AI news

1. NVIDIA Unveils "Rubin" R100 Architecture: The 1nm AI Milestone

Date: July 28, 2026

NVIDIA has officially announced the "Rubin" R100 GPU architecture, marking the industry's transition to 1nm--class process technology (via TSMC). Designed to power the next generation of AGI--scale models, the Rubin platform introduces HBM4 (High Bandwidth Memory 4) and a revolutionary "Optical NVLink" interconnect, delivering a 4x performance--per--watt improvement over the Blackwell architecture.

Metric Blackwell (B200) Rubin (R100)
Process Node 4nm (Custom) 1nm--class
Memory Type HBM3e HBM4
Interconnect Speed 1.8 TB/s 4.5 TB/s
Efficiency 1x 4x

This launch is viewed by analysts as a strategic move to secure NVIDIA's dominance in the "Reasoning AI" era, where hardware--level optimization for latent logic processing is critical.

Sources: NVIDIA Newsroom, Bloomberg Technology