Core Viewpoint - The GTC conference highlighted the transition from Power shelf to Power rack as the next-generation power solution for AI data centers, indicating a significant upgrade in the importance of power suppliers in the industry [2][4]. Group 1: Power Rack Development - The Power rack integrates multiple components such as Power shelf, BBU shelf, Super cap shelf, and PDU, enhancing the overall value and competitive barriers in the AI power supply industry [2][3][6]. - The shift to Power rack architecture addresses the increasing power demands of advanced computing nodes while optimizing space utilization within data center racks [4][6]. Group 2: SST Technology - SST (Solid State Transformer) is identified as the future power supply technology, capable of converting high voltages (10-34.5KV) to 800V DC, facilitating modular design and voltage adaptability across different regions [3][12]. - The implementation of SST technology is expected to reduce power loss significantly by increasing voltage levels, thus minimizing current and associated losses [15][19]. Group 3: Super Capacitor Benefits - The use of super capacitors can significantly reduce power supply volatility, with Delta's data showing a reduction in supply fluctuation from 73% to 6% [20][24]. - Different super capacitor technologies exhibit varying characteristics, with EDLC offering longer cycle life and LiC providing higher energy density, which can enhance the dynamic load capacity of GPUs in AI data centers [21][22]. Group 4: Future Roadmap - NVIDIA anticipates the gradual application of the Rubin and Feyman architectures between 2026 and 2028, which will further accelerate the adoption of Power rack solutions [11][12]. - Google's roadmap for data center power architecture evolution indicates a transition towards higher voltage levels and modular designs, aligning with the industry's shift towards SST solutions [13][15].