AI's voracious appetite for energy threatens climate goals as data centers consume record power

AI Energy Consumption Reaches Alarming Levels

The environmental cost of AI has emerged as a critical challenge to global climate goals, with new data revealing the technology's staggering resource demands. In 2025, data centers worldwide consumed 448 terawatt-hours of electricity, ranking them as the 11th-largest electricity consumer globally if they were a country2

2

3

3

. This figure is expected to more than double by 2030, potentially reaching 945 terawatt-hours annually3

3

. According to the International Energy Agency, US data centers alone consumed 224 terawatt-hours in 2025, representing more than 5 percent of the country's total electricity use—a sharp increase from 1.9 percent in 20181

1

. The AI environmental impact becomes clear when examining individual queries: processing a median-length text prompt with Google's Gemini consumes around 0.24 watt-hours1

1

. While this seems negligible—equivalent to watching TV for nine seconds—ChatGPT alone handles an estimated 2.5 billion prompts daily2

2

.

Water Crisis and Hidden Environmental Consequences

AI's environmental footprint extends well beyond carbon emissions. A groundbreaking United Nations report measured carbon, water consumption, and land usage together, revealing costs far larger than carbon figures alone suggested3

3

. By 2030, the electricity that data centers use would require nearly 2.5 trillion gallons of water for cooling—enough drinking water for the entire world for 1.7 years, according to Kaveh Madani, director of the United Nations University Institute for Water, Environment and Health2

2

. AI's energy and water consumption varies dramatically by task. Generating a complex video requires the equivalent of 42 hours of an efficient light bulb burning and uses a gallon of water, while a single AI image keeps a small LED bulb lit for about 17 minutes2

2

3

3

. The report also found that swapping coal for bioenergy can cut carbon footprints by roughly 70 percent, but this same swap can multiply water footprints more than 30-fold and land usage 100-fold3

3

.

AI and Data Centers Undermine Decarbonization Efforts

"AI is going in the opposite direction to decarbonization efforts," said Sasha Luccioni, co-founder and chief scientific officer of the Sustainable AI Group2

2

. Tech giants including Google, Meta, Amazon, OpenAI, Anthropic, Microsoft, and Oracle are investing tens to hundreds of billions of dollars to build AI-focused data centers1

1

. These new "hyperscale" facilities can use a gigawatt or more of electricity—roughly a tenth of Los Angeles's electrical capacity—compared to pre-AI data centers that consumed around 100 megawatts1

1

. Much of this demand is being met by fossil fuel plants, particularly gas, because data centers are often constructed in places without abundant renewable energy sources like hydropower, geothermal, solar, or wind1

1

. Without energy-saving strategies, US data centers could soon release the equivalent of 24 to 44 megatons of carbon dioxide annually—the latter equivalent to Norway's annual emissions1

1

.

Lack of Transparency Complicates AI Resource Usage Assessment

AI companies aren't transparent about the environmental consequences of their operations, forcing researchers to make estimates based on less common open source AI2

2

. "We have no way of knowing and getting a sense of the amount of energy," said University of Michigan computer science professor Mosharaf Chowdhury2

2

. Ana Pinheiro Privette, former top sustainability official for Amazon Web Services, emphasized that lack of transparency leaves consumers powerless: "We're really not choosing. We are being given whatever is being given to us"2

2

. Following the United Nations report, the UN called on AI companies to disclose their carbon, water, and land footprints and run data centers on renewable energy by 20303

3

. The report found that inference—the everyday work of answering prompts—consumes an estimated 80 to 90 percent of an AI system's energy, far exceeding the training phase that has dominated public concern3

3

.

Local Communities Bear Disproportionate Burden

The burdens of AI rarely land where the benefits do. In Ireland, data centers drew about 21 percent of all metered electricity in 2023, more than every urban home combined, prompting the grid operator to freeze new approvals around Dublin until 2028. Communities near data centers often suffer from air and noise pollution from gas plants and possible strain on local water resources1

1

. In Uruguay, a thirsty data center was planned just as a 2023 drought drained the capital's reserves and left tap water unsafe to drink3

3

. AI hardware could generate up to 2.5 million tons of e-waste annually by 2030, much of it shipped to poorer countries with few safeguards3

3

. Only about 32 nations host these data centers, with most computing power concentrated in just two countries3

3

.

What Users Can Do to Reduce AI's Environmental Footprint

Experts recommend using generative AI less frequently, especially for simple tasks. "The cleanest form of AI use is no use," said Madani2

2

. Luccioni advised against using AI for calculations, directions, store hours, recipes, or shopping lists—tasks that don't require AI but waste power and water2

2

. When making queries, users should be concise, as more information translates into more computing and resource depletion2

2

. Computer scientists and engineers are developing energy-saving algorithms and processor designs to address AI exacerbates climate change concerns1

1

. "AI's energy cost is not an accident: This is basically a product of how our systems are built," says Fengqi You, an expert in energy systems at Cornell University. "But with the right mix of solutions, we could really reshape the trajectory"1

1

.