How Much Electricity Does a Data Center Actually Need?
"Give me a free one-day tour of Beijing's hutongs without backtracking."
"Remove the people passing by in this photo, keep it natural."
"Replace the braised duck in the snow mountain fox rescue video with salted duck."
Which of these AI suggestions would consume more electricity?
A few years ago, the power load (capacity requirement) of a single data center was often in the tens of megawatts (MW) range. However, with the growth in AI computing power demand, the power demand of data centers has continued to climb. According to a report by the International Energy Agency (IEA), global data center electricity consumption was approximately 415 terawatt-hours (TWh) in 2024, and may rise to approximately 945 terawatt-hours (TWh) by 2030 (this is subject to uncertainty depending on the scenario).
To give a comparison, the Three Gorges Dam's annual power generation in 2025 was approximately 95.715 billion kilowatt-hours (95.715 terawatt-hours). Therefore, the global data center's electricity consumption is equivalent to about four Three Gorges Dams' annual power generation.
Returning to the initial question: the energy consumption of a single AI request does not have a fixed value and is affected by many factors such as model size, output length, inference chain, concurrency, hardware, and data center. The general trends are:
Plain text, short outputs, and fewer inference steps are generally more energy-efficient.Complex inference, multi-turn dialogues, and long outputs significantly increase computational load.
Complex image generation is generally more computationally intensive than plain text.
Video generation/long video editing tasks often show the most significant increase in energy consumption.
Power planning needs to be aligned with AI data center construction.
Traditional power planning often operates on a decades-long timeframe; however, the development cycle of hyperscale data centers is much shorter, requiring strict alignment between power supply and market windows. Common practical constraints in implementation include:
Data center projects can take years just to build, not including the time required to commission new power sources.
New gas-fired power plants are often subject to restrictions such as permits, fuel, and pipeline conditions, with commissioning cycles potentially reaching 8-10 years. The delivery time for critical equipment such as large transformers, circuit breakers, turbines, and prefabricated substations is long, often requiring planning years in advance.
From the power sector's perspective, capital and a sense of urgency cannot bypass physical construction and regulatory processes. Therefore, both sides need to adjust their strategies: data centers should signal their actual demand earlier, and power companies/utilities should intervene earlier in power supply solutions and design selection to reduce mismatches and iterations in the "planning-implementation" process.
Extreme weather exacerbates power demand.
Extreme weather has become increasingly frequent in recent years. Even before the full load of data centers is added, extreme weather is already continuously testing the limits of the power grid.
Residential cooling loads are surging, forcing system scheduling to make difficult choices under high pressure. With the growth of hyperscale data centers, without careful integration, large loads will make reliability assurance more difficult. Regulatory and legislative discussions are also underway regarding whether to enforce load reductions for large industrial users such as data centers in emergency situations to ensure priority for life-critical loads. This means power companies need institutional and technological arrangements with "reducible capacity" while balancing contracts, customer relationships, and long-term partnerships.
Facing the next wave of demand growth, power grids and data centers require "faster and more stable" infrastructure.
As energy demands from data centers and AI continue to rise, utilities must strike a more delicate balance between expansion and reliability.
In this process, Belden Hirschmann's connectivity and networking capabilities can support critical scenarios in data centers and the power sector, including reliable connections to critical power sources and distribution, industrial network communication and data acquisition, and visualization and management capabilities for operations and maintenance, helping customers improve deployability, maintainability, and operational security during expansion.
News Source: Belden Hessmann WeChat Official Account
Cited Source: Energy and AI - Analysis - IEA: https://www.iea.org/reports/energy-and-ai/
2025 Three Gorges Project Bulletin: http://www.mwr.gov.cn/sj/tjgb/sxgcgb/202606/t20260629_2131904.html
Post time: Aug-10-2026