Nvidia revives Rubin CPX chip with major redesign

The revival of the Rubin CPX marks a strategic pivot for Nvidia, coming after the company appeared to remove the chip from its roadmap at GTC 2026, where it instead highlighted Groq 3 LPUs and LPX racks. Kuo's industry checks suggest that Nvidia has not abandoned the prefill acceleration niche, but is repositioning it with a stronger architecture. The shift to standalone MGX ETL racks and the larger memory footprint indicate a focus on handling the growing demands of AI inference, where prefill performance is critical. By recommending a 1:1 ratio of CPX to Rubin GPUs, Nvidia is positioning the two chips as complementary rather than competing, with CPX handling the compute-heavy prefill stage and Rubin handling the decode stage. This division of labor could optimize overall system efficiency for customers deploying large-scale AI models. The fact that Nvidia did not immediately respond to requests for comment leaves room for further clarification, but Kuo's track record as a supply-chain analyst lends weight to his claims.