Nvidia aims to occupy every chip in AI data centers
NVIDIA
Intel Corporation
OpenAI
Nvidia is promoting its new Vera Rubin chip system, which combines GPUs and CPUs, positioning itself as a supplier of complete AI systems. The system promises 10x tokens per watt compared to Grace Blackwell and features liquid cooling, cable-free design, and faster installation. Early customers include Microsoft, OpenAI, and Oracle.
Nvidia is intensifying its push into the AI data center market with the Vera Rubin chip system, a successor to Grace Blackwell that combines one Vera CPU for every two Rubin GPUs. The company emphasizes that the system delivers 10 times more tokens per watt and nearly three times memory bandwidth compared to its predecessor. Nvidia claims the Vera CPU outperforms rival CPUs from AMD and Intel on agentic AI tasks, although benchmarks used slightly older competitor chips. The Vera Rubin NVL72 racks are designed for plug-and-play installation, reducing cable usage and setup time from hours to minutes, and are 100% liquid-cooled. Nvidia states that OpenAI already has a Vera Rubin rack in use. The system is scheduled to ship in the second half of 2026, with early customers including Microsoft, OpenAI, and Oracle. The promotion comes ahead of AMD's annual conference, where AMD is expected to detail its competing Helios AI chip rack.
- Сокращения
- GPU = Graphics Processing Unit — графический процессор
- CPU = Central Processing Unit — центральный процессор
- ARM = Advanced RISC Machines — архитектура ARM
Source: Wired AI —
original