
Copper: The Hidden Bottleneck Behind AI
We unpack why copper, not chips, may be the hidden bottleneck in the AI boom, tracing how the metal moves from mine to smelter to grid to data center. The episode also breaks down supply risks in Chile, Peru, and the Congo, and the demand surge from electrification, defense, and AI infrastructure.
Chapter 1
The Metal Under the Machine
Zach Martin
Welcome to K M K T I R Hub Radio. I'm Zach Martin, and this is Critical Capital, the show about the materials powering civilization and the capital behind them. This is episode one, and today's material is copper, the metal behind the AI revolution.
Zach Martin
Let's start with a question that Wall Street isn't asking loudly enough. Everybody is obsessed with AI companies. The chips, the models, the valuations. But what if the world can build the chips, and build the data centers, and still cannot get enough electricity to them? Not generate it. Deliver it. Because between a power plant and a rack of servers sits a long chain of physical things: wire, busbars, transformers, switchgear, substations, transmission lines, cooling loops. And under almost every one of those things is the same reddish metal.
Zach Martin
Copper sits under the wire, the transformer, the grid, the electric vehicle, the wind turbine, the factory floor, the radar array, and the electrification of the developing world. If the AI boom has a hidden dependency, I'd argue it's not a chip. It's a conductor.
Zach Martin
Quick housekeeping. This show is analysis and education, a map of how materials and capital connect. It is not investment advice. Nothing here is a recommendation to buy or sell anything. And in the season ahead, we'll follow the same method through uranium, rare earths, lithium, graphite, nickel, silver, antimony, tungsten, and a special on the critical mineral war. But today, copper.
Zach Martin
First, The Material.
Zach Martin
Copper is element number 29, symbol Cu. Among pure metals, only silver conducts electricity better. And copper pairs that conductivity with strong thermal conductivity, ductility, so you can draw it into wire thinner than a hair, and real resistance to corrosion. That combination is why it became the standard conductor for motors, transformers, inductors, and power transmission.
Zach Martin
Here's what I want you to sit with. Software can be rewritten. Algorithms can be compressed. Conductivity cannot. When electrons move through a material, they run into physical resistance, and no clever code changes that. That's the difference between a digital problem and a materials problem.
Zach Martin
Now, the obvious question: can't we just swap it for something else? Aluminum does compete, mostly in overhead high voltage transmission lines, because it's lighter. But for the same conductivity, aluminum needs a much larger diameter. It suffers from creep and thermal expansion, which can become a fire hazard in enclosed structures. And in tight enclosures, microelectronics, high performance transformers, and EV motors, it simply can't match copper. S&P Global's study, Copper in the Age of AI, published in January 2026, says the bulk of feasible substitution and thrifting, meaning using less copper per product, has already happened. So when we talk about copper demand, we're mostly talking about a material the world cannot easily route around.
Zach Martin
Next, The Supply Map. And I want you to picture this as a relay race, because copper has to be handed off several times before it's useful.
Zach Martin
Leg one is the mine. Chile is the powerhouse, roughly a quarter or more of global mine output, anchored by state miner Codelco and BHP's Escondida, the largest copper mine in the world. But Chile is aging. Ore grades are falling, with many deposits running below 0.6 percent copper, which means you move more rock to get the same metal. And the Atacama Desert is among the driest places on Earth, so miners build multibillion dollar desalination plants and pump seawater up to the high Andes.
Zach Martin
Peru is number two, with big porphyry deposits like Cerro Verde, Las Bambas, and Antamina. Peru's challenge is often above ground: community road blockades and shifting politics. Then there's the Central African Copperbelt, the Democratic Republic of Congo and Zambia. This is where the richest remaining deposits live, with grades often running three to five percent, several times what Chile now offers. Ivanhoe's Kamoa Kakula and CMOC's Tenke Fungurume are the headline names. But it's inland. Metal has to travel by rail and truck toward ports like Durban or Dar es Salaam, or along the developing Lobito corridor, and the regional power grids can be unstable. Richest ore in the world, hardest road to market.
Zach Martin
Leg two is the handoff most people never hear about: the smelter and the refinery. Mined concentrate is not something a wire maker can use. It has to be smelted into blister copper, about 99 percent pure, then refined into cathode, 99.999 percent pure. And here the map flips. Columbia University's Center on Global Energy Policy notes that China produces only about 8 percent of the world's copper concentrate, but about 48 percent of smelting output, and has 12 of the 20 largest smelters. About 75 percent of China's current capacity came online since the year 2000, and new smelters there can be built at up to a fifth of the cost. The United States, by contrast, has just two active primary copper smelters: Rio Tinto's Garfield in Utah and Freeport McMoRan's Miami in Arizona. Both are over a century old. A new one in the U.S. could cost up to five billion dollars.
Zach Martin
Leg three is the part you can see: cathode becomes rod, rod becomes wire and cable, and cable becomes the transmission grid, the substation, the busbar in a data center. Mine, smelter, refinery, cable, grid, server. Break any handoff and the baton drops.
Zach Martin
Now, The Demand Engine. S&P Global projects global copper demand rising about 50 percent, from 28 million metric tons in 2025 to 42 million by 2040. That growth runs along four vectors, and I think it's useful to separate them.
Zach Martin
Vector one is core economic demand. Construction, appliances, manufacturing. This is the old Dr. Copper, and it's still the largest piece. One vivid number from the study: the developing world could add as many as two billion new air conditioners by 2040. Every one of them is copper tubing and copper windings.
Zach Martin
Vector two is energy transition and addition. An EV uses about 2.9 times the copper of a conventional car, roughly 83 kilograms versus 23. Solar and wind are far more copper intensive per megawatt than fossil generation, and over 90 percent of new generating capacity installed worldwide in 2025 was solar and wind. Add the grid itself: global transmission and distribution modernization is estimated to need over 7.5 trillion dollars of spending by 2040.
Zach Martin
Vector three is the one in our title, AI and data centers. Industrial Info's Shane Mullins says a single one gigawatt AI data center can require up to 50,000 tons of copper for power distribution, grounding, and cooling. Older facilities used about 5,000 to 15,000 tons. AI ready campuses need roughly three to four times that, because of higher power loads and liquid cooling, which uses copper heat exchangers. And S&P Global says data centers could go from about 5 percent of U.S. electricity demand today to 14 percent by 2030.
Zach Martin
Now here's a fair caveat, because I'd rather be accurate than dramatic. S&P is clear that AI is not the largest source of copper demand. It's the newest. The reason the report is named for AI is that AI makes the underlying point impossible to ignore: expanded electricity supply needs copper.
Zach Martin
Vector four is defense. Drones, radar, electric propulsion, naval systems. This demand is price inelastic, because national security doesn't shop for discounts, and S&P projects defense driven demand to triple by 2040. And on the horizon, humanoid robots, which would be heavily wired. Estimates for their numbers range from tens of millions to a billion or more, so treat that one as a question mark, not a forecast.