{"id":12934,"date":"2025-12-27T15:02:31","date_gmt":"2025-12-27T15:02:31","guid":{"rendered":"https:\/\/www.yiaho.com\/water-pollution-consumption-what-is-the-true-environmental-impact-of-ai\/"},"modified":"2025-12-27T15:02:31","modified_gmt":"2025-12-27T15:02:31","slug":"water-pollution-consumption-what-is-the-true-environmental-impact-of-ai","status":"publish","type":"post","link":"https:\/\/www.yiaho.com\/en\/water-pollution-consumption-what-is-the-true-environmental-impact-of-ai\/","title":{"rendered":"Water, Pollution, Consumption: What is the True Environmental Impact of AI?"},"content":{"rendered":"

Artificial intelligence sparks many discussions, not only about its implications for employment but also about its effects on the environment.<\/p>\n

Beyond technological advancements, AI relies on infrastructures and components that demand significant quantities of energy, water, and rare materials. This article, written by the Yiaho team, factually examines these aspects, based on data and observed trends in the sector. <\/p>\n

Data Centers: Energy-hungry Giants<\/h2>\n

Technology companies are investing heavily in data centers to support AI development. For example, ambitious projects aim to create large-scale facilities, comparable in size to significant portions of major cities and in power to several nuclear production units. <\/p>\n

These structures, often located in rural or suburban areas, mark an evolution in investment priorities: in the United States, funds allocated to these infrastructures could soon surpass those for traditional office spaces, reaching billions of dollars. This AI race seems to be redefining international economic dynamics. In terms of electricity consumption, these data centers are expected to see a significant increase in demand. <\/p>\n

According to estimates, the total installed power could lead to a tripling of electricity consumption by the end of the decade.<\/p>\n

Currently, these facilities account for approximately 1.5% of global electricity consumption. In the United States, this share could rise from 4.4% to a range between 7% and 12% by 2028. In Europe, projections indicate an increase from 2.5% today to 7.5% by 2035. These figures highlight increasing pressure on electricity grids, without implying an immediate shortage, but calling for increased planning. <\/p>\n

AI’s Thirst for Water<\/h2>\n

At Yiaho, we’ve already wondered how much water AI consumes<\/a>. And the data is updated over time and with research. <\/p>\n

In addition to electricity, AI consumes significant amounts of water, primarily for server cooling and component production. In 2023, the volume of water used by data centers exceeded 5 trillion liters, an amount equivalent to the total potable water extracted annually in a country like France. <\/p>\n

This demand intensifies with the proliferation of data centers, raising questions about water resource management in the regions where they are located.<\/p>\n

Also read on this topic: How much water does ChatGPT consume for one query? Sam Altman’s answer<\/a><\/p>\n

Rising Greenhouse Gas Emissions<\/h2>\n

The carbon footprint of data centers is another key point. Projections vary, but they indicate that emissions related to these infrastructures could double or triple in the coming years, depending on growth scenarios. <\/p>\n

This is explained by the reliance on energy sources that are not always renewable, although efforts are underway to integrate more green energy.<\/p>\n

Key Components: GPUs and Rare Materials<\/h2>\n

At the heart of AI are Graphics Processing Units (GPUs<\/a>), essential for complex calculations. Their deployment is expected to increase significantly, with their number multiplying fivefold over a ten-year period. A detailed analysis of a popular GPU, such as those produced by Nvidia, reveals the presence of more than twenty different metals, including rare earth elements. <\/p>\n

These materials, extracted from mines often located in specific regions of the globe, are crucial for performance, but their supply poses logistical and environmental challenges.<\/p>\n

The market for GPUs and AI-dedicated chips is expanding rapidly, with forecasts reaching over $280 billion by 2029.<\/strong><\/p>\n

But the manufacturing of these components is resource-intensive: it requires energy, water, metals, and chemical substances. To improve computing power, elements must be miniaturized and purified to the extreme, which increases the use of potentially harmful products. <\/p>\n

Chip Production: A Resource-Intensive Process<\/h2>\n

Semiconductor factories, like those of Taiwanese leader TSMC, illustrate this intensity.<\/p>\n

The creation of a single 12-inch wafer, on which chips are etched, can require more than 7,000 liters of water. With an annual production equivalent to 16 million such wafers, the total volumes are considerable. <\/p>\n

Although high recycling rates, around 88%, are applied, this does not entirely eliminate the impacts on local resources. In summary, AI, while offering promising innovations, implies increased consumption of natural resources.<\/p>\n

Current data and future projections call for a balanced reflection on how to balance technological progress and environmental sustainability, without dramatizing or minimizing established facts. Initiatives to optimize efficiency and integrate greener practices could mitigate these long-term effects. <\/p>\n

In short, does AI pollute or not?<\/h2>\n

Artificial intelligence does indeed pollute, in the sense that its development and operation lead to very significant consumption of electricity, water, and raw materials, as well as an increase in greenhouse gas emissions.<\/p>\n

This environmental footprint is real, measurable, and growing rapidly with the expansion of infrastructures and the most powerful models.<\/p>\n

But it still remains relatively modest on a global scale (a few percent of global electricity consumption today) and it is not inevitably destructive: progress in energy efficiency, cooling, water recycling, the use of decarbonized energies, and the design of more efficient chips can significantly limit this impact.<\/p>\n

AI pollutes today, it pollutes more than yesterday, and it risks polluting even more tomorrow if nothing changes, but the final level of this pollution will largely depend on the technical and energy choices made in the coming years.<\/p>\n

Source:<\/p>\n