By Mari Hansen 
At first glance, the current AI boom does not appear to be code. It resembles buildings. You begin to notice them when you drive through Northern Virginia on a muggy afternoon: windowless warehouses humming softly next to suburban areas and highways. Racks of servers that power the internet are housed in these buildings. Algorithms create images, chatbots respond to inquiries, and recommendation systems choose what users should watch next. It’s hard not to wonder how much electricity is passing through those walls when you look at the size of these facilities.
One of the most energy-intensive technologies ever developed is artificial intelligence. Large models can use a lot of power just to train. According to a frequently cited example, the GPT-3 model’s training required approximately 1,287 megawatt-hours of electricity, which, when powered by a traditional grid, produced hundreds of tons of carbon emissions. That was just the start. After training, millions of users begin to ask questions, and the system must instantly respond while continuously performing calculations. This is when the real demand starts.
| Category | Information |
|---|---|
| Industry Focus | Artificial Intelligence Infrastructure |
| Key Companies | Microsoft, Google, Amazon |
| Primary Infrastructure | Hyperscale Data Centers |
| Estimated Global Data Center Energy Use | ~1–3% of global electricity |
| AI Training Energy Example | GPT-3 training required ~1,287 MWh electricity |
| Major Challenge | Balancing AI growth with sustainability targets |
| Key Technologies Involved | GPUs, CPUs, TPUs, NPUs |
| Environmental Concerns | Carbon emissions, water consumption, grid strain |
| Global Investment in Data Centers | Estimated $550+ billion under construction |
| Reference Source | https://www.iea.org |
The data center, which serves as the process’s physical foundation, is rapidly expanding throughout the globe. According to industry analysts, the value of properties under construction worldwide is estimated to be over $550 billion. There is a race to increase computing capacity as construction cranes now loom over locations in Virginia, Malaysia, Ireland, and the American Southwest. Investors appear to be certain that the need for AI infrastructure will only increase. However, the expansion leads to an awkward paradox.
Over the past ten years, the technology sector has made commitments to renewable energy, net-zero targets, and climate responsibility. Simultaneously, AI’s computational appetite continues to grow. Data centers and digital networks already account for about 1% of the world’s energy-related emissions, according to the International Energy Agency. Although that figure might not seem like much at first, demand is rising rapidly. According to some estimates, the amount of electricity used in data centers worldwide may double by 2026. Even the biggest tech firms seem to be still trying to figure out how to balance those realities.
Amazon, Google, and Microsoft have all publicly acknowledged the conflict. Their sustainability reports, which celebrate purchases of renewable energy while acknowledging that emissions from data centers are still rising, read almost like careful balancing acts. For example, Google recently revealed a discernible rise in greenhouse gas emissions, primarily due to the expansion of supply chains and infrastructure in areas where access to clean energy is more difficult.
The reasons become evident when you enter a hyperscale data center. Massive cooling systems circulate chilled air throughout the building as rows of servers stretch into the distance, blinking lights reflecting off polished floors. These devices produce a lot of heat, and it takes constant energy to keep them cool. Evaporative cooling systems use a lot of water every day in many buildings.
That has started to provoke quiet local discussions in water-stressed areas like Arizona or parts of Ireland. Sometimes locals ask a straightforward question: should fresh water be used to cool search engines or to support households and agriculture? Seldom is the solution simple.
The majority of AI’s environmental impact comes from operations rather than construction. According to studies, the electricity used to run servers and cooling systems is responsible for about 97% of data center emissions. Additionally, there are significant regional variations in electricity. Compared to facilities that rely on coal-heavy grids, those powered by hydroelectric or wind energy may run with significantly lower emissions. Businesses are being forced by this reality to reconsider the locations of their data centers.
Suddenly, choosing a site has become a climate strategy. Emissions can be significantly decreased by placing facilities close to renewable energy sources, such as wind corridors, solar farms, and hydroelectric dams. According to some estimates, a suitable location can reduce operational carbon emissions by up to 60%. However, relocating infrastructure is more difficult than it seems, particularly in light of the rapidly increasing demand for computing power worldwide. The equation is further complicated by hardware choices.
Graphics processing units (GPUs), which are far more effective than conventional CPUs for machine-learning workloads, are a major component of AI models. However, when running at scale, GPUs also use a lot of electricity. In an effort to extract more computation per watt, engineers are experimenting with alternatives such as neural processing units, tensor processing units, and specialized accelerators.
However, pressure from competitors frequently clashes with efficiency gains. Seldom do businesses striving to create the most potent models want to halt experiments just to conserve energy. Within the industry, there is a belief that the person who grows the fastest will control the next stage of computing.
For the time being, investors appear at ease with that risk. In 2025, data centers attracted tens of billions of dollars in acquisitions, making them the most sought-after real estate sector for investors. Infrastructure funds and private equity firms are still investing heavily in the industry, hoping that the demand for AI will surpass environmental concerns.
It’s difficult to ignore the paradox as this expansion takes place. Methane leak detection, climate change modeling, and renewable energy grid optimization could all be aided by artificial intelligence. At the same time, massive amounts of infrastructure and power are needed by the systems carrying out those tasks. The environmental impact of AI in the future might rely more on electrical grids than on algorithms.
In order to secure clean power, such as nuclear and hydroelectric sources that can run continuously, some tech companies are entering into long-term contracts. Others are experimenting with facilities constructed right next to renewable energy projects, carbon-aware computing schedules, and liquid cooling systems. It’s unclear if those initiatives will keep up with the development of AI.
The scope of the problem becomes apparent when one stands outside a contemporary data center and feels the warm air drifting from ventilation units and hears the slight mechanical hum. The devices within might be a symbol of the upcoming technological era. However, they also serve as a reminder that electricity is still necessary for intelligence, including artificial intelligence. Additionally, there is always a source of electricity.