The AI Metals Trade: What the Bullish Case Leaves Out
July 8, 2026
You have probably read the pitch by now, because almost everyone is making it.
AI needs data centres. Data centres need power, cooling, wiring, and chips. All of that needs metal, copper most of all, silver close behind. Supply is tight, mines take years, a deficit is coming. So buy the metal, buy the miners, buy the ETF.
I am not going to tell you that story is wrong.
What I want to do is point out what that story leaves out.
One. AI is a much smaller slice of the demand than the headline makes it sound.
The scary numbers you see are usually about all the new demand for a metal at once, with AI quietly standing in for the whole thing. Pull the figures apart, though, and AI is the small part.
The real engines of copper demand are electrification more broadly, electric vehicles, grids, and transmission, and AI rides along in the framing because it is the exciting word.
Even J.P. Morgan, in the middle of a bullish note, has pointed out that data-centre copper demand is growing from a small base. Silver tells the same story. The Silver Institute and Metals Focus only recently began tracking data-centre silver as its own category, with early estimates somewhere around fifteen to twenty million ounces a year, a thin sliver of an industrial market that runs to hundreds of millions of ounces, and one where solar, not servers, is the bigger pull. A lot of what gets sold as "AI metals" is really just electrification wearing a more fashionable costume.
Two. The demand forecasts themselves are shakier than the people quoting them admit.
The metals world treats the big power-demand projections as settled fact. The people who actually sell that power do not. The chief executive of Constellation Energy, a company whose business is supplying electricity to these data centres, has said publicly that the projected load is being overstated and that the industry should slow down.
Utilities have flagged a phantom-demand problem, where the same data-centre project gets counted over and over as developers shop it around to different power providers, inflating the totals. Fortune and the Associated Press ran a piece arguing the clearest sign of an AI bubble might be hiding in those very electricity forecasts. If the demand at the top of the funnel is padded, then so is the deficit math built on top of it. Yet I rarely see a mining write-up so much as mention that the foundation is contested.
Three. You can be completely right about the thesis and still get hurt in the stocks.
This is the one that costs people real money. Copper spent the last stretch hitting record prices, and plenty of the miners still had a brutal time. Freeport-McMoRan cut its sales guidance just days after the stock touched an all-time high, after operating setbacks at its Grasberg mine. Disruptions, accidents, rising costs, and write-downs have landed on producers even while the metal itself climbed. Some market watchers go further and argue the deficit has been promoted so effectively that traders front-ran it and priced the shortage in already. The lesson is uncomfortable but old: being right about copper and making money in copper stocks are two different sentences. A mine is an operating business, full of weather, geology, permits, politics, and cost, not a clean little tracker that rises whenever the price does.
So no, I am not bearish on copper or silver, or on the buildout. I think the world is going to need an enormous amount of both.
But "the world will need this eventually" and "this is a good trade, at today's price, in today's stocks" are not the same claim, and the pitch you are being handed blurs them on purpose. The genuinely scarce thing in this cycle may not be copper, or silver, or any element on the periodic table.
It might just be someone willing to tell you what the trade leaves out.