As expected by the market, AI technology and its applications have been accelerated in 2024 and ushered in explosive growth. While reshaping the computing power demand pattern, it has also set off an unprecedented wave of innovation. As the core force of industrial transformation, data centers, accelerated computing infrastructure, power supply and distribution, and cooling systems are also evolving to meet the digital needs driven by AI workloads. In addition, with the vigorous development of AI technology, sustainable development has become a strategic focus of digital infrastructure construction. More and more data centers are turning to renewable energy and implementing effective demand management measures to reduce carbon emissions from digital infrastructure. Globally, almost all enterprises and hyperscale data center operators are actively adopting more efficient cooling technologies, setting net zero carbon emission goals, and making the construction of environmentally friendly data centers the top priority for development.
Looking back at the development trend of digital infrastructure in 2024 and the continued surge in AI demand, Schneider Electric believes that although the future is always full of uncertainty, we are still convinced that 2025 will still be a year full of change, innovation and growth, and all industries and fields will usher in new development opportunities.
AI computing power demand is mainly divided into two core functions: training and reasoning. Among them, training workloads are used for model building and optimization, while inference focuses on tasks such as decision making, content generation, and comprehensive automation. In the field of data centers, the application of AI inference technology has made significant progress, especially in the context of the growing demand for edge computing. The demand for real-time data processing has prompted it to be as close to the data source as possible to effectively improve operational efficiency. However, the current development trajectory of AI deviates from early expectations, especially for those companies that have built large training clusters for AI models (such as large language models) with strong accelerated computing capabilities. When these clusters have completed training tasks, they may transform into platforms for performing tasks such as inference, decision making, and content creation.
Previously, the industry generally expected that after the AI model training was completed, a micro and efficient inference cluster would be established close to the user side to achieve rapid deployment of edge AI. However, surprisingly, AI service providers did not deploy small edge AI systems near the data source, but tended to continue to use large centralized training clusters to perform inference tasks. This model undoubtedly promoted the rise of "inference data centers". For inference tasks that do not require high computing power, the use of training clusters may lead to the emergence of excessive computing power. Looking ahead to 2025, the shift to edge computing for reasoning will gradually increase as edge devices become more efficient, latency is reduced, data security is enhanced, and applications become more customized. Of course, this shift will be a gradual process, and companies will need to adjust their infrastructure to accommodate the growing demand for real-time AI applications. In the meantime, data center reasoning will remain the primary solution, even if it means using a lot of resources to handle smaller tasks.
Schneider Electric notes that in recent years, all parties are actively working to strengthen the collaboration between data centers and public utilities in forecasting and trend analysis, especially in optimizing power efficiency. Looking ahead, as the data sharing mechanism between public utilities and data centers matures, AI technology will drive data centers to be more deeply integrated into the public utility power ecosystem and play a key role. This will enable data centers to more reasonably choose power supply modes and seamlessly switch to off-grid mode when needed, switching to backup power. It is expected that in 2025, the demand for stable renewable energy (such as wind power, solar power, etc.) in data centers will become increasingly urgent, and the introduction of battery energy storage systems (BESS) will become the key to solving power shortages. Cooperation between data centers and utilities will deepen. Data center operators will optimize BESS charging cycles based on the availability of renewable energy, ensure sufficient power reserves, and discharge when relying on fossil fuel power generation or regulating power fluctuations. As the industry progresses, such cooperation is increasing, deepening and standardizing.
In the coming years, the increasing importance of power supply to data centers will prompt the industry to more actively explore a diversified energy structure, including renewable energy. The diversification of energy structure can not only meet the rising energy demand of data centers, but also effectively improve energy self-sufficiency. At present, many technology giants are increasing their investment in large-scale energy projects, showing that companies are paying more attention to ensuring the stability of power supply. In addition, many companies plan to install gas turbines on site to enhance the flexibility and reliability of energy supply. In this transformation process, the potential of small modular nuclear reactors (SMRs) is particularly prominent, which not only promises waterless operation, high safety and a smaller footprint, but also has the potential to use recycled uranium for power generation. Once SMRs are fully tested and approved by regulators, they may lead data centers and other industries to completely change the way energy is produced.
Looking ahead to 2025, data center operators and hosting service providers are working hard to build accelerated computing capabilities to better serve artificial intelligence applications. This trend may lead to a relative reduction in the construction demand for traditional enterprise-level and cloud-hosted data centers. The surge in market demand for data centers with accelerated computing and artificial intelligence capabilities has triggered fierce competition for power resources among related companies. Companies are actively negotiating with power suppliers to ensure adequate power supply before obtaining construction permits. In addition, it is worth noting that many commercial real estate companies that have not previously been involved in the data center field are entering this market, showing the attractiveness of the data center industry and the growth potential of its infrastructure investment.
There is no doubt that AI technology will continue to lead the future of the data center industry, not only promoting technological innovation and structural optimization, but also accelerating the realization of carbon neutrality goals, optimizing water resource utilization, almost 100% use of green materials, deploying advanced liquid cooling solutions, and widely applying AI technology to the design, maintenance, power supply and distribution system, backup power supply, and refrigeration system automation management of data centers are also key directions for the development of the industry. The rapid development of technology has prompted data centers to continue to iterate and evolve to adapt to changes in market demand and promote disruptive innovation and breakthroughs. We have every reason to believe that we will witness more exciting surprises and achievements in the future.