SemiAnalysis claims NVIDIA Rubin beats GB300 NVL72 on vLLM inference economics

SemiAnalysis says NVIDIA's Rubin delivers 3.2x better profit per gigawatt and up to 10x better performance per dollar than GB300 NVL72 when running the vLLM production LLM engine.

The claim is presented as the first part of a three-part thread that SemiAnalysis says will explain the results; the post's own evidence so far is its headline statement, and the underlying methodology and test conditions have not yet been published.

The figures are SemiAnalysis's claims, not independently verified results. Because the explanation is still forthcoming, readers should treat the 3.2x and 10x numbers as unexplained until the follow-up parts appear.