—
Every time you think about AI or technology, something significant is happening, not just on personal devices but in massive data centers filled with servers and GPUs that require substantial energy. The current surge of AI is pushing the power grid to its brink, with systems like ChatGPT processing around a billion inquiries each day, demanding far more energy than a typical device ever would.
The energy required to support AI is skyrocketing. In fact, several coal plants in the U.S. are delaying their closures due to this rising need, with more postponements on the horizon. Some experts are sounding alarms that this AI arms race is surpassing our infrastructure’s ability to manage it. Others, however, believe this could spur innovations in clean energy over time. AI is changing not just apps and search engines, but also how we approach building, fueling, and regulating the digital landscape. The urgency to expand AI capabilities is outpacing our ability to support it with adequate energy.
The intricacies of AI’s energy demands are intriguing. Running large-scale AI requires immense computing power. Unlike everyday internet usage that merely accesses stored data, AI performs real-time processing. Whether developing large language models or responding to queries, AI systems rely heavily on specialized hardware like GPUs. These units consume significantly more power compared to older server technology. For example, a single Nvidia H100 GPU can draw up to 700 watts, and running thousands of them over weeks just to train a single model like GPT-4 introduces staggering power requirements. In contrast, traditional data centers might use about 8 kilowatts, while AI-optimized setups can demand more than 45 to 55 kilowatts.
Cooling the hardware adds another layer of complexity to energy consumption. If AI servers overheat, cooling can account for up to 55% of a data center’s total power usage. Although advanced cooling methods exist—like liquid immersion—they’re not yet widespread.
Nevertheless, AI researchers are developing more efficient systems, such as the “mixed expert” model, which activates only parts of a model as needed, thereby conserving energy without sacrificing performance.
In 2023, global data centers consumed around 500 TWh, enough to power every home in California, Texas, and Florida for a year. Experts predict AI could drive that number to triple by 2030. For context, the average home uses about 30 kilowatt-hours daily, and one terawatt hour can power about 33 million homes for a day.
The rapidly growing demand for AI is outpacing the capability of existing power grids. It’s estimated that in the U.S., data center power usage could exceed 600 TWh by 2030, which would mean needing the equivalent of 14 new power plants to meet that demand. Massive AI data centers can require 100 to 500 megawatts each, with some exceeding 1 gigawatt. That’s comparable to the output of nuclear plants; a single 1 GW data center can consume more electricity than the entire city of San Francisco. When you multiply that by multiple campuses, the picture of escalating demand becomes clear.
In response, utilities are reluctantly holding off on retiring coal plants while expanding natural gas projects. Some states, like Utah and Georgia, are even approving new fossil fuel investments directly related to the demands created by growing data centers. By 2035, these centers could account for about 8.6% of U.S. electricity consumption, up from 3.5% today. The irony is that, despite commitments to sustainability, tech companies might be inadvertently prolonging reliance on fossil fuels. For everyday consumers, this shift could lead to higher electricity costs, less local energy availability, and hinder clean energy goals.
Major tech companies like Microsoft, Google, Amazon, and Meta claim to be moving toward net-zero emissions, meaning they aim to balance greenhouse gas emissions with removals or offsets. They are buying large quantities of renewable energy to mitigate their consumption, and firms like Microsoft are even partnering with startups to secure clean energy.
However, critics argue these actions don’t necessarily reflect the real-world energy picture. Since the grid is interconnected, fossil fuels often fill the gaps created when renewable sources are insufficient, regardless of the clean energy purchases made by these giants. Some researchers have pointed out that the current corporate model benefits accounting more than it does climate progress.
As for the initiatives toward new energy sources—like data centers, geothermal systems, and small nuclear reactors—they come with their own technical and regulatory challenges. For instance, while fusion energy holds promise, it still hasn’t achieved commercial viability. Public concerns about safety and environmental impact complicate the deployment of new nuclear technologies, and the construction timelines for facilities can stretch to several years.
At the heart of the matter is whether AI will ultimately be beneficial for the environment or detrimental. Proponents suggest that AI could optimize energy usage and hasten advancements in clean technologies. However, skeptics warn that current trends are unsustainable and could lead to a rise in greenhouse gas emissions. Recent projections estimate that AI might contribute 1.7 gigatons of carbon dioxide to global emissions between 2025 and 2030.
Moreover, as AI infrastructure grows, there are emerging issues related to water usage, rare minerals, and land disputes. Cooling vast data centers demands millions of gallons of water annually, while the mining of essential materials like lithium and cobalt compounds existing supply chain stresses. Certain communities are resistant to zoning changes for tech development due to various concerns.
The rapid turnover of AI hardware also poses environmental challenges. Outdated GPUs and components create significant electronic waste, much of which ends up in landfills because of insufficient recycling programs.
Ultimately, the question remains—can AI become cleaner over time, or can we build the necessary infrastructure without reverting to fossil fuels? Meeting this challenge demands cooperation among tech firms, utility providers, and policymakers, as experts caution that AI’s impact on climate change will largely depend on how we guide the evolution of computing technology.
—
