Summary of Key Points
This news article highlights the issue of the surge in global data center electricity consumption amid the AI revolution, emphasizing that relying solely on traditional power grids or fossil fuels is no longer sustainable. A sustainable path for the co-development of AI and the electric power system requires expanding the use of renewable energy, innovating energy storage technologies, and making data centers more flexible partners in regulating the power grid.
1. Data Centers’ Insatiable Energy Demand: One Out of Every Five Kilowatt-hours Will Be Used by Data Centers
The electricity consumption of data centers is growing exponentially. According to the International Energy Agency, global data center energy usage will reach 945 terawatt-hours by 2030, equivalent to Japan’s annual electricity consumption. Over the next five years, 18% of the world's electricity growth will come from data centers, meaning that one out of every five kilowatt-hours of electricity will be used by them.
China and the United States are leading the increase in energy consumption: In the U.S., data center electricity usage will account for 14% of total electricity consumption by 2030, nearly half of industrial electricity use; in China, it is estimated to account for 3%-5%, which is equivalent to the annual electricity consumption of the entire Guangdong province or the energy demand of the aluminum and chemical industries. Even Mongolia is under pressure—its national grid capacity is only sufficient to power a large data center, and adding more computing projects would squeeze residential and industrial electricity use.
2. AI Data Centers Cannot Rely on Traditional Power Grids; Renewable Energy Is the Solution
Experts agree that relying on fossil fuels to support AI data centers is not viable. Fossil fuel usage exacerbates carbon emissions and contradicts sustainable development goals, and it also puts pressure on residential and industrial electricity use (as seen in Mongolia’s case).
However, renewable energy has a drawback: wind and solar power are intermittent (e.g., there is no electricity at night or when the wind doesn’t blow), while data centers require continuous power supply 24/7. This necessitates a solution to stabilize the fluctuations in renewable energy.
3. Energy Storage Is Key for Renewable Energy, but Lithium-Ion Batteries Are a Limitation; Sodium-Ion Batteries Are on the Way
Energy storage is essential for supplying data centers with renewable energy. Ningde Times’ chairman, Zeng Yuqun, noted that energy storage systems for AI data centers must withstand high voltage, high power, and high temperatures while maintaining safety. Additionally, the supply chain for energy storage must be stable. Currently, lithium mines, which are crucial for lithium-ion batteries, are subject to export restrictions by some countries, potentially limiting energy storage capacity. Therefore, Ningde Times is investing in sodium-ion batteries, planning large-scale production by 2026 to reduce reliance on lithium-ion batteries. However, it will take another 3-5 years before sodium-ion energy storage becomes widely available.
4. Data Centers Are More Than Just Power Consumers; They Can Also Help Regulate the Power Grid
While data centers were once seen as power consumers, they can now play a role in regulating the grid. For example, large AI models can be trained in remote areas with abundant renewable energy, but for real-time tasks (such as answering questions), data centers need to be located near users, which increases energy consumption and makes power demand more unpredictable, putting pressure on urban grids. Experts suggest transforming data centers into flexible load sources—using less electricity during peak times and more during off-peak times, and combining this with distributed energy storage (e.g., nearby battery arrays) and vehicle-to-grid interaction (recharging electric vehicles to provide power back to the grid). They also recommend developing more energy-efficient models.
5. A Multidisciplinary Approach Is Needed to Find a Solution
Solving this issue requires cooperation among governments, businesses, and regulatory authorities. Experts from DNV GL suggest that while AI can assist in grid planning, it cannot overcome institutional and physical limitations. Actions needed include:
- Regulatory Authorities: Treating AI data centers as a separate category of load and adjusting grid connection rules and management.
- Businesses: Assessing the grid’s capacity before building data centers to avoid last-minute adjustments.
- Policy Makers: Accelerating grid expansion while establishing transitional mechanisms to ensure reliable power supply and alleviate public concerns about electricity shortages.
Only by working together can we achieve a sustainable balance between AI development and the electric power system.