Data Centers Will Consume as Much as Japan by 2030
Data centers worldwide will consume 945 terawatt-hours of electricity in 2030, equivalent to Japan’s annual consumption. In the United States, they will capture half of the growth in electricity demand and consume more than aluminum, steel, cement, and chemicals combined. This energy explosion from artificial intelligence raises a major political question: who will power this technological revolution?
The answer is striking in its brutality. While technology giants multiply their carbon neutrality commitments, 175 terawatt-hours of additional electricity will come from natural gas. The race for AI becomes an electric battle, and gas is winning it against renewables.
The Essentials
- Global data center consumption will double by 2030 to reach 945 TWh, approximately 3% of global electricity
- In the United States, they will represent 50% of electricity demand growth between 2023 and 2030
- 175 TWh of additional electricity will come from natural gas, compared to 85 TWh for renewables
- This demand exceeds the combined consumption of the four largest American industries
The United States Concentrates the Demand Explosion
The geography of AI is redesigning the American electrical map. By 2030, data centers will consume 260 terawatt-hours in the United States, a 160% increase compared to 2022. This increase is equivalent to adding Florida’s entire electricity consumption to the existing grid.
The International Energy Agency documents this shift in its “Energy and AI” report published in April 2025. Data centers will represent 6% of American electricity consumption in 2030, compared to 4% today. Even more striking: they will capture half of the country’s total growth in energy demand.
This geographic concentration amplifies tensions. Virginia, the global hub for data centers with 5 gigawatts of installed capacity, is seeing electricity demand grow by 5% per year. Texas follows with 3.2 gigawatts, attracted by competitive electricity rates and favorable regulation. These two states concentrate 40% of American intensive computing capacity.
The comparison with traditional industry reveals the magnitude of the shift. By 2030, data centers will consume more electricity than the four most energy-intensive sectors combined: aluminum, steel, cement, and chemicals. This virtual industry now surpasses traditional industry in the American energy hierarchy.
Natural Gas Wins the Race Against Time
Faced with this explosive demand, the United States makes a pragmatic choice that contradicts its climate ambitions. 175 terawatt-hours of additional electricity will come from natural gas by 2030, more than double the 85 terawatt-hours expected from renewables.
This dominance of gas is explained by deployment speed. A gas power plant is built in 2 to 3 years compared to 5 to 7 years for an offshore wind farm. Data center developers, pressured by demand, prioritize short-term energy certainty over long-term sustainability.
The Biden administration is betting on decarbonized electrification. The Infrastructure Investment and Jobs Act allocates $65 billion to modernizing the electrical grid. But authorization procedures slow down renewable projects: 8 years on average to connect a wind farm to the grid compared to 18 months for a gas plant.
This temporal asymmetry mechanically favors fossils. Microsoft, which commits to being carbon-negative by 2030, signs 10-year gas supply contracts to power its Azure data centers. Amazon follows the same logic with its AWS: carbon neutrality promised for 2040, gas plants under contract until 2035.
The Technology Industry Faces Its Energy Contradictions
Technology giants navigate between climate promises and energy reality. Google now consumes 24 terawatt-hours per year, more than Portugal as a whole. Meta reaches 18 terawatt-hours, Microsoft 16. Together, these three companies consume more electricity than 100 countries.
Their carbon neutrality commitments clash with the urgency of AI deployment. Recent estimates suggest that ChatGPT consumes approximately the same amount of energy as a Google search, contrary to initial assessments that estimated it much more energy-intensive. Training GPT-4 required 50 gigawatt-hours, equivalent to the annual consumption of 4,600 American households.
This contradiction fuels a parallel market for green certificates. Technology companies buy renewable energy guarantees to offset their fossil consumption. In 2024, they represent 60% of the global market for renewable energy certificates, worth $23 billion in purchases.
But this accounting compensation masks a physical reality: the electricity consumed comes from the local grid, predominantly fossil-based in 30 American states.
The American Electrical Grid Under Pressure
This additional demand tests the limits of the American electrical grid. The Electric Reliability Council of Texas (ERCOT) anticipates growth of 80 gigawatts by 2030, with 40% attributed to data centers. This is equivalent to 80 nuclear power plants.
Northern Virginia illustrates this saturation. Dominion Energy, the area’s main utility, is postponing new data center connections due to insufficient capacity. The queue reaches 50 gigawatts of projects, ten times the current installed capacity in the state.
This congestion pushes developers toward less saturated regions. Ohio now attracts projects thanks to competitive industrial rates and available grid capacity. Iowa follows, attracting Meta and Google with tax incentives and favorable electricity mix.
But this geographic dispersion does not solve the fundamental problem: the expansion pace exceeds the grid’s adaptation capacity. American Electric Power Company estimates that 40% of new connection requests come from data centers, compared to 10% in 2020.
Toward an Energy Geopolitics of Artificial Intelligence
This energy dependence is reshaping geopolitical balances. The United States imports 20% of its natural gas from Canada to power its digital industry. This energy interdependence creates a new form of technological sovereignty.
China follows an inverse strategy by favoring coal for its data centers. With 40% of global computing capacity, it powers 70% of its centers with coal, worsening its carbon footprint. But this approach guarantees it the energy autonomy needed for its AI ambitions.
Europe is attempting a third path by conditioning its public investments in AI on the use of renewable energy. The Horizon Europe program allocates €4 billion to AI research under carbon neutrality conditions. But its computing capacities remain marginal compared to American and Chinese giants.
This energy divergence could determine tomorrow’s technology leaders. AI requires abundant and reliable energy. Countries that solve this energy equation will gain an advantage in the technological race. The United States is betting on transitional gas, China on autonomous coal, Europe on renewable sovereignty.
The question remains whether this energy battle compromises the climate objective that partly justifies the technological innovation itself.