Summary of Conference Call Records Industry Overview - The records discuss the evolution of data center power and infrastructure architecture in the context of the AI era, highlighting the transition from traditional data centers to new Artificial Intelligence Data Centers (AIDC) [1][2][4]. Key Points and Arguments Transition to AIDC - Traditional data centers are facing disruptive changes due to the exponential growth in computing power demands driven by AI models. The new AIDC architecture includes a power base layer, platform service layer, model enabling layer, and industry application layer, emphasizing vertical integration within each layer [2][4]. - AI servers consume significantly more power than traditional servers, with projections indicating that total power consumption in data centers will double by 2030. The U.S., China, and Europe account for over 80% of global data center capacity, with the U.S. expected to contribute over 50% of the growth in global data center power demand [1][3]. HVDC Technology - The 800V HVDC technology is identified as a key solution to meet the megawatt-level power cabinet demands of the AI era. This technology replaces traditional lead-acid batteries with lithium-ion battery backup units (BBUs) and eliminates the need for uninterruptible power supply (UPS) systems, significantly improving power transmission efficiency [1][3][6]. - NVIDIA plans to transition to 800V HVDC data center power infrastructure starting in 2027 to support IT racks of 1 megawatt and above [3][6]. Cooling Technologies - As chip power consumption and rack power density increase, liquid cooling technology is becoming crucial for data center thermal management. The cooling approach is shifting from pure air cooling to a hybrid air-liquid model, with cold plate liquid cooling being the most widely used solution [1][8]. - The effective cooling boundary for air cooling is approximately 40-60 kW per rack, beyond which liquid cooling is considered more efficient and cost-effective [3][8]. AI Industry Development Stages - The AI industry has progressed through three key stages: 1. **Exploration Phase**: Initial theoretical and conceptual developments in AI. 2. **Foundation Phase**: Technological accumulation and breakthroughs, leading to practical applications in various fields. 3. **Industrial Boom Phase**: Significant growth in AI applications, driven by advancements in computing power and infrastructure, leading to a transformation of traditional data centers into AIDCs [4][5]. Market Projections - The AI data center market is expected to see substantial growth, with IT energy consumption projected to reach 77.7 terawatt-hours by 2025. The liquid cooling server market in China is anticipated to grow at a compound annual growth rate (CAGR) of 48.3% from 2023 to 2028, reaching a total market size of $10.5 billion by 2028 [2][10]. Supply and Demand Dynamics - The upgrade of computing infrastructure is primarily driven by the demand for computing power and the supply-demand gap. Since 2012, the computational requirements for AI training have doubled approximately every 3.4 months, significantly outpacing Moore's Law [11]. Additional Important Insights - The transition to AIDC is not only a technological shift but also a necessary evolution to support the increasing demands of AI workloads, which are pushing the limits of traditional data center capabilities [2][4]. - The development of high-speed optical modules is also highlighted, with the global market expected to reach approximately $14.4 billion by 2024, growing at a rate of 52% [12][13]. - Domestic companies in the AI chip sector are actively pursuing advancements and partnerships to catch up with global leaders, indicating potential investment opportunities in the evolving landscape of data centers and AI technologies [14].