AI Expansion and the Energy Crisis: A Balanced Perspective on Powering Data Centers

The Looming Energy Crisis and the Expansion of AI

Recently, President Trump declared an energy emergency, highlighting concerns regarding the growing electricity demands of artificial intelligence (AI) data centers. This declaration was largely motivated by the projected need for substantial power to fuel the expansion of up to 5,000 new data centers across the United States. To address this potential crisis, the administration is turning to natural gas and gas-fired power plants.

Chris Wright, the Secretary of Energy, emphasized the perceived limitations of renewable energy sources like wind and solar, stating, “Beyond the obvious scale and cost problems, there is simply no physical way that wind, solar and batteries could replace the myriad uses of natural gas.” He questioned the reliability of renewables, particularly in comparison to the established infrastructure of fossil fuels.

However, this perspective doesn’t tell the whole story. Data from other countries demonstrate the commercial viability of solar and wind energy, especially when paired with grid-scale battery systems—a combination often referred to as green SoWiBess (Solar, Wind, and Battery Energy Storage System). The Secretary also suggested small nuclear reactors and geothermal resources, but these options currently contribute a minimal amount to overall electricity generation in the United States.

What is the most realistic approach to meeting the escalating energy demands of AI data centers? This article aims to present a more balanced perspective by drawing on data and facts from multiple independent sources.

Fossil Fuels vs. Renewables: A Balanced View

Examining the JP Morgan view based on a Roger Pielke analysis gives these energy realities:

1. The Dominance of Fossil Fuels: Globally, coal, oil, and gas consumption have grown linearly since 1965. Each fuel source provides about one-third of the world’s total energy.
2. Fossil Fuel Consumption by Country: Fossil fuels make up a significant portion of total energy consumption: Asia (excluding Japan and China) at 92%, Japan and China at 87.5% each, the U.S. at 87%, and Europe at 79%. However, fossil fuel consumption is declining in most of these regions, most dramatically in Europe.
3. The Slow Rise of Renewables: Carbon-free renewables accounted for 14% of global consumption in 2012, increasing to 18% in 2024. The annual growth rate of renewable sources is relatively slow, approximately 0.3-0.6%. Roughly $9 trillion has been invested globally in carbon-free sources such as wind, solar, battery storage, and EVs.
4. Solar’s Growing Share: Solar power supplies approximately 7% of global energy generation, with projections to reach 14% by 2027. Solar’s progress accelerated in 2024, primarily driven by China.
5. Energy Intensity of Economic Prosperity: According to the JP Morgan report, steel, cement, fertilizers, plastics, glass, and other industrial products that many consider essential depend on fossil fuels for 80-85% of their energy consumption.
6. Global Temperature Rise: Since the 1950s, the global temperature has increased by nearly 1°C. Projections to 2100 suggest temperatures will be higher than in a million years. Is the world willing to take this risk?
7. Sea Level Rise: The average sea level rise due to global warming is currently 2-3 mm per year. If the current rate continues, sea levels could rise by approximately 3 feet by 2100, which will devastate many coastal countries. Singapore is investing $74 billion over the next century to strengthen its borders against rising sea levels, equivalent to $740 million each year.
8. Coral Reef Degradation: If global warming continues unchecked, 90% of the world’s coral reefs could be functionally degraded by 2050. These reefs provide food, livelihoods, and cultural heritage to approximately half a billion people worldwide.
9. Extreme Weather Patterns: The “killer quad” of droughts, wildfires, flooding, and hurricanes, have not worsened over the last fifty years, even though global temperatures have risen 1°C.

Headwinds for New Electricity for AI Data Centers

1. A data center is built on banks of computers that require significant electricity. The additional supply can come from gas-fired or green energy, but the expansion will require larger transmission lines and more transformers, etc.
2. Infrastructure Bottlenecks: The expansion of transmission lines in the U.S. faces delays, and the acquisition of new transformer equipment is currently experiencing 2-3 year backlogs. This situation impedes the delivery of electricity (regardless of its source) to AI data centers.
3. Cost Comparisons: Electricity is generally costlier than natural gas in the U.S. The difference reflects the total costs of electricity/gas production and distribution. In Texas and California industrial plants, electricity often costs about five times more on an equivalent energy basis, while residential costs are 2.8 and 3.6 times more, respectively. However, new SoWiBess installations could cost less than new-build facilities that burn gas.

Policy for AI Power: Gas or Green?

1. State and International Leadership: Several states and countries have surpassed global averages in green power adoption. South Australia stands out, increasing its electricity from SoWiBess renewables from 1% to 74% in just 16 years. Projections forecast this figure to reach 85% by 2025/2026 and potentially 100% by 2027.
2. Advancements in Battery Storage: Battery energy storage systems (BESS) have increased storage capacity to almost 8 hours of dispatchable energy, while prices dropped by 40% in the last year. Australia has emerged as a leader in grid batteries. A new battery project at Stoney Creek (125 MW/1000 MWh) is designed for eight-hour power supply, a significant improvement over the typical 2-4 hours. The New South Wales (NSW) government is also supporting the development of five 8-hour batteries to facilitate the state’s shift away from coal-fired power. The cost of battery technology at Stoney Creek declined by 40% in the past year due to price changes in China.
3. Renewable Capacity Growth in the U.S.: Renewables (excluding nuclear) accounted for roughly 90% of newly installed capacity in 2024. Renewables and battery storage (SoWiBess) now constitute 30% of U.S. power capacity. In the U.S., installed solar capacity now totals 220 GW and can provide 7% of overall U.S. electricity. BESS doubled in 2024 to 29 GW and is expected to climb almost 50% in 2025. Batteries make U.S. power more stable.
4. SoWiBess vs. Gas Plant Timelines: SoWiBess can be deployed much faster than gas-fired power plants. Gas-fired plant construction might be problematic with the AI’s urgent need for power due to existing backlogs for gas turbines. GE’s backlog of gas turbines was $73 billion as of January 1, 2025, which illustrates the problem. Creative suggestions propose linking new renewable sources to existing peaker gas plants, in a concept called “power couples.” When SoWiBess power dips at night, the gas peaker steps in—but the gas peaker model is more expensive at $200/MWh, compared to SoWiBess at $40/MWh.
5. Prioritizing SoWiBess for AI Power: The case for expanding SoWiBess over natural gas, and other options like SMRs or geothermal, seems clear, due to commercial and cost factors. Building more gas-fired power plants and pipelines is also not the quickest route to new supply. Mark Christie, the head of FERC, is pushing combined-cycle gas turbines that offer consistent baseload supply. However, with the most recent data, levelized cost of electricity (LCOE) data show that SoWiBess is most cost-effective.
6. Inefficiency and Pollution of Mobile Gas Turbines: Data centers are increasingly utilizing mobile gas turbines to meet power needs, but this method is both inefficient and environmentally damaging. Elon Musk’s AI company, xAI, is utilizing mobile turbines in its new Colossus data center in Memphis, Tennessee, for backup power. These turbines emit pollutants like nitrous oxide and formaldehyde, but public agencies also have not authorized their use.

In summary, while the energy demands of AI are considerable, and require careful planning and investment, renewable energy sources, especially coupled with battery storage, offer a viable, cost-effective, and environmentally responsible path forward. It will be imperative for policymakers and industry stakeholders to prioritize and accelerate the deployment of SoWiBess and other sustainable energy infrastructure to meet the needs of the growing AI sector, minimizing environmental impact and long-term costs.