Nvidia GB200 NVL72

Nvidia GB200 NVL72 is the rack-scale [[NvidiaBlackwellPlatform|Blackwell]] system used as the main incumbent comparison in 国产 AI 算力能凭「超节点」弯道超车吗？｜WAIC 深度观察 S10E23. The source describes it as 72 B200 accelerators connected in one cabinet-scale domain, with NVSwitch trays and high rack power.

The existing E230｜1万亿收入预期背后：英伟达的巅峰与软肋 source treats NVL72 as part of Nvidia’s platform shift from chip sales toward full AI infrastructure. The WAIC source adds the China comparison: Huawei CM384 can exceed NVL72 in cited aggregate compute by using more chips, but that does not erase Nvidia’s efficiency, CUDA, reference architecture, and ecosystem advantages.

Connections

• Nvidia, Nvidia Blackwell Platform, and CUDA — company, platform, and software ecosystem.
• AI Accelerator Supernode and Scale Up AI Interconnect — system category and interconnect logic.
• Huawei CM384 — domestic comparison case.
• Token per Watt, Data Center Power Bottleneck, and AI Infrastructure Full-Stack Moat — efficiency and deployment constraints.