Core Insights - The 2025 Asia Data Center Summit and Exhibition in Hong Kong highlighted the importance of liquid cooling technology, with Shuguang Data's innovative solutions gaining significant attention [1][2] - The demand for computing power is increasing due to advancements in AI models and high-density computing, with global data center installed capacity reaching 46 GW in 2023 and expected to grow to 87 GW by 2027 [1][2] - Traditional air cooling methods are facing challenges in meeting the performance, energy efficiency, and reliability needs of future data centers, leading to a shift towards liquid cooling solutions [1][2] Industry Trends - Liquid cooling technology offers superior thermal conductivity compared to air cooling, enabling greater computing power deployment in the same space while significantly reducing energy consumption of auxiliary equipment [2] - The value of liquid cooling is evolving from merely efficient heat dissipation to becoming a strategic support for overall data center operations [2] - Despite its advantages, the implementation of liquid cooling faces challenges such as server selection mismatches, lack of industry standards, and conflicts with traditional HVAC concepts [2] Company Developments - Shuguang Data has established a comprehensive one-stop delivery system that covers the entire lifecycle of liquid cooling data centers, addressing challenges from planning and design to operation and maintenance [2] - The company has developed the SLiquid intelligent operation and maintenance platform, transitioning from manual operations to intelligent monitoring, enhancing system stability and management efficiency [2] - With over a decade of experience in liquid cooling technology, Shuguang Data has positioned itself as a leader in the field, moving from concept validation to large-scale application [3]

英伟达CEO黄仁勋近期动作频繁，试图在技术管制与市场拓展间寻找平衡。5月19日，他在 Computex2025宣布 联合台积电、富士康在中国台湾建立AI超级计算机 ，并推出个人AI计算机 DGXSpark，强调"AI工厂"时代的到来。此前，他于4月访华期间重申中国市场的重要性，并透露英伟达 正扩建上海研发中心，计划引入2000名工程师，聚焦定制化芯片设计与自动驾驶技术开发。 Blackwell架构的技术革新集中体现在 B300芯片 上。作为Blackwell Ultra系列的核心，B300采用台积电 4NPT SMC工艺制造，通过10TB/s的片间互联技术将GPU裸片连接成统一计算单元，实现了1.8TB/s的双 向NVLink带宽，支持多达576块GPU的无缝协作。其FP4性能较前代提升1.5倍，配合第二代Transformer 引擎，可在4位浮点精度下实现算力与模型规模的翻倍。然而，为满足美国出口管制要求，Blackwell特 规芯片将 放弃HBM高带宽内存，改用GDDR7 ，这一调整虽降低了性能上限（预计较原版Blackwell下 降约30%），但仍显著优于被禁的H20。 对比H20的96GB HBM3内 ...