Nvidia unveils Rubin CPX GPU for million-token contexts

At its AI Infrastructure Summit on Tuesday, Nvidia announced a new GPU, the Rubin CPX, specifically engineered to handle context windows larger than 1 million tokens. The breakthrough marks a major step in enabling AI models to process and reason over unprecedented volumes of data.

Built for long-context AI

The Rubin CPX belongs to Nvidia’s forthcoming Rubin series of GPUs and is designed to power workloads that demand extended context comprehension. These include video generation, complex coding tasks, and other long-sequence applications where existing GPUs struggle.

Nvidia described the chip as a cornerstone of its “disaggregated inference” strategy, which separates AI processing tasks across specialized hardware to maximize performance and efficiency.

For end users, the Rubin CPX promises faster, more reliable results on long-context inference tasks, expanding what’s possible in both consumer and enterprise AI applications.

Nvidia’s growth engine

The announcement comes as Nvidia continues to dominate the AI hardware market. In its most recent quarter, the company reported $41.1 billion in data center revenue, driven largely by demand for GPUs powering large-scale model training and inference.

The Rubin CPX will not arrive immediately, however. Nvidia said the chip is scheduled for release at the end of 2026, giving developers and partners time to prepare for the next wave of AI infrastructure.

The bigger picture

By pushing the boundaries of long-context inference, Nvidia is positioning itself at the center of the next frontier in AI computing — where models are expected to handle massive context windows, multimodal inputs, and increasingly sophisticated reasoning tasks.

As the arms race in AI infrastructure accelerates, Rubin CPX could become the defining hardware platform for developers building the next generation of intelligent systems.