图 点击 上方 硬AI 关注我们 当地时间5月21日，纳微半导体公司宣布与英伟达合作开发下一代800V高压直流(HVDC)架构，为包括Rubin Ultra在内的 GPU提供支持的"Kyber"机架级系统供电。Navitas的氮化镓和碳化硅技术将在这一合作中发挥关键作用。周三美股盘 后，纳微股价一度暴涨200%。 硬·AI 当地时间5月21日，纳微半导体公司宣布与英伟达合作开发下一代800V高压直流(HVDC)架构，为包括 Rubin Ultra在内的GPU提供支持的"Kyber"机架级系统供电。 纳微的氮化镓和碳化硅技术将在这一合作中 发挥关键作用 。 有媒体指出，这代表了数据中心基础设施领域的重大技术飞跃，特别是在支撑吉瓦级AI计算负载方面， 纳 微的技术将一定程度上大幅提升能效并降低铜材使用量 。周三美股盘后，纳微股价一度暴涨200%。 01 催A生I时革代命的性电解力决需方求案 现代AI数据中心需要吉瓦(GW)级的电力来满足不断增长的AI计算需求。 作者 |鲍奕龙 编辑 | 硬 AI 英伟达正与纳微半导体(Navitas)展开战略合作，共同开发800V电力架构，有望彻底改变AI数据中心的电 力供应系统 ...

Core Insights - Nvidia is collaborating with Navitas Semiconductor to develop an 800V power architecture aimed at transforming the power supply systems for AI data centers [1][2] - The partnership leverages Navitas's GaN and SiC technologies, which are expected to significantly enhance energy efficiency and reduce copper usage in data center infrastructure [1] - The current data center architecture, which operates on a traditional 54V system, is reaching its physical limits as power demands increase beyond 200kW [1] Group 1 - Nvidia's solution involves converting 13.8kV AC grid power directly to 800V high-voltage DC, utilizing solid-state transformers and industrial-grade rectifiers to maximize efficiency and reliability [2] - The implementation of 800V HVDC allows for a reduction in copper wire thickness by up to 45%, addressing sustainability concerns for next-generation AI data centers that require gigawatt-level power [2] - Traditional 54V DC systems require over 200kg of copper to supply power to a 1MW rack, which is unsustainable for the increasing power needs of modern AI applications [2]