NVIDIA GB200 NVL72
An entire rack of 72 GPUs wired together with NVLink to act like one giant GPU.
GPU memory
13,400 GB
GPU memory is the "workspace" a graphics card uses to hold an AI model while it runs. If a model is bigger than the available memory, it simply will not fit — like trying to fit a large book on a shelf that's too small. More in the wiki →
72 GPUs × None GB each.
Power draw
120,000 W
Watts measure how much electricity something draws while it's running. More powerful hardware needs more electricity and produces more heat that has to be cooled. More in the wiki →
Ideal cooling
Liquid-cooled (required)
Direct-to-chip liquid cooling fed by a facility coolant loop - air physically can't move this much heat out of one cabinet. Running flat out, it turns 120,000 W of electricity into heat that has to go somewhere. We stock matching cooling gear →
Price
$3,000,000
This is a current typical market price, not a fixed list price — real prices move with supply and demand. NVIDIA does not publish a rack price. $2–3 million is the range most consistently repeated by industry press; we use $3,000,000.
CPU
36x NVIDIA Grace CPUs (one per Grace Blackwell Superchip)
The CPU is the general-purpose "manager" chip. It runs the operating system, moves data around, and hands the heavy AI math off to the GPUs. More in the wiki →
What this means in everyday terms
Running one of these draws as much power as about 100.0 average homes running around the clock. A full day of running one uses about 32.00 electric-car batteries worth of energy. An average home draws roughly 1,200 watts around the clock. This compares a build's power draw to that baseline. More in the wiki →
NVIDIA GB200 NVL72 is sold to run inside a server or rack — it isn't something you plug into a home PC.
Where these numbers come from
Official NVIDIA GB200 NVL72 product page (72 GPUs, 13.4TB total HBM3e memory, ~120kW rack power). Price per Tom's Hardware / TechSpot industry reporting.