Despite being the epitome of modern technology, artificial intelligence (AI) runs mainly on antiquated fossil fuels. As tech giants race to build data centers, the energy required to sustain this AI boom locks the U.S. into a dangerous dependence on natural gas — one with dire consequences for the climate.
Most AIs are large language models (LLMs), software tools that are able to analyse linguistic patterns in billions of text sources and from context, predict the next word. The most capable LLMs are those that generate text, such as ChatGPT, Claude and Gemini.
AI consumes energy, most of it as electricity, in two major ways: training and inference. During training, billions of parameters, which determine the weight of the various probabilities in word prediction, are adjusted in an LLM using vast amounts of data. The high-performance computing chips used for training require immense power — for OpenAI’s GPT-4, the initial training used the equivalent of San Francisco’s total energy consumption over three days — to keep the hardware, electricity, cooling, and maintenance running.
Yet, about 80 to 90 percent of the total energy consumption comes from inference, which is when users interact with the AI. AI models rely on graphics or tensor processing units for fast parallel processing, meaning many computations occur simultaneously, to perform the large matrix operations essential to manipulating and transforming data. These chips consume significantly more power than the central processing units used in almost all modern electronic devices and non-AI data centers. Indeed, querying an LLM costs ten times more energy than a standard Google search and emits 340 times more carbon dioxide.
Nonetheless, the International Energy Agency estimated that by 2026, AI electricity will still account for only around 3% of global electricity demand. Additionally, AI-related computer chips have grown over 100 times more energy efficient in under two decades. Indeed, it is not the quantity of energy that AI consumes that is the primary issue, but where it comes from.
A major challenge with sourcing energy from renewables is that, as CEO of Mawson Infrastructure Group, Rahul Mewawalla, stated, AI data centers require “constant power, 24/7, 365 days a year.” This restriction precludes most renewable energy sources like solar and wind power due to fluctuating weather conditions. Even nuclear, with its advantages as a consistent, environmentally-friendly source, is being rejected because of a perceived clash between the decade it takes to build a nuclear power plant and the immediacy of the energy need.
Natural gas, however, is readily available and in 2024, it reached historic low prices in the U.S. AI companies also benefit from its predictability and ease of production and scaling. Thus, AI companies are resorting to natural gas to meet their exploding energy demands, careless of environmental consequences. According to a study by Harvard’s T.H. Chan School of Public Health, the energy used by AI data centers releases 48% more greenhouse gases than the average U.S. energy output. Yet, even this does not capture the full extent of the greenhouse gas outputs: Beyond burning fossil fuels, drilling natural gas and installing pipelines releases methane, which traps 28 times more heat than carbon dioxide. The relative “dirtiness” of data center energy is expected to rise as Big Tech pours money into new data centers in states like Virginia, where natural gas accounts for 55% of electricity generated.
Similarly, in Louisiana, Meta is planning the construction of its largest data center yet, which was originally designed to consume two gigawatts (GW) of power — enough to fuel 1.6 million U.S. homes. To provide this energy, Meta is collaborating with Entergy, the largest utility company in Louisiana, Arkansas and Mississippi, to build three new natural gas plants totaling 2.3 GW in capacity and costing $3.2 billion. With a lifetime of around 30 years, Entergy’s plants imperil the state’s goal of carbon neutrality by 2050, particularly since 72% of Louisiana’s energy is still generated by natural gas.
In July 2025, Meta announced plans to expand the Louisiana data center from two to five gigawatts with an area the size of Manhattan. Resistance to the plants is not quiet. Various nonprofit organizations filed a lawsuit to oppose Entergy’s plans due to the lack of information regarding their economic and environmental consequences. A judge dismissed the challenge on the basis that there lacks justification for Meta to divulge details.
In other Southeastern states like Virginia, Georgia, South Carolina, and North Carolina, utilities are adding more than 20 GW of power from natural gas, with data centers driving 65 to 85 percent of forecasted grid load growth. Entergy is also building a new 754 megawatt natural gas power plant in Mississippi, the first of its kind in half a century, to serve two Amazon data centers.
