Nvidia Adjusts AI Chip Roadmap at GTC 2026: Rubin CPX Project Shelved, Focus Shifts to Groq LPU/LPX

At the 2026 GTC conference, Nvidia unveiled its latest AI chip roadmap, where the originally planned Rubin CPX was not included in the Rubin series. Later, company representatives confirmed that the Rubin CPX project has been shelved, and current efforts are focused on Groq LPU/LPX.

Both Rubin CPX and Groq LPU are AI inference accelerators for the Vera Rubin system, but the former focuses on the pre-filling phase (context and prompt processing), while the latter concentrates on the decoding phase (output generation). Since the AI industry currently values inference output efficiency more, Groq LPU is more favored. Additionally, given Nvidia's "two-year generation change, one-year update" rhythm, introducing two new chips in the same generation is quite challenging, which is another reason for the adjustment.