AI Data Centers Use Lots of Copper. Why Watch Tin and Gallium?
Data centers use 60 to 75 tons of metal per megawatt of power. Yet by 2035 they may account for a far larger share of global gallium demand than of copper demand.

The 20-second read
1. Will AI favor copper most?
Picture an AI data center and you think of huge servers and miles of cable. Copper does run through nearly every point that touches power. It is natural to assume that more AI facilities mean more copper demand.
The World Economic Forum (WEF) estimates that a data center uses about 60 to 75 tons of metal per megawatt of power capacity. From the building to the power and cooling equipment, even the invisible digital economy needs a great deal of material.
But knowing a metal is used heavily does not tell you how much its market will be shaken.
2. A big lake and a small fish tank react differently
Pour the same bucket of water into a large lake and into a small fish tank. The lake's level barely moves. The tank's level changes a lot.
Metal markets work the same way. Copper is used in large volumes, but its global market is also large. Gallium, germanium and tin are relatively small markets, so new demand from data centers can show up as a bigger share.
Estimated share of global metal demand in 2035 from inside data centers
| Metal | Share |
|---|---|
| Gallium | 6% |
| Germanium | 2.4% |
| Tin | 1.2% |
| Copper | Under 0.9% |
Note: Figures cover demand inside data center facilities only. Demand from external grid expansion is excluded. The copper figure is an upper bound, not an exact value.
That gap is the point. This is not a claim that copper is irrelevant. The question is how much the new demand from AI changes each metal's existing market.
3. Does that mean copper has little to do with AI?
Many readers jump to the opposite conclusion: with a small share of demand inside data centers, copper must matter little for AI.
The figures above leave something out: the power grid outside the data center. For AI facilities to use electricity, transmission lines, substations and generation capacity are needed. Copper is an important metal there too.
So AI-driven demand for raw materials should be read in three parts.
- 1. Inside the facility (inside the data center): Metal demand from servers, electrical equipment and cooling systems.
- 2. Outside the facility (grid expansion): Added demand from building more transmission lines, substations and generation. This is where copper's importance rises again.
- 3. Supply side (production that is hard to expand): Constraints from concentrated producing countries, export controls and the time needed to develop new mines.
This is why the inside-the-facility numbers alone cannot show how much copper benefits from AI.
4. Supply makes the final difference
If production can be raised easily, the market shock from higher demand may be small. If production is concentrated in a few countries and tangled in export controls, even a small change in demand can make the supply chain sensitive.
Gallium and germanium are metals where China holds large influence over supply. Tin also belongs to a market smaller than copper's. When looking at AI raw materials, it pays to examine not only what these metals are used for but also who supplies them and how fast output can rise.
High market sensitivity does not mean a given metal will deliver higher investment returns. Understanding the structure of supply and demand is different from forecasting prices.
- It may rise. But the global copper market is large, so data centers' share of demand inside the facilities is estimated to be small.
- Data centers' in-facility share of global copper demand in 2035 is estimated at under 0.9%.
- That is hard to say. This estimate leaves out demand from expanding the external power grid.
- Copper is also used heavily in transmission lines, substations and power generation equipment.
- Their markets are small, so data center demand can account for a larger share.
- The estimated in-facility shares of global demand in 2035 are 6% for gallium, 2.4% for germanium and 1.2% for tin.
- The World Economic Forum (WEF) estimated that about 60 to 75 tons of metal are used per 1MW of power capacity.
- That figure includes the building, power equipment and cooling equipment.
- You can't assume that. A market being sensitive is not the same as returns being high.
- Understanding the structure of supply and demand is a separate matter from predicting prices.
What to watch
Insight Times Editorial Desk





































