NVIDIA VR200 NVL144 · 2H 2026 PLATFORM

From sand grain
to 3.6 exaFLOPS in a rack.

A Vera Rubin NVL72 rack contains 72 Rubin GPUs, 36 Vera CPUs, 576 GB of HBM4 per accelerator, and roughly 1.2 million components — sourced from 30+ countries and integrated through eight manufacturing stages before it arrives at a hyperscaler data center.

8
Supply Stages
~6 mo
Wafer → Rack
~2 t
Per Rack
120 kW
Power Draw

The 8-Stage Supply Chain

Each stage represents one step in NVIDIA's manufacturing pipeline.

CLICK ANY NUMBERED CIRCLE BELOW TO EXPAND

Who builds what

25+ companies across 30+ countries. Color stripe on each card maps to the stage they own.

USE THE PILLS BELOW TO FILTER BY STAGE

Building a Rack, Live

Watch all 18 trays — power, compute, NVSwitch, networking — install into a single VR200 NVL72 chassis.

PRESS PLAY TO START THE BUILD
VR200 NVL72 · 44U Oberon
Live Build
Press Play to begin assembly.

From Foxconn floor to AI factory

The final step: completed racks travel from integration sites to the data centers running them. Below are the typical routes and the customers receiving Vera Rubin.

✈ Air freight ~$50K/rack 🛃 Customs 7–14 days 🏭 Final L12 deploy
10 GW
OpenAI commitment
100k+
GPUs · Microsoft Fairwater
~1.5M
Rubin GPUs · 2026 target
8 racks
Per SuperPOD