As a global leader in infrastructure construction and digital services for data centers and critical industry applications, since 2018 Since the year , Schneider Electric has consecutively 7 At the beginning of each year, Schneider Electric releases the “Visible Future—New Trends and Breakthroughs in the Data Center Industry” series of insights, pioneering forward-looking analysis of industry trends and consistently guiding the direction of future transformation, thereby injecting strong momentum into the data center sector. Grounded in deep industry insights and practical experience, Schneider Electric is committed to illuminating the changes that will unfold in the data center industry in the coming year, the value and significance of these changes and trends for data center operators, as well as its perspectives and value proposition regarding these industry shifts. Below is the perspective from Schneider Electric’s Global Data Center Research Center on 2024 Forecast of annual development trends.

Trend 1: Intelligent Computing Centers Will Lead Data Center Development

Over the past decade, cloud computing has been the primary driver behind the construction and evolution of data centers, with the goal of providing the general-purpose computing power needed for digital transformation across society. However, AI The outbreak has created enormous demand for computing power, in order to meet AI To train and deploy large-scale models for inference, we need to build a substantial number of intelligent computing centers. Based on global data center electricity consumption, Schneider Electric, Graphics Processing Unit Chips and AI Based on future server shipment data and other metrics, the current global power demand of intelligent computing centers is estimated to be 4.5 GW , accounting for the total of data centers 57 GW of 8% , and predict that 2028 This year it will 26%-36% the compound annual growth rate, ultimately reaching 14.0 GW To 18.7 GW , accounting for the total 93 GW of 15%-20% . This growth rate is the annual compound growth rate of traditional data centers ( 4%-10% ) of 2 To 3 times. The distribution of computing power will also shift from the current centralized deployment (centralized versus The edge is 95%:5% ) Migration to the edge ( 50%:50% ), which means that intelligent computing centers will lead the trend in data center construction. According to the Ministry of Industry and Information Technology’s plan, the share of intelligent computing power in our country will be 2025 Reach the year 35% , with an average annual compound growth rate of 30% That concludes the discussion. Schneider Electric believes that, compared with traditional data centers, the construction of intelligent computing centers must prioritize sustainable development and forward-looking design while ensuring high energy efficiency and high availability—specifically, minimizing environmental impact and, in particular, enhancing adaptability to meet future needs. IT Demand for technology (high-power chips and servers).

Trend Two  ： A I To put   Promote   Sharp increase in cabinet power density

Cabinet power density has a significant impact on data center design and construction costs, including power distribution, cooling, and IT The layout of the equipment room, among other factors, has always been one of the key design parameters of concern for data centers. Uptime Research conducted over the past few years has shown that the power density of server cabinets has been steadily but gradually increasing. The average power density of cabinets is typically below 6 kilowatts; most operators do not exceed 20 kilowatt-rated cabinets. Factors driving this trend include Moore’s Law, which has kept the thermal design power of chips at a relatively low level ( 150 watts), while high-density servers are typically deployed across multiple racks to reduce infrastructure requirements. However, AI The outbreak will reverse this trend, according to research by Schneider Electric, which found that the resources used for training AI The power density of the cabinet can be as high as 30-100 kilowatts (depending on the chip type and server configuration). The reasons for this high density are multifaceted, including rapidly increasing CPU/GPU Thermal Design Power, Central Processing Unit for 200-400 Watt, Graphics Processing Unit for 400-700 The level will continue to rise in the future; AI Server power consumption is typically around 10 Around several kilowatts, because Graphics Processing Unit It works in parallel. AI Servers must be deployed in a compact, clustered configuration to minimize network latency between chips and storage. The sharp increase in cabinet power density will pose significant challenges to the design of data center physical infrastructure.

Trend 3: Data centers are transitioning from air cooling to liquid cooling.

Air cooling has always been the standard in data centers. IT The mainstream approach to data center cooling, when properly designed, can support cabinet power densities of over ten kilowatts. However, as the demand for AI In the relentless pursuit of enhanced training performance, developers have continually pushed the thermal design power limits of chips, rendering air cooling for these chips increasingly impractical. Although some server vendors are constantly pushing the boundaries of air-cooling technology by redesigning chip heat sinks, increasing server airflow, and optimizing inlet–outlet temperature differentials, they are configuring 40-50 Kilowatt air-cooled AI cabinets, but this will cause the fan’s power consumption to increase exponentially. For example, AI Server fans can consume up to 25% server power, but the typical value for traditional servers is only 8% Schneider Electric believes that chip cooling is the primary driver of liquid cooling. 20 The power density of a kilowatt-class cabinet marks the relatively reasonable dividing line between air cooling and liquid cooling. When AI When the power density of a server cabinet exceeds this threshold, liquid-cooled servers should be given serious consideration. Compared with air cooling, liquid cooling offers numerous advantages, including enhanced processor reliability and performance, improved energy efficiency, reduced water consumption, and lower noise levels, among others. Currently, for high-density AI Servers: vendors typically offer both air-cooling and liquid-cooling solutions, but for the next generation Graphics Processing Unit , liquid cooling will be the only option.

