Artem Voloskovets
| Platform | Nvidia revenue opportunity per GW |
| Hopper | ~$18B |
| Grace Blackwell | ~$25B |
| Vera Rubin | $40B+ |
That is a 122% increase from Hopper to Rubin, showing how rapidly the value of hardware installed behind each gigawatt of power is rising.
The number also changes how Nvidia’s growth opportunity should be viewed. The constraint is increasingly not GPU demand, but power availability and the capital required to build gigawatt-scale AI infrastructure.
At Nvidia’s stated Rubin economics, 5 GW represents more than $200 billion of potential Nvidia revenue opportunity, while 10 GW represents more than $400 billion. These are not revenue forecasts, but they illustrate the scale of spending associated with large AI clusters.
Rubin increases Nvidia’s revenue density partly because the company now captures more of the AI system. The platform extends beyond GPUs to CPUs, NVLink interconnects, networking and other infrastructure.
For hyperscalers, the same trend has the opposite implication: Nvidia’s revenue is their capex. As spending per gigawatt rises, Microsoft, Amazon, Alphabet, Meta and other AI infrastructure buyers need increasingly large returns from cloud services, inference and AI products to justify new capacity.
The key bottleneck is therefore shifting toward electricity. GPUs can be manufactured faster than new power generation, transmission and grid connections can be built.
Huang’s $40B+ figure captures both sides of the AI infrastructure boom: Nvidia can extract substantially more revenue from every available gigawatt, while the cost of competing at the AI frontier continues to rise.
Artem Voloskovets
