Usher in the New Era of AI Reasoning
Driven by the three scaling laws — pretraining scaling, post-training scaling, and test-time scaling — AI continues to advance rapidly. It is now entering a new phase marked by test-time scaling, also known as long thinking, where more compute is applied during inference to improve accuracy. This enables the rise of AI reasoning models, which mimic human thoughts to break down complex problems step by step before reaching an answer.
Designed for this era, the NVIDIA GB300 NVL72 is a liquid-cooled, rack-scale solution that integrates 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace™ CPUs, all interconnected with fifth-generation NVIDIA NVLink™ for fast, seamless communication. Delivering 1.5× more AI performance than the previous-generation GB200 NVL72, the GB300 NVL72 is the ultimate compute engine to power AI factories at scale and drive the next frontier of AI reasoning and generative workloads

Highlights
- 36 NVIDIA Grace™ CPUs
- 72 NVIDIA Blackwell Ultra GPUs
- CPU and GPU Connected by Fifth-generation NVIDIA NVLink™
- Integration with NVIDIA Quantum-X800 InfiniBand and Spectrum-X™ Ethernet
- Built for the Age of AI Reasoning
Rack-scale Architecture
-
Design Details






2 x Management Switches
Key Components of NVIDIA GB300 NVL72
NVIDIA GB300 Grace Blackwell Ultra Superchip
Connecting two NVIDIA Blackwell Ultra GPUs, one Grace™ CPU, and two ConnectX®-8 SuperNICs, the ultra-powerful superchip acts as the central core of the NVIDIA GB300 NVL72. With NVIDIA NVLink™ Switch technology and NVIDIA BlueField®-3 DPUs, up to 36 superchips are interconnected into one powerful unified GPU that redefines AI performance.


NVIDIA GB300 Compute Tray
Each GB300 compute tray combines two GB300 Grace Blackwell Ultra Superchips, which serve as a computing building block of the GB300 NVL72. Built to fuel the AI reasoning era, this tray powers the GB300 NVL72 with the scale, efficiency, and performance needed for tomorrow’s most demanding AI innovations.
The Next-Gen AI Data Center Solution

Liquid-to-Air Solution
- Support Superior Cooling Capacity up to 80kW
- All Heat Dissipation Removed by Fans
- Ideal to Upgrade Existing Air-Cooled Data Center

Liquid-to-Liquid Solution
- Provide Extreme Cooling Capacity up to 2500kW, supporting 10 or more AI Racks*
- Main Heat Dissipation through Facility Liquid
- Enable High-density Computing while Reducing Energy Consumption Significantly
*Depending on the power consumption of IT racks