Power Consumption of AI Computing Servers

AI servers are among the fastest-growing electricity consumers, with individual AI racks drawing over 130 kW today and global AI-related data center consumption projected to nearly double by 2030.

Current Energy Use

AI data centers are consuming energy at a rapidly increasing rate. In 2024, U.S. data centers alone used approximately 183 TWh, representing 4.4% of total U.S. electricity, and this is expected to rise to 325–580 TWh by 2028, or 6.7%–12% of U.S. electricity generation, primarily due to AI workloads . Globally, data center electricity consumption is projected to reach 945 TWh by 2030, with AI-accelerated servers accounting for almost half of the net increase .

AI Server Power Requirements

Modern AI servers rely heavily on GPUs and specialized accelerators. For example, a single NVIDIA GB200 NVL72 AI rack draws 132 kW, more than 16 times the power of a traditional server rack, and by 2028, AI racks are projected to reach 1 MW per rack . These high-power racks require substantial infrastructure, including oversized UPS systems and battery energy storage to handle rapid power spikes, as AI workloads can surge from idle to full power in milliseconds .

Energy Distribution in Data Centers

Electricity consumption in AI data centers is distributed across several components :

• Servers and accelerators: ~60% of total consumption
• Storage systems: ~5%
• Networking equipment: up to 5%
• Cooling and environmental control: 7–30%, depending on efficiency
• Backup power systems: rarely used but essential for reliability High-performance AI workloads, such as video generation, multi-step reasoning, and autonomous agent tasks, consume hundreds to thousands of times more energy per query than simple text generation, offsetting efficiency gains from per-query improvements .

Trends and Implications

The growth of AI servers is driven by three competing trends :

1. Improving efficiency: Energy per query has dropped significantly for simple tasks.
2. Surging adoption: Active users and AI workloads are increasing rapidly.
3. Energy-intensive use cases: Complex AI tasks require massive computational power. This surge in demand is reshaping data center planning, grid infrastructure, and energy management strategies. Utilities and operators must consider renewable integration, waste heat utilization, and demand-flexibility mechanisms to maintain grid stability while supporting AI growth .

Summary

AI servers are transforming the energy landscape of data centers. Individual racks now consume over 130 kW, with projections reaching 1 MW per rack. Globally, AI-driven electricity demand is growing at roughly 30% per year, and by 2030, AI infrastructure could account for nearly 3% of global electricity consumption. Efficient hardware, cooling, and energy management are critical to sustaining this growth without overwhelming power grids .