Is AI Forcing a Return to Fossil Fuels? 50% Energy Surge Explained

Is AI Forcing a Return to Fossil Fuels? 50% Energy Surge Explained

The AI revolution, often touted as a cornerstone of a green future, harbors a dirty secret: its skyrocketing energy demands are quietly forcing a significant resurgence in fossil fuel reliance, particularly natural gas. In my research, I found that in 2025 alone, electricity demand from AI-focused data centers surged by an astonishing 50%, dwarfing the global electricity demand growth of just 3%. This unprecedented acceleration is creating a critical bottleneck in our energy infrastructure, compelling utilities and tech giants alike to prioritize immediate, dispatchable power over long-term clean energy goals. What I discovered is that this isn't just a minor blip; it's a profound shift that could redefine our energy landscape for decades.

The Grid's Breaking Point: A Systemic Crisis

The fundamental challenge isn't merely the sheer volume of electricity AI requires, but the speed at which this demand is materializing and its need for 24/7 firm, reliable power. AI-driven computing is expanding data center growth faster than existing grids can deliver new capacity, making power, not land or chips, the primary limiting factor for new facilities. In regions like Northern Virginia, a global data center hub, new grid connections face multi-year delays, while places like Ireland and Singapore have even imposed restrictions on new data center approvals due to grid constraints. My findings show that this grid strain is more than just an inconvenience; it's a systemic crisis.

The U.S. alone could see its data center demand reach 74 GW by 2028, facing a projected shortfall of approximately 49 GW in available power access. Utilities, accustomed to predictable demand growth, are now grappling with unprecedented spikes, leading to clogged interconnection queues and heightened reliability risks. I've seen projections that data centers now consume 4.4% of U.S. electricity, more than some entire states, and could reach double or even triple that amount by 2028. The PJM electricity market, which covers a large part of the eastern U.S., projected 32 GW of growth in its 2025 Long-Term Load forecast, but entering 2026, aggregated requests more than doubled that to 70 GW over the next 15 years—equivalent to adding the power used by Texas or California. Analysts like Gartner are even predicting that power shortages will restrict 40% of AI data centers by 2027, a direct consequence of demand outstripping local grid capacity.

The Silent Return to Gas: Fueling the AI Boom

Facing these immediate power constraints and the imperative for continuous operation, the industry is making a stark pivot. Recent data from 2025-2026 reveals a startling trend: planned non-renewable additions to the power grid, predominantly natural gas, surged by 71% during this period. In stark contrast, renewable growth flattened to a mere 2% over the same timeframe. Natural gas has emerged as the most reliable baseload energy source, capable of providing the constant power AI infrastructure desperately needs. I believe this shift is driven by the lower grid-connection costs for natural gas (averaging $24/kilowatt (kW) from 2017-2022) compared to solar ($253/kW) and offshore wind ($335/kW), as well as higher project completion rates.

This isn't just about utility-scale projects. Many data centers are bypassing grid delays by building their own natural gas power plants on-site, a trend I've observed gaining significant momentum. For example, Google is involved in a planned gas-fired power plant in Armstrong County, Texas, which could generate up to 4.5 million tons of carbon dioxide annually—more than the city of San Francisco. This facility, a partnership with Crusoe Energy for Google's "Goodnight" data center campus, is designed to operate off the grid. I've also found that Meta is funding the construction of seven new natural gas-fired plants to fuel its massive Hyperion data center in rural Louisiana, and Amazon has several multi-gigawatt data centers powered by gas. Microsoft, too, announced a new gas project for a data center in West Virginia and signed a deal with Chevron in 2026 to build a 2.5-gigawatt gas power plant in west Texas. Furthermore, NRG Energy, in partnership with GE Vernova and Kiewit Corp., plans to expedite natural gas power plant projects for data centers, with four plants totaling 5 GW planned for Texas and mid-Atlantic states. Oracle and Project Stargate are collaborating on a multibillion-dollar agreement to develop 4.5 GW of AI data center capacity, powered by a dedicated onsite natural gas facility in Abilene, Texas.

The Hidden Environmental Cost: Water and Carbon

Beyond the immediate energy demands, I've uncovered two critical, often overlooked environmental impacts: water consumption and the true carbon footprint, which together form what I call the "Green AI Paradox."