Trend Four ： The safety and reliability of power distribution are even more critical in intelligent computing centers.

For traditional data centers, the probability that different workloads simultaneously reach their peak levels is extremely low. For example, the typical peak-to-average ratio in a large data center is usually 1.5–2.0 or higher. However, in intelligent computing centers, due to AI Lack of variation in training load (peak-to-average ratio close to 1.0 ), the workload can operate at peak power for several hours, days, or even weeks. As a result, the likelihood of upstream large circuit breakers tripping increases, along with the risk of system downtime. At the same time, as cabinet power density rises, it becomes necessary to use circuit breakers, row-end cabinets, and busbars with higher current ratings. Meanwhile, as resistance decreases, the fault current that can flow also increases, which means IT The risk of arc flash in the data center room also increases, making the safety of personnel in this area a critical challenge that must be addressed. Schneider Electric recommends using simulation software during the design phase to conduct arc flash risk assessments of the power system, analyze potential fault currents, and evaluate system reliability, thereby enabling the development of the optimal solution tailored to the specific site. This analysis should cover everything from medium-voltage switchgear down to the cabinet level, and it is further recommended that, for new data centers, IT of the computer room AI Training workload exceeds 60-70% The rating of the main circuit breaker shall be determined based on the sum of the ratings of all downstream feeder circuit breakers, and the simultaneity factor is no longer considered in the design.

Trend Five   Standardization Will Be Key to Advancing Liquid Cooling

Cold-plate liquid cooling and immersion liquid cooling are the two mainstream approaches for liquid cooling in data centers. The question of which liquid-cooling method to choose and how to achieve rapid deployment has long been a hotly debated topic in the industry. As more and more AI The servers employ cold-plate liquid cooling, which is also more easily compatible with traditional air-cooling systems and thus favored by many data center operators. However, server manufacturers offer a wide variety of liquid-cooling designs, including quick-connect fittings, blind-mate connectors, and Manifold There are numerous compatibility issues. IT The boundaries of responsibility for infrastructure are also unclear, which significantly limits the acceptance and adoption of liquid cooling in data centers. Compared with cold-plate liquid cooling, immersion liquid cooling that uses fluorocarbon-based fluids is not only relatively more expensive, but many fluorocarbons are synthetic chemicals that are harmful to the environment, subjecting this technology to increasing industry regulation and policy pressure. Consequently, aside from oil-based coolants, the range of viable fluorocarbon-based fluids available for immersion liquid cooling is steadily shrinking. Schneider Electric recommends IT Manufacturers provide more standardized design solutions, covering fluid temperature, pressure, flow rate, equipment interfaces, and other key parameters, while also defining clearer responsibility boundaries. Schneider Electric will release a white paper on liquid cooling in the first quarter to help data centers better deploy liquid-cooling technologies.

Trend Six   : Data centers will pay more attention to Water Use Efficiency

Water scarcity is emerging as a serious issue in many regions, making it increasingly important to understand and reduce the water consumption of data centers. Previously, one major reason why data center water use was overlooked was that water costs were typically negligible compared with electricity costs; indeed, many data centers even increased their water consumption in order to improve energy efficiency. However, data center water use has now drawn significant attention from local governments, particularly in water-scarce areas, where authorities are introducing policies to restrict and optimize water usage by data centers. These measures include… Water Use Efficiency As a design metric for data centers, a dual water-and-electricity control policy is adopted. Consequently, reducing water consumption will become a key area of focus for many data center operators in the future. Through its research on water usage in the data center industry, Schneider Electric believes that data centers’ Water Use Efficiency Value is in 0.3–0.45 L/kWh This represents a relatively optimal value. Schneider Electric recommends striking a balance between electricity and water usage based on the local water resource availability, climate conditions, and data center type. The industry can leverage various technological innovations—such as adiabatic evaporation, indirect evaporative cooling, and liquid cooling—to reduce direct water consumption. Data center operators should Water Use Efficiency Report water usage as part of the Sustainable Development Goals. / Conserve water while also paying attention to the indirect water use associated with electricity consumption.

Trend 7: Enhancing power distribution capacity will become a new requirement for intelligent computing centers.

