Nvidia launches Rubin CPX chip for AI and video.
The company unveils the Rubin CPX, a chip that promises to revolutionize AI tasks such as video creation and software development, with a planned launch in 2026.
247 - Nvidia Corp., the tech giant known for its central role in developing chips for artificial intelligence (AI), has announced the launch of a new product, the Rubin CPX. The information was released in a company statement and it will be launched in late 2026.
This new system aims to optimize demanding tasks such as video generation and software creation. Rubin CPX will be available as boards that can be integrated into existing servers or used in discrete computers operating autonomously in data centers.
Nvidia, known for its rapid pace in launching technological innovations, believes that the Rubin CPX is crucial for improving the efficiency of certain AI processes. The new chip is an evolution of the Rubin line, scheduled for next year, and comes with a different approach than conventional graphics chip solutions, or GPUs.
Currently, a single GPU handles both understanding inputs and generating responses. With the new Rubin CPX architecture, this process will be divided, allowing for greater efficiency in handling large volumes of data. The company states that by investing US$100 million in hardware with the new chip, customers can generate up to US$5 billion in revenue, a metric that, according to CEO Jensen Huang, has gained prominence as the industry seeks to quantify the return on investment in new technologies.
Huang also highlighted that CPX is the first chip developed specifically to handle AI models capable of performing complex reasoning with large amounts of data, known as tokens. Regarding software creation, the new system will allow platforms to evolve from simple code suggestions to the understanding and development of large-scale software projects.
In the field of video generation, the Rubin CPX promises to be an advanced solution, capable of performing decoding, encoding, and processing on a single chip, simplifying processes that currently require multiple components.
(With information from Bloomberg)


