GTC大会前瞻：重点关注哪些领域？

Core Viewpoint - The article discusses the upcoming advancements in AI computing infrastructure, particularly focusing on the developments expected at the NVIDIA GTC conference, including new GPU architectures and power supply innovations [5][21]. Power Supply Architecture - The competition in AI computing power is fundamentally about power density, with significant advancements expected in power supply and cooling technologies. NVIDIA's Rubin chip is projected to exceed 2000W, while the upcoming "Feiman" chip aims for over 5000W, necessitating a shift to 800V high-voltage power systems [9][10]. - The transition to 800V high-voltage direct current (HDVC) is expected to reduce current and alleviate power loss issues, with NVIDIA's Rubin Ultra generation likely adopting this technology [9][10]. - Modular power supply solutions are anticipated to enhance current capacity and reduce PCB layout space, with significant implications for manufacturing complexity and costs [10][11]. - The value of inductors in power supply components is expected to increase significantly, from approximately 3 yuan in traditional designs to 8-10 yuan in modular designs, with potential for further increases due to technological advancements [10][11]. AI Computing and Data Transmission - The demand for higher bandwidth, lower latency, and reduced power consumption in AI computing is driving the evolution of optical communication technologies from traditional modules to Co-Packaged Optics (CPO) and Near-Packaged Optics (NPO) [14][15]. - The GTC conference is seen as a critical point for the commercialization of CPO technology, with expectations for significant production increases by 2027 [14][15]. - The integration of optical engines with network cards in the Rubin Ultra design is expected to lead to a surge in demand for optical components, enhancing the overall market for optical communication [15]. Liquid Cooling Technology - As GPU power levels rise above 2000W, liquid cooling is transitioning from an optional feature to a mandatory requirement, with projections indicating that liquid cooling will dominate the market by 2026 [16][17]. - The deployment of the "Feiman" chip is expected to significantly increase the demand for liquid cooling components, with a focus on upgrading cooling materials and thermal interface materials (TIM) [17][19]. - Innovations in cooling materials, such as diamond-based thermal conductors and liquid metal TIMs, are anticipated to enhance cooling efficiency and reduce total cost of ownership (TCO) for GPU servers [19][20]. Industry Supply Chain Developments - Domestic companies are making significant strides in integrating into the supply chains of major players like NVIDIA, Amazon, and Google, with several firms already validated for key components [12][13]. - The article highlights the importance of companies with core technologies and established supply chain relationships as key investment targets in the evolving AI infrastructure landscape [21].