NVIDIA announced the Rubin CPX GPU for million-token coding and generative video apps. The Vera Rubin NVL144 CPX platform offers 8 exaflops of AI performance and $5B in token revenue for every $100M invested. Cursor, Runway, and Magic are exploring Rubin CPX for AI acceleration.

Rubin CPX is a new GPU class for massive-context processing, working with Vera CPUs and Rubin GPUs. Vera Rubin NVL144 CPX platform provides 8 exaflops of AI compute and 100TB of fast memory in a single rack. Rubin CPX enables unprecedented AI performance and revenue generation for companies.

Built on the Rubin architecture, Rubin CPX offers up to 30 petaflops of compute with NVFP4 precision and 128GB of GDDR7 memory. It delivers 3x faster attention capabilities compared to previous systems. Rubin CPX can be integrated with NVIDIA InfiniBand or Ethernet networking platforms for scalability.

AI innovators like Cursor, Runway, and Magic are leveraging Rubin CPX for software development, video generation, and AI agent automation. Cursor sees improved developer productivity, while Runway aims to enhance visual effects creation. Magic is developing foundation models for AI agents with large context windows.

NVIDIA Rubin CPX will be supported by the complete NVIDIA AI stack, including the Dynamo platform for AI inference scaling. Nemotron multimodal models and NVIDIA AI Enterprise software will also be compatible with Rubin CPX. The platform extends NVIDIA’s developer ecosystem with CUDA-X libraries and a large community of developers.

Availability of NVIDIA Rubin CPX is expected by the end of 2026. Learn more about the Rubin CPX GPU by watching NVIDIA’s Vice President of Hyperscale and High-Performance Computing Ian Buck’s keynote at the AI Infra Summit on Sept. 9 at 10am PT.

Read more at Nvidia: NVIDIA Unveils Rubin CPX: A New Class of GPU Designed for Massive-Context Inference