The Massive Environmental Impact of AI Revealed

⚡

Key Takeaways

1In 2023, data centers consumed 4.4% of electricity in the United States, a figure that could triple by 2028.

2Training GPT-3 required 1,287 megawatt-hours, generating 552 tons of CO₂, equivalent to the annual energy use of 120 American households.

3AI systems could consume as much water as the global bottled water supply by 2025, exacerbating shortages in arid regions.

💡Why it matters — The rise of AI threatens to worsen energy and water crises, with significant environmental and social impacts.

Interactions with artificial intelligence, while appearing seamless and effortless, hide a concerning energy and environmental reality. Behind every request made to a chatbot or every image generated from a simple text instruction lies a massive and energy-intensive infrastructure, the ecological cost of which is constantly rising.

Exponential Energy Consumption

Data centers, which host the servers necessary for AI operations, consume a significant portion of electricity. In 2023, they accounted for 4.4% of total electricity consumption in the United States, a figure that could triple by 2028. Training models like GPT-3 is particularly energy-intensive: it required 1,287 megawatt-hours of electricity, generating 552 tons of CO₂, equivalent to the energy used by 120 American households over a year.

To provide some context, a simple text search on ChatGPT has been estimated to use nearly ten times more electricity than a standard search on Google. With billions of daily queries, the impact becomes colossal and hard to ignore for the general public.

The Water Footprint of AI

In addition to energy, AI also consumes significant amounts of water to cool its infrastructure. According to a study published in Nature Sustainability in November 2025, the water footprint of AI servers in the United States could reach between 731 and 1,125 million cubic meters per year from 2024 to 2030. This consumption could equal the total bottled water consumed worldwide by 2025, putting a strain on water resources in already vulnerable regions.

Moreover, the deployment of AI servers in the United States could generate an annual carbon footprint ranging from 24 to 44 million tons of CO₂ equivalent each year. This adds further pressure on the environment, already affected by greenhouse gas emissions.

A separate study from VU University Amsterdam highlighted AI's water consumption even more strikingly. Many data centers are located in areas where freshwater availability is already limited, including parts of Latin America, the southwestern United States, India, and Australia.

The Cost of Hardware and Electronic Waste

The manufacturing of hardware components, such as GPUs, is also a source of pollution. A single high-end GPU produces about 200 kg of CO₂, equivalent to driving a gasoline car for over 800 miles. With thousands of GPUs deployed in data centers, the environmental impact is significant, even before a single request is processed.

This issue is exacerbated by the rapid pace of obsolescence of AI-specific hardware. Unlike general-purpose servers, which can remain in service for five to seven years, AI-specific hardware often becomes obsolete in two to three years, driven by the relentless performance gains of each new generation of chips. This leads to a growing electronic waste crisis that is rarely addressed in the conversation about AI sustainability.

Geographic Disparities and Lack of Transparency

The impact of data centers is not evenly distributed. In Virginia, these facilities are expected to consume between 36 and 51% of the state's total electricity by 2030. In South Carolina, the figure reaches 65 to 70% of all new energy consumption. The effects are already being felt, with companies like American Electric Power in Ohio suspending all new data center connections due to insufficient electrical infrastructure.

Meanwhile, communities in The Dalles, Oregon, are opposing Google's expansion due to water consumption concerns. One of the frustrating obstacles is that the companies best positioned to shed light on this issue are not always the most transparent. Tech giants currently do not publish specific AI-related figures regarding energy and water usage.

Existing data suggests that the scale is significant. AI systems could have a carbon footprint equivalent to the total annual emissions of New York, ranging from 32.6 to 79.7 million tons of CO₂. These figures are estimates, as the underlying data remains opaque.

Towards Nuclear Energy?

In the face of these challenges, giants like Microsoft, Google, and Amazon have made notable bets on nuclear energy. Over the past year, they have signed agreements for more than 10 gigawatts of new nuclear capacity in the United States. This seems to be a clean solution, but there is a timing issue. Even as these companies promote plans for nuclear energy, they will continue to rely on fossil fuels in the meantime.

This means keeping coal plants operational and, in some cases, building new natural gas plants that could remain in operation for decades. Beyond the data, this story has a more personal dimension. The environmental cost of AI has become part of a much broader cultural conversation about climate anxiety, personal responsibility, and what it means to live with the knowledge that your daily actions have costs you cannot easily see.