At the intelligent computing center, as cabinet power density increases and AI Clustered deployment of server cabinets, IT The data center’s power distribution system faces the challenge of having a rated capacity that is too small. For example, in the past, a 300 kW The power distribution module can support dozens, even hundreds, of server cabinets. Yet today, the same power distribution module often cannot even meet the power requirements of a minimum-configured system. NVIDIA DGX SuperPOD AI Cluster (single row 358 kW of 10 One cabinet, per cabinet 36 kW ). The power distribution module specifications are too small; using multiple power distribution modules not only results in waste IT Space also becomes impractical. Compared with a single high-capacity power distribution module, using multiple modules further increases costs. Returning to the essence of power distribution, the primary way to increase distribution capacity is to raise the current. Schneider Electric recommends selecting power distribution modules with sufficiently high specifications during design to enable flexible deployment, thereby accommodating future power distribution needs—specifically, to support at least an entire rack of servers. For example, at the rated voltage, 800 A The distribution module is currently applicable to all three types of power distribution ( PDU ， RPP Standard capacity sizes for busbars, which can provide 576  kilowatts (reduced to 461  kilowatts). For end-of-line distribution, small busbars can be used, thereby avoiding the need to customize ones larger than 63 A Cabinet with rated current PDU . Where space permits, multiple standardized cabinets may be used. PDU As a transition.

Trend Eight: AI Empowering energy-efficient upgrades for data centers

Data centers provide AI Computing power is driving human society toward more sustainable pathways—such as automation, digitalization, and electrification—thereby empowering the transportation, manufacturing, and power-generation sectors to reduce their environmental impact. Conversely, AI It can also empower the optimization of data center energy use, thereby reducing its own environmental impact. For example, AI Machine learning techniques can be applied to the control of data center cooling systems and air-conditioning terminal units. By analyzing historical data, these techniques enable real-time monitoring of airflow distribution within the data center and, based on the data center… IT Real-time matching of appropriate cooling capacity to accommodate load variations. By automatically adjusting the operating modes of precision air conditioners and fans at the terminal units, on-demand cooling is dynamically delivered, thereby reducing hotspots and lowering data center energy consumption and O&M costs. Schneider Electric believes AI The application of technology in data center air-conditioning cluster control systems enables intelligent monitoring and control of indoor environmental parameters, and through automatic adjustment and optimization, enhances energy efficiency and system reliability, thereby achieving energy conservation and emission reduction. With AI With the continued widespread adoption of technology and the ongoing national requirements for energy conservation and consumption reduction in data centers, whether for new construction or renovation projects, AI Technology will receive greater attention and wider application in data center air-conditioning cluster control systems.

Trend Nine ： The footprint of the distribution system will attract attention.

In data center design, the pursuit of IT Maximizing the utilization of data center floor space—specifically, minimizing the footprint of auxiliary equipment—is consistently one of the primary design objectives for data centers. For traditional data centers, IT The ratio of the data center area to the distribution room area is typically 1.5:1 Left and right. As AI Drive IT The increasing density of server cabinets, with more and more IT The data center employs liquid cooling; liquid cooling. IT The ratio of the data center’s area to the distribution room’s area will reverse, and 0.6:1 left and right. At this point, the footprint of the distribution room will attract greater attention from data center designers, and optimizing the distribution room’s footprint will inevitably become a key development direction for the industry. Schneider Electric believes that increasing the power supply capacity of distribution and power equipment within a smaller footprint is one of the most effective approaches. For example, reducing UPS The system’s footprint, including modularity with higher-power power modules. UPS , achieving megawatt-level power per cabinet; meanwhile, by replacing lead-acid batteries with lithium-ion batteries, the floor space required for battery installation can be reduced. 40-60% Centralized deployment of power supply and distribution equipment (such as power skids) can also reduce the floor space required for switchgear rooms; furthermore, adopting compact, modular switchgear cabinets and pooling diesel generators as emergency power sources are also effective measures.

Trend 10: Energy Storage Systems   The value of data centers is increasingly evident.

Enter 2024 This year, the focus of the data center industry will shift from traditional data center construction to the development of intelligent computing centers, leveraging continuous technological innovation to achieve the sustainable development and next-generation adaptability of these centers. IT Technology is the key.

The above forecasts on emerging trends are sourced from Schneider Electric’s Global Data Center Research Center, which was established in the 1990s. The center has consistently upheld the mission of “exploring technological and developmental trends in the data center industry and promoting best practices.” It achieves this by publishing accessible white papers and decision-support tools that help data center operators enhance availability, optimize energy efficiency, foster sustainable development, and maximize business value. As of 2023 This year, the Schneider Electric Research Center team has already published a white paper. 230 The remaining articles, each year there are more than 40 Ten thousand downloads; trade-off tool 30 individuals, with more than each year 2 Tens of thousands of users are currently online. All white papers and trade-off tools are provided free of charge to the entire industry for learning and use, thereby not only advancing the development of the data center sector but also firmly cementing Schneider Electric’s position as a thought leader in the field.