Trend 1: Prefabrication and modular construction are becoming mainstream.

The rise of large models has triggered an explosive surge in demand for intelligent computing power across industries. According to IDC data, from 2022 to 2026, the annual compound growth rate of AI computing power in China will reach 52.3%. Amid this boom, users are increasingly seeking Data center The requirements for delivery cycles are becoming increasingly stringent. However, traditional Data center Construction involves multiple phases, including design, civil engineering, MEP installation, and commissioning, to build a facility that supports ten thousand cards. Data center Under normal circumstances, a turnkey delivery cycle from the civil construction phase onward typically takes about 12 months; in regions of northern China with permafrost, the timeline is even longer. Moreover, AI and other IT technologies are advancing at a breakneck pace, driving a rapid increase in data center rack density: just three years ago, the typical power capacity per standard rack was 5–10 kW, whereas today even a single AI server supporting large models can consume more than 10 kW—far outstripping the pace of traditional construction.

Consequently, prefabricated and modular construction approaches have emerged. Prefabricated data centers integrate the principles of prefabricated building with those of modular data center design, using shipping containers as the primary platform to house a wide range of integrated systems. They offer rapid deployment, flexible scalability, and high energy efficiency. Guo Rensheng, Senior Vice President of Prologis China, notes that these prefabricated and modular methods can deliver projects on time, to specification, and with assured quality in significantly shorter timelines, making them the future mainstream. Inspur Information has launched a prefabricated AIDC solution that employs a prefabricated container-based modular design—allowing for assembly much like building with LEGO bricks—thereby decoupling the project from on-site civil works and removing constraints imposed by site conditions. All mechanical and electrical equipment is pre-integrated and pre-tested at the factory, requiring only connection to the grid to become operational. The prefabrication and assembly rate reaches 90%, and the construction schedule is substantially shorter than that of traditional civil engineering projects. Computer room Construction time can be reduced by more than 60%, with projects completed in as little as three months and equipment delivered and installed within 17 days. Huawei employs a prefabricated modular construction approach, enabling parallel factory fabrication and on-site assembly, which cuts delivery timelines by 50%. This method not only accelerates project delivery but also improves the yield of high-quality equipment, ensures the quality of data center construction, and helps reduce carbon emissions across the entire lifecycle.

Trend 2: Continued Deepening of Green Technologies

With the advancement of the “dual carbon” strategy, the green transformation of data centers has permeated every aspect of their operations. The year 2025 marks a critical milestone in achieving carbon neutrality for data centers. From a policy perspective, numerous documents have set forth clear requirements regarding key performance indicators such as the Power Usage Effectiveness (PUE) of data centers. For instance, the “Guiding Opinions on Accelerating the Establishment of a Nationally Integrated Big Data Center Collaborative Innovation System” stipulate that by 2025, data centers across the country should have established an integrated infrastructure landscape characterized by rational layout, green practices, and intensive resource utilization. Additionally, the “Action Plan for Enhancing Industrial Energy Efficiency,” jointly issued by the Ministry of Industry and Information Technology and five other departments, specifies that by 2025, the PUE of newly built large and ultra-large data centers must be better than 1.3.

From the perspective of energy consumption share, IT equipment accounts for over 60% of total consumption, while among non-IT equipment, air-conditioning systems represent the largest share, exceeding 20%. Consequently, at present, the focus of energy-saving and emission-reduction technologies in data centers is squarely on cooling systems, including liquid cooling, indirect evaporative cooling, natural cooling, and magnetic-levitation chillers. Notably, liquid cooling is poised for explosive growth in 2024, a trend that is expected to persist into 2025. Inspur Information unveiled its “All-in Liquid Cooling” strategy as early as 2022, spearheaded the release of China’s first set of core-component standards for cold-plate liquid-cooled data centers, and has built Asia’s largest liquid-cooling facility with an annual production capacity exceeding 300,000 units. Beyond hardware vendors, IDC service providers such as Qinhuai Data have also begun developing products and solutions in the liquid-cooling space.

Trend Three: High-Density Development

As IT computing power continues to evolve and CPU and server power densities keep rising, growing demand for AI applications is further increasing the share of AI workloads in overall data center compute. To strike a balance between efficiency and cost, the trend toward higher-density data centers is inevitable. Currently, the average power per rack stands at 6–8 kW; by 2025, 15–20 kW per rack is expected to become the norm. The shift to high-density AI workloads is driving GPUs to increasingly replace CPUs as the core of high-performance computing, which in turn places greater demands on data center power supply and cooling systems. AI racks require UPS systems, battery banks, and power distribution and switching equipment with even higher power densities to handle substantial load fluctuations. At the same time, hybrid cooling solutions will see wider adoption, with liquid cooling systems working in tandem with high-density UPSs; moreover, more servers will be shipped from the factory with integrated liquid-cooling systems, thereby improving overall system energy efficiency.

Trend Four: Widespread Adoption of Elastic Architectures

IT equipment typically has a lifecycle of 3 to 5 years, with power density doubling every 5 years, while data center infrastructure has a much longer lifecycle of 10 to 15 years. This necessitates that data center infrastructure be highly resilient, supporting architectural flexibility and phased investments to optimize CAPEX while accommodating the evolution of 2 to 3 generations of IT equipment. Moreover, given the diverse nature of the IT workloads they support, data centers must also enable mixed deployments of IT equipment with varying power densities. For instance, some data centers have adopted hyper-converged infrastructure (HCI) and software-defined networking (SDN), breaking away from traditional siloed architectures to achieve elastic resource allocation, reduce redundant energy consumption, and meet the need for flexible resource orchestration across different business scenarios.

