NVIDIA announces Vera Rubin supercomputing platform (Rubin GPU + Vera CPU + liquid cooling) delivering 7+ exaflops AI and 5 PF FP64 per system, 144 GPU/rack density, shipping Q4 2026.

NVIDIA has announced the Vera Rubin supercomputing platform, a new system designed for scientific computing that combines a custom Rubin GPU, Vera CPU, and liquid cooling technology. The platform is scheduled to ship in Q4 2026.

The Vera Rubin system delivers 7+ exaflops of AI performance and 5 petaflops of FP64 performance per system, with a density of 144 GPUs per rack. This configuration represents NVIDIA's push toward denser, more integrated compute solutions for high-performance computing workloads.

The platform's specifications position it as a significant addition to NVIDIA's infrastructure offerings, particularly for scientific computing applications requiring both massive AI throughput and traditional HPC floating-point performance. The Q4 2026 timeline aligns with the broader acceleration of GPU deployment across data centers supporting AI model training and inference at scale.