By David Bluestein
Photo by Kari Shea on Unsplash
In the past few years, generative artificial intelligence (AI) has gone from an obscure idea to a very real part of everyday life for many people. AI has been integrated into Google searches, smartphone assistants, and countless applications in professional and academic settings. Without regulation, many people worry that AI will replace humans in certain jobs, but there are also significant environmental concerns related to the explosion in AI’s popularity.
Modern AI uses a massive amount of energy and water. While companies such as Google and Microsoft had been reducing their environmental footprints in many ways, even aiming to eventually reach “net zero” emissions, their adoption of AI has completely derailed those plans and rapidly increased their environmental footprints. So, let’s look at what it is about AI that makes it so detrimental to the environment.
The environmental impacts of AI come primarily from the physical locations of the information and processing it uses, which are known as data centers. Data centers have been around for a long time, but many more AI-dedicated data centers have been – or are being – built since the release of ChatGPT in 2022. Unfortunately, data centers use a lot of electricity to keep powerful processors and servers running, and they consume immense amounts of water to cool the hardware. The rapid growth of data centers has led experts to predict that by 2030, data centers may account for up to about 10% of US electricity use.
The high level of electricity use at AI data centers comes from two main processes. The first is training: AI requires the input and complex processing of large amounts of data simply to provide the system with the tools it needs to generate responses. Processing that data uses a large amount of electricity. Second is the use of electricity when the AI receives a query. At that point, another large amount of electricity is used in processing the query and generating a response. In both cases, the process generally utilizes a collection of energy-intensive advanced processors called GPUs (graphics processing units).
The popularity and extensive use of AI have created so much electricity demand that entire power plants are being brought online, including coal plants that were otherwise supposed to be shut down. As a result, the electricity consumption of data centers has led to a rapid increase in carbon dioxide (CO2) emissions. To account for this rapid rise in electricity use and the associated CO2 emissions, Microsoft and Amazon have both started working to open nuclear power-plants. Regardless, most data centers still rely on fossil fuels for their electricity.
The other large factor in AI’s environmental footprint is water. At data centers, the large amount of electricity flowing through processors and servers also generates a lot of heat. For the equipment to continue functioning, the data centers need to dissipate that heat and cool the equipment quickly and constantly. For this, they use water, and a lot of it. In 2023, Google’s data centers alone used about the same amount of water as 207,000 average Americans. Since climate change is expected to make water scarcities more frequent in certain regions, AI’s water use is a significant concern in areas prone to droughts.
While AI has immense promise in a wide variety of uses, we need to account for the toll its use will have on our environment. Efforts are already underway to improve the energy efficiency of the processors and servers used by data centers, which would reduce both energy and water usage. Similarly, using renewable energy or nuclear power to provide electricity for data centers can reduce CO2 emissions. As individuals, we can also simply be mindful and avoid using AI. Nonetheless, AI is here to stay, so the first step is to understand it and the consequences it brings.
David graduated from Duke’s Master of Environmental Management program with a focus on Business and Environment in 2024.
