NVIDIA Reveals Vera Rubin The 1.3 Million-Part Supercomputer Sourced from 20 Countries.

 

Inside NVIDIA 1.3 Million-Part Puzzle: A First Look at the Vera Rubin Prototype

In an exclusive segment with CNBC, NVIDIA opened its doors to showcase the prototype of its next-generation AI powerhouse: the Vera Rubin chip architecture. The tour offered a rare glimpse into the staggering complexity of the NVIDIA DGX Vera Rubin NVL72 server cabinet, highlighting a supply chain that spans the globe.

A Global Masterpiece of Engineering

NVIDIA revealed that a single DGX Vera Rubin system is comprised of over 1.3 million individual components. These parts are sourced from specialized factories across 20 different countries.

• Thermal Management: The sophisticated liquid-cooling system is provided by Delta Electronics.

• Redundancy and Sourcing: To ensure a steady supply, NVIDIA utilizes multiple vendors for critical parts; for instance, there are ten different manufacturers for a single type of chip cooling plate.

• Final Assembly: While components come from around the world, the final integration of the massive cabinets is handled by Foxconn.

Strategic Transition: Blackwell Meets Rubin

Although the Vera Rubin systems are slated for delivery in the second half of 2026, NVIDIA continues to ship several thousand Blackwell units per week. The company expects customers to run both architectures side-by-side for the foreseeable future, delegating different AI workloads to each system based on their specific performance profiles.

Distributing components to 20 countries isn't just for show; it's a "China Plus One" strategy, mitigating over-reliance on any single country. If a trade war or natural disaster strikes a particular area, NVIDIA's supply chain will remain operational.

The collaboration with Delta Electronics (a Taiwanese giant) reflects the world's advancement beyond fan-based cooling. The Vera Rubin servers consume enormous amounts of power, requiring thermal throttling to maintain peak performance.

The DGX cabinets seen by reporters are reference designs. NVIDIA acts as the architect, providing a structural model. Manufacturers like Dell, HP, or Supermicro then modify the internal components according to their budgets, while retaining the core NVIDIA chip.

It's anticipated that Blackwell will remain the primary component for inference processing, while Vera Rubin is designed for training massive Long-Term Infrastructure Models (LLMs) requiring significantly more memory bandwidth.

 

 

The Galaxy S26 series smartphones come with Buds 4 wireless earbuds that include three AI assistants. 

 

Source: CNBC