Nvidia has announced the latest breakthrough in its Rubin CPX GPU, AI hardware.
Built to handle extremely demanding tasks like video generation and software development assistance, this new chip shows how Nvidia is shaping the future of artificial intelligence.
Unveiled on September 9, 2025, the Rubin CPX GPU is designed for large workloads and has become one of the most powerful AI processors ever created.
The chip is built on Nvidia's new Rubin architecture for workloads that require extremely long context windows. It can process up to 1 million tokens, equivalent to one hour of video.
This makes it especially valuable for advanced generation tasks that require consistent across large datasets, from long video generation to complex coding sessions.
Key Technical Highlights
- 30 PFLOPS of NVFP4 computing power per GPU
- 128 GB GDDR7 memory for efficient and cost-effective scaling
- Integrated video encoding/decoding support
- Rubin CPX is optimized to handle contexts while Rubin CPX handles contexts and standard Rubin GPU manages generation and decoding
Vera Rubin NVL144 CPX
The Rubin CPX GPU is not designed to run on its own. It is part of Nvidia's Vera Rubin NVL144 CPX platform.
- 144 Rubin CPX GPU
- 144 Rubin GPU
- 36 Vera CPU
According to NVIDIA, the setup will deliver 7.5x performance for previous generation Blackwell systems and set up a new benchmark for AI infrastructure.
Nvidia believes new chips will also be a wise investment. The company estimates that a $100 million setup can return as much as $5 billion in revenue from AI tasks such as training and running large models.
This reflects not only the efficiency of new architectures, but also the growing demand for AI models that process and generate at large scale.
availability
The chips and systems are expected to be on the market by the second half of 2026 once production is completed at TSMC. Nvidia will provide them as standalone GPUs and integrate them into full-rack scale deployments.
For industries relying on video, coding assistants, and large-scale generation AI, this innovation can redefine both performance and cost-effectiveness.
