NVIDIA Unveils Rubin CPX GPU with 128GB GDDR7, Set for AI and Enterprise Debut in Late 2026

NVIDIA has officially announced the Rubin CPX GPU, a powerhouse graphics processor equipped with an unprecedented 128GB of GDDR7 memory. Projected to launch at the end of 2026, Rubin CPX isn’t targeting gamers, but rather, it’s engineered for the ever-expanding world of artificial intelligence, machine learning, and advanced enterprise workloads.

Built for AI’s Next Frontier

At its core, the Rubin CPX adopts a monolithic die based on NVIDIA’s Rubin architecture. The GPU is capable of up to 30 petaFLOPs of compute performance using the new NVFP4 precision, which is vital for fast, long-context inference operations—a key requirement for large language models and generative AI. The inclusion of 128GB GDDR7 memory means Rubin CPX can hold vast contexts, making it ideal for million-token inference, long-form video processing, research, software development, and more.

Architecture: Rubin (monolithic die)
Compute Performance: Up to 30 petaFLOPs (NVFP4)
Memory: 128GB GDDR7
Media Engines: 4× NVENC + 4× NVDEC for heavy-duty video tasks
Primary Use: Long-context inference, large-scale AI research, and video analytics

NVIDIA positions the Rubin CPX not as a gaming GPU, but as a data center and AI solution for groundbreaking projects that need seamless performance across enormous streams of data.

A New Era for AI Infrastructure

The Rubin CPX makes its debut in the Vera Rubin NVL144 CPX rack platform, where each rack boasts a potent configuration: 144 Rubin CPX GPUs, 144 Rubin GPUs, and 36 Vera CPUs. This hardware ensemble delivers an incredible 8 exaFLOPS of computing power and an astonishing 100TB of rapid-access memory, all connected via NVIDIA Quantum-X800 InfiniBand and Spectrum-X Ethernet with SuperNICs.

The Rubin series also debuts NVLink Gen6, which doubles inter-GPU communication speeds, setting the stage for the near future of scalable AI infrastructure. NVIDIA has teased Rubin Ultra for 2027 (with HBM4E), and Feynman for 2028, pushing connectivity and performance even further.

What This Means for the Industry

With its extraordinary memory and computing capacity, the Rubin CPX will enable data scientists and enterprises to tackle projects that were previously limited by hardware constraints. From generative AI and language models to advanced video analytics and research, NVIDIA’s new platform sets a new bar for what’s possible in AI inference at scale.

Expect Rubin CPX shipments to begin in late 2026, powering the next wave of AI-driven transformations in data centers and research facilities worldwide.