Nvidia wants to own every chip inside AI data centers
Nvidia’s Vera Rubin system signals a shift toward fully integrated AI data-center hardware, promising tighter optimization across CPUs and GPUs and faster, more efficient AI workloads.
At a glance
- Vera Rubin is Nvidia’s next-generation system, described as a one-CPU-for-every-two-GPUs configuration.
- In an NVL72 rack, Nvidia claims 36 Vera CPUs are paired with 72 Rubin GPUs, marketed as “cable-free compute” and easily hot-swappable for rapid deployment.
- Nvidia asserts Vera Rubin delivers up to 10x the tokens-per-watt vs. its Grace Blackwell system and offers nearly three times the memory bandwidth with its localized memory subsystems.
- The company also notes the Vera Rubin NVL72 system is 100% liquid-cooled and emphasizes reduced cabling to speed up rack installation.
The story
Nvidia is pushing deeper into AI infrastructure with Vera Rubin, a hardware system designed to pair CPUs with GPUs in a tightly integrated package for data-center workloads. The Vera Rubin NVL72 configuration is described as a one CPU for every two GPUs setup, with a 36‑CPU to 72‑GPU ratio per rack.
Nvidia executives showcased NVL72 racks as being more plug-and-play than earlier offerings, and the company highlighted that OpenAI already operates a Vera Rubin rack in use at a data-center facility. The announcements were made during a workshop in Santa Clara as part of a broader push into complete AI systems rather than chips alone.
Nvidia claims that Vera Rubin delivers an order of magnitude more tokens-per-watt than its Grace Blackwell system and features localized memory subsystems that provide nearly three times the memory bandwidth of Blackwell. The company also touts a cable-free design and hot-swappability to simplify installation and maintenance, along with 100% liquid cooling to reduce energy use.
While Nvidia has repeatedly said Vera Rubin is ramping to full production and will ship on schedule, the workshop did not provide a precise delivery date, instead emphasizing strategic positioning ahead of industry events.