First, the Water Footprint of AI: Data centers require vast amounts of water for cooling their processors, which generate enormous heat. My research shows that data centers powering AI systems consumed approximately 17 billion gallons of water in the U.S. in 2023, with projections showing usage surging to 68 billion gallons by 2028—a staggering 300% increase in just five years. A single Meta data center in Newton County, Georgia, consumes 500,000 gallons of water per day, roughly 10% of the entire county's supply. In Iowa, one facility consumed a billion gallons in 2024, enough to cover the state's residential water needs for five days. Texas faces an even steeper trajectory, with a study projecting the state's data centers will consume 49 billion gallons of water in 2025, potentially rising to 399 billion gallons by 2030. This massive demand is a pressing problem, especially in regions already facing drought or depleting water supplies.

Second, The Green AI Paradox: While AI is often presented as a tool for sustainability, its own energy intensity creates a significant paradox. I found that training a single large AI model can emit as much carbon as five gasoline-powered cars over their entire lifetimes. A single query to a generative AI model like ChatGPT can use nearly 10 times more electricity than a conventional Google search. If Google were to power its 9 billion daily searches with its most advanced AI, it could require up to 30 times the energy it currently uses for search operations. My analysis of major tech firms' 2025 sustainability reports revealed a profound disconnect: while electricity consumption and actual location-based emissions are skyrocketing, market-based figures can present a fiction of progress. For instance, between 2020 and 2024, Microsoft's electricity consumption grew by 176% and its location-based emissions by 130%, yet its reported market-based emissions actually fell by 56%. This suggests a form of "greenwashing" that masks the true environmental cost of the AI boom.

Geopolitical Stakes: The AI Energy Race

I believe the energy demands of AI also carry significant geopolitical implications, reshaping global competition for resources. Nations increasingly view strategic control over compute, data, and digital infrastructure as central to national power. The race for AI dominance, particularly between the United States and China, is now inextricably linked to the ability to secure reliable and abundant energy. While the United States currently has an advantage in cutting-edge AI semiconductors, China holds a significant lead in overall energy generation, producing more than twice as much electricity as the U.S. China's electricity demand for data centers is also expected to more than double over the next five years, reaching around 277 TWh by 2030, a growth that is unlikely to constrain China given its rapid pace of energy expansion. This "electron gap" could potentially reshape the balance of compute for AI between the two countries.

What This Means For Investors, Entrepreneurs, and Professionals

For Investors: I see clear opportunities, but also significant risks. The immediate need for dispatchable, 24/7 power means continued investment in natural gas infrastructure, including new power plants and pipeline capacity, will likely yield returns. Companies like NRG Energy and Williams Companies, which are strategically positioned in regions with high data center growth and existing infrastructure, are examples of this trend. However, I also believe that the long-term imperative for sustainability will drive investment into advanced renewable energy solutions that can provide baseload power. Corporate power purchase agreements for zero-carbon electricity reached 29.5 GW in 2025, with a growing share of nuclear, hydropower, and geothermal contracts as tech giants double down on clean, baseload power for AI data centers. I think investors should look for companies innovating in long-duration energy storage, small modular reactors (SMRs), and advanced geothermal, as these technologies can offer the firm, clean power AI demands.

For Entrepreneurs: This energy crisis presents a fertile ground for innovation. I see opportunities in developing more energy-efficient AI algorithms and hardware, as well as novel cooling technologies that drastically reduce water consumption in data centers. Solutions for optimizing existing grid infrastructure, such as AI-powered grid management systems, will also be in high demand. Furthermore, entrepreneurs focusing on distributed energy resources, microgrids, and "behind-the-meter" power generation solutions that can circumvent grid delays will find a ready market among data center operators.

For Professionals: I believe the landscape is changing rapidly, requiring new skill sets. Energy professionals will need to understand the unique demands of AI workloads and how to integrate them into existing and future grids. IT professionals will increasingly need to grasp the energy implications of their computing choices. For sustainability professionals, the challenge is immense: moving beyond market-based accounting to location-based emissions will be crucial for accurately reporting and reducing the environmental impact of AI. I anticipate increased scrutiny on corporate sustainability reports and a growing demand for experts who can navigate the complex intersection of AI, energy, and environmental responsibility.

Bottom Line

The AI revolution, while promising immense progress, has ignited an unprecedented energy demand that is undeniably forcing a short-term resurgence in fossil fuel reliance, particularly natural gas. As I've explored, this isn't just about electricity; it's also about a massive, often hidden, water footprint and the geopolitical implications of an intense global energy race. Addressing this challenge requires immediate, concerted efforts in grid modernization, sustainable energy development, and a transparent reckoning with AI's true environmental cost, lest the very technology we hope will save us from climate change becomes a significant contributor to it.