Nvidia Rubin NVL72 Rack Could Cost Around $8M

The cost of building next-generation AI infrastructure is climbing fast.
Nvidia’s upcoming Rubin NVL72 system could reportedly cost around $8 million per rack.
That would represent roughly a 17% increase versus comparable current-generation systems as higher memory, networking and component costs work through the supply chain.
At hyperscale, the impact becomes enormous.
For a 1 GW AI data center, the higher hardware pricing could add at least $5 billion to the cost of the Nvidia compute alone.
Cloud providers are unlikely to absorb all of that themselves.
Some of the increase will likely flow through into higher prices for AI training and inference.
The AI boom is still expanding.
But the economics are getting more expensive at every layer:
Chips.
Memory.
Networking.
Power.
And now the full rack itself.