Trend 5: Rapid Delivery Becomes the Norm

The internet business has experienced rapid growth, leading to a sharp surge in data and traffic demand within a short period—thus necessitating the swift commissioning of data centers. Moreover, as data centers transition from support systems to production systems, faster time-to-market translates into quicker revenue generation. Currently, the typical TTM (time to market)—the duration from planning to operational deployment—is 9 to 12 months; going forward, this is expected to be reduced to under 6 months. The rise of prefabrication and modular construction approaches is precisely aimed at meeting the need for rapid delivery. In addition, some data centers are adopting a fully modular construction philosophy, extending modularity from individual components all the way to architectural modules. Computer room Modularization ultimately enables full modularity of data centers, delivering advantages such as rapid deployment, elastic scalability, simplified operations and maintenance, and high energy efficiency, thereby further accelerating data center delivery.

Trend Six: Full Digitalization and AI-Powered Intelligent Operations

Digitalization and intelligentization are inevitable milestones in the evolution of data center infrastructure. As IoT and AI technologies continue to mature, data centers will transition from domain-specific digitalization—covering operations and maintenance, energy efficiency, and overall management—to full-lifecycle digitalization and autonomous operation spanning planning, construction, O&M, and optimization, with AI being widely adopted across the board. For example, by leveraging digital twin technology to create virtual data center models that simulate operational states in real time and provide early warnings of equipment failures, Meta has reduced its fault-response time by 50%. Similarly, through the integration of AIoT—combining sensor networks with edge AI analytics—data centers can achieve millisecond-level monitoring of equipment, enabling Singapore’s STTelemedia data center to cut O&M costs by 15%. In addition, autonomous inspection robots are increasingly replacing human workers in hazardous environments such as high-temperature and high-radiation zones, thereby reducing safety risks.

Trend 7: Ultra-Simplified Power Supply and the Shift from Lead-Acid to Lithium Technologies

Traditional data center power distribution systems suffer from system fragmentation and complexity, large footprints, and difficulty in fault localization. A minimalist power architecture reduces the number of conversion stages, shortens power delivery distances, minimizes floor space requirements, and enhances rack utilization and overall system energy efficiency. Meanwhile, lithium-ion batteries offer significant advantages over traditional lead-acid batteries in terms of footprint and service life. As lithium-ion battery costs continue to decline, they are poised for large-scale deployment in data centers. For instance, several newly built data centers have adopted minimalist power architectures and selected lithium-ion batteries as backup power sources, effectively improving space utilization and reducing long-term operational costs.

Trend Eight: The Fusion of Wind and Liquid, and “Wind Advances While Water Retreats”

The increasing adoption of GPUs and NPUs is driving a rise in high-density computing environments, making liquid-cooling systems ever more prevalent. However, certain storage and compute workloads still operate in low-density settings; to accommodate the uncertain future demands of IT operations, cooling solutions must be capable of supporting both air-cooling and liquid-cooling architectures. Moreover, chilled-water systems feature complex infrastructure, which hinders rapid deployment and efficient operations and maintenance. In contrast, modular indirect evaporative cooling systems can significantly shorten deployment timelines, simplify O&M, and fully leverage natural cooling resources, thereby substantially reducing the power consumption of cooling systems. In regions with suitable climates, such systems are gradually replacing chilled-water solutions. Some data centers have adopted hybrid air–liquid cooling approaches, enabling flexible selection between air cooling and liquid cooling based on specific workload requirements, while also integrating indirect evaporative cooling to effectively enhance cooling efficiency and reduce O&M costs.

Trend Nine: BitWalt Integration

Reducing PUE does not necessarily mean that a data center’s overall energy consumption is optimized; energy use must be evaluated and optimized as a holistic system, rather than focusing solely on the energy infrastructure. By fostering full-stack, collaborative innovation across energy, IT, chips, data, and cloud, we can achieve seamless coordination between bits and watts, enabling dynamic energy savings and optimizing system-level energy efficiency. For example, some companies leverage AI-driven dynamic optimization, using machine learning to predict workload fluctuations and adjust cooling and power-supply strategies in real time, thereby reducing PUE to below 1.1. Others are innovating at the chip level by developing low-power chips that cut energy consumption at the source, then integrating these advancements with data-center energy-management systems to achieve synergistic optimization of bit and watt performance.

Trend 10: Security and Trustworthiness Are Highly Prioritized

As the level of intelligence in data center infrastructure continues to rise, the cybersecurity threats it faces have also increased exponentially. Data centers must possess six key attributes—resilience, security, privacy, safety, reliability, and availability—to mitigate threats posed by environmental factors and malicious actors, including network intrusion threats. Particularly in today’s environment, where ransomware attacks are increasingly frequent, cybersecurity experts, network administrators, and data centers Operator It is imperative to stay abreast of emerging trends and develop cutting-edge AI security technologies to address the challenges ahead. Some data centers have adopted a zero-trust architecture, leveraging the SASE (Secure Access Service Edge) framework to ensure seamless protection between edge nodes and the core data center. Meanwhile, other organizations are employing quantum encryption to safeguard sensitive data and utilizing federated learning and homomorphic encryption to perform analytics without exposing raw data, thereby complying with regulatory requirements and ensuring robust data security.

The data center industry in 2025 is brimming with both opportunities and challenges, as emerging trends drive the sector toward greater efficiency, sustainability, intelligence, and security. Industry peers, are you ready to embrace these changes? Let’s come together to explore how we can strategically position ourselves in light of these new trends—so that our data center business can reach even greater heights!