As the demand for AI computing power heats up, how is power management in data centers evolving?

In the past, discussions regarding energy in data centers focused primarily on total electricity consumption. However, with the rapid expansion of AI computing power, the conversation has shifted to a more specific question: exactly where is that power being used?
Data from the IEA indicates that global data center electricity consumption stood at approximately 485 TWh in 2025 and is projected to reach nearly 950 TWh by 2030—effectively doubling. Meanwhile, AI is driving a rapid increase in rack power density. For data centers, the growth in computing power entails not only higher electricity consumption but also fundamental changes to the entire power supply and distribution system.

Look at the Big Picture First, Then the Details

In early September, Dubai, Beijing, Jakarta, and Melbourne successively hosted exhibitions focused on data centers and energy. Topics such as high-density UPS systems, liquid cooling, 800V high-voltage DC power, and power quality frequently took center stage; meanwhile, metering capabilities began extending beyond the main power input to encompass UPS units, power distribution units (PDUs), individual rack circuits, and cooling equipment.
As data centers scale up and workloads become increasingly complex, relying solely on main utility meters makes it difficult to answer practical operational questions: Which segments carry the heaviest loads? Where are energy losses occurring? How much energy is consumed by cooling systems versus IT equipment?
Consequently, energy management is shifting from a focus on aggregate totals to an analysis of consumption structure.

When it comes to electricity, a single, aggregate figure is no longer sufficient.
Metering the main power input, UPS units, branch circuits, and cooling systems separately allows for the monitoring of fluctuations across different loads, providing a granular data foundation for energy efficiency analysis and PUE calculations. For liquid cooling systems, the power consumption of auxiliary equipment—such as Coolant Distribution Units (CDUs) and circulation pumps—is also increasingly becoming a focal point.
Underlying this is a simple reality: as data centers grow more complex, relying on a single total figure for energy data is no longer adequate.
The *Implementation Plan for Promoting the Synergistic Transformation and Development of Digitalization and Greenization (2026–2030)*, released on September 4, further advocates for supporting liquid cooling technologies, exploring 800V high-voltage DC power architectures, and promoting the deployment of single racks exceeding 100 kW in capacity. It also calls for the real-time monitoring, forecasting, and dynamic adjustment of energy efficiency in computing infrastructure.
From technological evolution to policy mandates, the requirements for energy data in data centers are becoming increasingly specific.

Metering Must Evolve with Change

As power distribution architectures, cooling methods, and rack power densities shift, metering can no longer rely on outdated, single-point monitoring.
For data centers, the priority is not merely adding more electricity meters, but establishing a data framework that is accurate, continuous, and granular—and capable of integrating with energy management systems.
Addressing various power distribution levels, Eastron offers the MCS multi-circuit metering solution and the SDM630MCT-ML series. These products enable sub-metering for branch circuits, row-based power distribution units (PDUs), and auxiliary equipment, while connecting to upper-level systems via communication protocols such as RS485 Modbus and Ethernet.
From a single meter to a dataset, and ultimately to a comprehensive energy management system, intelligent metering is playing an increasingly vital role.
AI drives up data center power demands, yet simultaneously makes it more critical than ever to truly understand that energy usage.