Overview
- The rack-scale system integrates 72 Rubin GPUs and 36 Vera CPUs connected by NVLink 6 for about 260 TB/s of rack bandwidth, using 18 modular compute trays for fast serviceability.
- Nvidia says Vera Rubin delivers roughly 10 times the performance per watt versus Grace Blackwell and is the company’s first fully liquid‑cooled system, which it says reduces data center water use.
- The company reports the platform is in full production and expects customer deliveries in the second half of 2026, noting the new racks weigh nearly two tons and carry about 1,300 microchips.
- Built from roughly 1.3 million components sourced from more than 80 suppliers in at least 20 countries, the system faces pressure from soaring memory costs linked to a global shortage.
- Meta plans to deploy Vera Rubin by 2027, and Nvidia lists expected customers including OpenAI, Anthropic, AWS, Google, Microsoft and Oracle, while rivals such as AMD prepare competing rack-scale systems.