Table of Contents

Hiding in Plain Sight: Is There More to AI Than We’ve Been Led to Believe?

Image Credit: Medium

Over 1.5% of all electricity and billions of gallons of potable water are spent on AI every year. When one thinks of “AI”, they don’t think of all the massive data centers pouring energy, water, and money just to generate a response to a simple question. Instead, we rarely visualize AI as something physical. We interact with AI on an app, a website, the internet – and most people are content to stop there. But this massive leap in technology may be costing us more than we thought. As AI systems rapidly expand, the data centers powering them are driving a surge in global energy demand. Because many data centers rely on fossil fuels, rapid expansion of these resources could heavily increase greenhouse gas emissions if the grids continue to rely on unsustainable sources of energy. However, AI may be extremely useful to fight climate change. By using detailed AI algorithms, we can improve renewable energy systems and production, reduce waste, and accelerate scientific research. However, the use of AI must be used responsibly. As data centers continue to expand, stretching our resources thin, it is more important than ever to weigh the costs and benefits of AI.

Hundreds of millions of people all over the world use AI every day – and yet, not many think of the infrastructure behind every AI application. Every AI Model is run inside of enormous data centers, each filled with high-performance computers running constantly in a concentrated space^[¹^,²^,³^,⁴^]. These facilities house and run thousands of specialized processors known as GPUs, or Graphics Processing Units^[⁵^]. GPUs can process many pieces of data simultaneously, and are therefore essential for AI^[⁶^]. Data centers use specific GPUs designed to handle the massive demands of AI systems. While traditional internet searches generally retrieve stored information and data, AI systems create new outputs in real time, performing billions or even trillions of calculations in just one prompt^[⁴^,⁷^]. Not only must AI systems compute massive computations every prompt, but they must do it quickly, accurately, all while generating entirely new content simultaneously.

These insane computations require equally immense amounts of electricity. According to the International Energy Agency (IEA), global data centers consumed over 415 terawatt hours (TWh) of electricity in 2024^[²^,⁸^]. To put this in context, this is about the same amount of energy that would be used to continuously power over 38 million US homes for an entire year. This number represents more than 1.5% of all electricity used worldwide, and is more electricity than most countries consume every year^[³^,⁹^]. Not only is this a massive number annually, but IEA additionally found that electricity demand has actually increased dramatically since 2017. Electricity demand from data centers has grown by approximately 12% every year^[²^,¹⁰^]. IEA projects that global data-center electricity demand could rise to over 945 TWh by 2030, largely due to AI^[²^,⁸^]. This projected number – 945 TWh is over double the energy demand recorded in 2024. The infrastructure required to power AI has become so energy intensive that some modern data centers consume as much electricity as cities.

This extreme strain on resources is a serious environmental concern – one of the first being greenhouse gas emissions. Although many tech companies are heavily invested in renewable energy, electrical grids in most regions still significantly rely on coal, oil, and natural gasses^[¹¹^,¹²^]. If AI-related electricity demand grows faster than clean energy infrastructure, the expansion of AI will increase fossil fuel consumption and carbon emissions. Recent investigations found that fossil fuel “peaker plants” scheduled to be shut down have remained operational longer than originally decided because of increasing demand from large data centers^[¹³^,¹⁴^]. Peaker plants are power plants only turned on during times of extreme electricity demand, or “peak demand”, such as heat waves or major surges in grid usage. Many are powered by natural gas or oil, meaning that as the peaker plants remain operational, more and more fossil fuels are dumped into Earth. This creates a contradiction often overlooked. Companies expand energy-intensive AI infrastructure while claiming to publicly support climate goals.

Another negative effect is the massive amount of water consumed by data centers. AI servers generate massive amounts of heat in data centers. To prevent overheating, many data centers rely on cooling systems which use water evaporation^[¹⁵^,¹⁶^]. A research paper published in the journal Nature found that some data centers obtain over half of their cooling water from potable drinking water supplies. Data centers use billions of gallons of water every year to cool the computers^[¹⁷^,¹⁸^,¹⁵^]. However, not only are massive amounts of water consumed directly through cooling systems, but data centers also indirectly consume even more water from the electricity generation process itself. Power plants themselves often require enormous amounts of water to cool turbines and generate electricity^[¹⁹^,²⁰^]. This means that even if a data center is not directly using water, the power plant supplying its electricity is likely using an equivalent amount. As AI infrastructure expands, water use by major technology companies has risen at unprecedented rates. Reports indicate that Microsoft’s water consumption increased by over 34% during the recent AI boom^[²¹^], and Google’s water usage has risen by approximately 22%^[²²^]. A single large data center can consume millions of liters of water every day, creating tensions in nearby communities, particularly in drought-prone regions already experiencing water shortages worsened by climate change. Worse, data centers are not evenly distributed. The IEA found that almost half of U.S. data center capacity is concentrated in five small regional clusters, meaning that resources like electricity and water are stretched even more than we might initially have expected^[³^,²³^].

However, while the infrastructure used to power AI is a large issue, AI can provide some important solutions to climate change. Many researchers believe that AI could become one of the most powerful tools available to address climate change – if deployed responsibly^[²⁴^,²^]. AI systems are already being used to improve renewable energy forecasting, optimize electricity grid efficiency, reduce energy waste, model extreme weather patterns, accelerate battery development, improve transportation, and detect methane leaks^[²⁵^,²⁶^]. One real-world example of AI in action to solve climate issues is renewable energy consistency. Renewable energy sources such as wind or solar power are naturally variable, as weather conditions are constantly changing. AI systems can help electrical grids predict fluctuations in energy production, in turn allowing them to distribute power more efficiently. This could also potentially reduce reliance on fossil-fuel backup systems. Scientists are also using AI to process massive climate datasets much faster than traditional computing methods, allowing researchers to improve climate models and identify environmental changes much faster^[²⁷^,⁹^].

The benefits and issues associated with AI creates a wavering tension between AI technological advancement and environmental preservation. If AI continues to grow without sustainable infrastructure, it may significantly increase global energy demand and environmental pressure. On the other hand, this same technology posing so many environmental concerns could become essential to manage renewable energy systems, improve efficiency, and accelerate climate solutions. Whether AI becomes a burden or a tool to climate change will depend heavily on how quickly governments, companies, and energy systems can transition to low-carbon electricity and sustainable infrastructure over the next decade.