In response, the Institute for Energy Economics and Financial Analysis (IEEFA) notes that there is a “serious risk of overbuilding” as Big Tech’s high AI energy growth projections could be an overestimate. The problem becomes particularly pressing when we consider the elevated environmental costs of carbon emissions from natural gas which, according to the IEEFA, are “incompatible with … mitigating the worst effects of climate change.” As reported by the Rhodium Group, if natural gas remains the default source for data centers, the U.S. will emit an additional 278 million metric tons of carbon annually by 2035, which is roughly equivalent to Florida’s yearly emissions.
Due to these increased emissions, Big Tech companies are deserting their carbon neutrality commitments en masse. In an attempt to make this abandonment seem less distressing, some companies, such as Meta, Google, and Amazon, have instead pledged to double global nuclear capacity by 2050. However, this timeframe is too slow to meet their near-term demand; “clean” power promises remain vague. In addition, in Louisiana, Meta said it would help finance 1.5 GW of renewable power, but Meta nor Entergy has yet to file for state utilities permission or announced specific projects.
Noticing this discrepancy, Sen. Sheldon Whitehouse, D-R.I., sent Meta CEO Mark Zuckerberg a letter pressing the company to deliver documents detailing the expected greenhouse gas emissions of the Louisiana data center; explanations for the use of natural gas and the lack of carbon capture facilities; and analyses on the compatibility of the gas plants with Meta’s goal of net zero emissions by 2030, stating that the tech giant’s current plan “flies in the face of Meta’s climate commitments.” As of August 2025, Meta has not confirmed the release of such information.
Though corporate climate pledges surged during 2022 and 2023 before the commercial release of AI, with half of the world’s 2000 largest companies committing to net zero, only 4% of those companies are projected to reach their targets. Microsoft, Google, and Amazon have stated that AI is making it difficult to meet sustainability goals.
In fact, in 2024, Google’s parent company discontinued its carbon neutrality program, pointing out that, “reducing emissions may be challenging due to … the greater intensity of AI compute.” AI’s expansion led to nearly a 50% rise over the past five years. This spike lies in stark contrast to Google’s 18% reduction in emissions from 2019 to 2020.
But data center energy does not need to grow in carbon intensity. A Duke study demonstrates that if data centers cut their energy usage by 50% for a few hours per year during spikes in grid demand they could be integrated into the preexisting grid without needing additional power plants built. This “load flexibility” would make AI energy more affordable and climate friendly. In fact, Google recently announced such agreements in Indiana, Michigan and Tennessee.
Cumulatively, Big Tech is worth $17.9 trillion: Tech giants have the resources and influence to invest in clean energy. Now, with the Trump administration rolling back climate protections, corporate responsibility is more important than ever. Sustainable energy alternatives exist. With storage, they too can provide firm electricity sources.
Indeed, the Emirati renewable energy company Masdar and Emirates Water and Electricity Company (EWEC) kickstarted a 5.2 GW solar power project with battery storage to produce 1 GW of round-the-clock power. EWEC CEO Othman Al Ali praised the facility for “supporting AI, ensuring their power needs are met sustainably and reliably.” Costing $6 billion and expected to start running in 2027, this project is better value for money than the Louisiana Entergy plants, with $1.15 billion and $1.39 billion per GW of power production respectively, and will take two fewer years to construct. By utilizing similar U.S. desert landscapes, a similar result could be achieved.
Most importantly, our lawmakers must demand transparency from Big Tech. It is the general public who bears data centers’ climate costs. Reporting annual carbon emissions is not enough. We need honest commitments, data on ongoing and future projects, and acknowledgement of the local and global consequences of expanding the AI sector relentlessly. And if the answers received do not meet climate standards, our representatives should stand up to Big Tech and prioritize the interests of citizens.
Associate Science and Technology Editor
