Why Data Centers Are the Power Bottleneck for AI – Trends, Costs & Green Solutions

Data Center Energy Consumption Overview

Data centers, as the operating platform for compute services, typically have electricity costs accounting for more than 50% of total operating expenses. The IEA predicts global data center electricity demand will rise to about 945 TWh by 2030, more than double the 2024 level of 415 TWh, with a compound annual growth rate of roughly 15% from 2024‑2030.

In China, the China Academy of Information and Communications Technology (CAICT) forecasts that by 2030 data center electricity demand will reach 300‑700 billion kWh, representing 2.3%‑5.3% of national electricity consumption and a CAGR of 10.4%‑27.1%.

Regional Distribution and Market Evolution

Since the 1990s, mainland China’s data center market has evolved from scattered small‑scale rooms to large‑scale facilities driven by the rise of the Internet, cloud computing, and AI workloads. The “East‑Data‑West‑Compute” strategy allocates low‑latency, high‑bandwidth services to eastern hubs and power‑intensive, latency‑tolerant tasks to western hubs, forming eight major hubs and ten clusters nationwide.

From 2023 data, the average Power Usage Effectiveness (PUE) of Chinese data centers fell to 1.48 (down from 1.52 in 2022). Ultra‑large centers achieve an average PUE of 1.33, while planned facilities target a design PUE of about 1.29. By the end of 2024 the national average PUE is expected to reach 1.46.

AI’s Growing Power Demand

AI workloads dramatically increase power consumption. Training a GPT‑3‑scale model in a Microsoft super‑computing center consumes roughly 190 MWh. A 100 k GPU cluster can exceed 150 MW, with annual electricity use approaching 1.6 billion kWh. The Uptime Institute projects AI services will account for up to 10% of global data‑center electricity use by 2025, up from 2% today.

Green Computing Initiatives

Green computing aims to improve the efficiency of compute devices (chips, servers, storage) and the data‑center infrastructure (cooling, power distribution, waste‑heat recovery). Key metrics include PUE, Water Usage Effectiveness (WUE), renewable energy utilization rate, and green‑certificate procurement. Over 140 Chinese data centers have achieved a green‑low‑carbon rating of 4A or higher, and many now recycle waste heat for district heating or agricultural use.

Advanced technologies such as high‑efficiency chips (e.g., Nvidia’s Blackwell, offering 1 000× AI performance with 350× lower energy per operation compared to Pascal) and integrated storage‑compute architectures further reduce energy intensity.

Future Outlook

By 2030, global data‑center electricity consumption is expected to exceed 945 TWh, with China and the United States leading the growth. Sustainable development will depend on continued PUE improvements, renewable energy integration, and coordinated compute‑energy planning.