The server room is just the appetizer
Everybody loves the AI story when it’s about chips, servers, and shiny new model launches. But Wood Mackenzie says that’s like looking at a concert and only counting the speakers. The bigger bill is the whole circus around it: grid upgrades, transmission lines, substations, on-site power, cooling, and all the heavy metal that makes the thing actually run.
And yes, that means aluminum and copper. A lot of it.
The hidden infrastructure tax
The report argues that once you add in the power layer, data centers can pull in 3x to 4x more aluminum and copper than the building itself. That’s not a rounding error. That’s the difference between “interesting theme” and “holy smokes, where do we get the metal?”
A few nuggets that matter:
- Internal data center aluminum use is heavily tied to cooling, racking, and enclosures
- Copper is essential for electrical systems, high-density compute nodes, substations, and underground connections
- Annual data center-driven power additions could jump from 15–20 GW today to 30–33 GW in the early 2030s
Meta is acting like a utility now
The piece uses Meta as the poster child for this new reality. Instead of relying on the old “plug it into the grid and pray” playbook, hyperscalers are building their own power ecosystems — solar, wind, batteries, gas turbines, fuel cells, and even small modular reactor talk.
That’s a very fancy way of saying: the AI race is now part software sprint, part industrial buildout. And industrial buildouts eat metal for breakfast.
Why investors should care
If Wood Mackenzie is right, the winners may not just be the obvious AI names. The ripple effects hit:
- copper miners and copper ETFs
- aluminum producers
- transmission and utility infrastructure suppliers
- grid equipment makers
Big picture: the AI trade may be wearing a hard hat now. And the market might still be underpricing just how much copper, aluminum, and power plumbing it takes to keep the robots caffeinated.