Nvidia just reported $96.2B in quarterly revenue, with Data Center revenue reaching $89B, up 117% year over year.
Hold up.
The next constraint may not be the GPU. It may be the infrastructure required to keep those GPUs running.
The IEA estimates global data-center electricity consumption could nearly double from 485 TWh in 2025 to 950 TWh by 2030, while electricity use from AI-focused data centers could triple.
That changes how I look at the AI trade.
The opportunity isn’t necessarily another company making faster chips. It could be the businesses supplying the physical layer underneath them: transformers, grid equipment, power generation, cooling, storage and data-center infrastructure.
But there’s a catch.
Projected demand doesn’t guarantee every planned data center gets built. Financing, permitting, grid connections and supply constraints can slow the entire buildout. And AI efficiency is improving quickly.
So I’m not saying power replaces chips as the AI trade.
I’m saying the bottleneck can move.
When AI demand meets the physical limits of the grid, the question becomes much more interesting:
Who owns the infrastructure everyone suddenly needs?