AI’s Energy Problem Is Growing

Data-center operator NextDC reported that its Power Usage Effectiveness increased from 1.44 to 1.49.

That may look like a small change, but with PUE, higher is worse. It means a larger share of electricity is being consumed by cooling and other infrastructure rather than the computing equipment itself.

Water efficiency also deteriorated.

And this isn’t just a one-year fluctuation. Both metrics have now worsened for three consecutive years.

That becomes interesting when you look at the broader AI story.

We constantly hear about more powerful chips, larger models, and enormous new data centers. But all of that computing power has to connect to the physical world somehow.

It needs electricity.

A lot of it.

And supplying that electricity means more substations, transformers, transmission lines, switchgear, cooling systems, and backup infrastructure.

This is where the AI boom starts becoming as much a commodities story as a technology story.

Copper is the most obvious beneficiary because of the enormous amount of electrical infrastructure involved. But silver also plays an important role across electrical and electronic supply chains thanks to its exceptional conductivity.

The interesting part is that efficiency improvements were supposed to offset some of the massive growth in computing demand.

NextDC’s numbers are a reminder that real-world data centers may not be becoming more resource-efficient as quickly as expected, especially while they are rapidly expanding capacity.

One company doesn’t represent the entire global industry.

But if demand for AI computing continues growing at anything close to the current pace, while power and cooling requirements remain this high, the amount of physical infrastructure required could be enormous.

And that means the AI investment cycle could quietly become another long-term source of demand for copper, silver, and other electrical metals.