AI hardware, particularly high-performance processors, generates significant emissions that contribute to environmental and health risks.
The Hidden Health Cost of AI Expansion
A comprehensive emissions analysis conducted by the researchers found that data center pollution had severe health consequences in 2023. Their findings include:
- 490 deaths in the United States due to pollutant exposure from AI-related data center operations.
- $7.6 billion in public health damage, a cost equivalent to more than a third of all vehicle emissions-related damage in California.
- By 2030, if current trends persist, data center emissions could lead to 600 new asthma cases and 1,300 premature deaths annually.
The study also quantified the carbon footprint of training large-scale AI models. According to the researchers, training Meta’s Llama 3.1 model alone is estimated to generate emissions equivalent to 10,000 round trips between New York and Los Angeles.
As AI continues to revolutionize industries, its environmental and health impact is becoming harder to ignore. While data centers are essential for AI development, experts warn that unless clean energy adoption accelerates, the toll on public health and the environment will grow exponentially.
