A lack of copper has an almost illuminating quality. Software, cloud computing, and intellect contained in data rather than matter are all examples of how the digital economy frequently presents itself as weightless. The illusion of weightlessness swiftly vanishes when a metal that has been mined since antiquity becomes the obstacle between a hyperscale AI campus and its completion date.
Even well-positioned commodity merchants were taken aback by the variety of cash copper prices, which had risen beyond $13,500 per ton by early 2026 and were occasionally trading close to $14,500. The entire picture is not revealed by the number alone. What matters more is what caused it to happen: a supply chain that was already running with little slack was hit by a collision of demand sources that all arrived at around the same time, rather than a single disruption.
The largest new variable in the copper market is the development of AI infrastructure, and its demand is greater than most outsiders realize. For power distribution and cooling alone, a single NVL72 AI server rack—the type being installed in thousands of hyperscale facilities—needs about 1,850 kilos of copper. When you scale that up to a whole campus, the numbers quickly become significant: up to 40 tons of copper per megawatt of installed capacity, with major facilities using at least 50,000 tons overall throughout their liquid cooling loops, backup generation, and switchgear. These are industrial-scale material requirements associated with what is still occasionally referred to as “digital infrastructure” in the IT sector.
The price insensitivity of this demand makes it very challenging for the market to meet. Even at record prices, copper accounts for less than 0.5% of a multibillion dollar hyperscale campus’s overall budget. It’s not difficult to decide between paying high copper prices and postponing a plant that will cost billions of dollars and start making money the moment it opens for business. Tech buyers are willing to pay. Compared to the manufacturing and construction industries, which react to price signals by delaying purchases or looking for alternatives, that type of buyer is different. As a result, unlike typical industrial customers, this demand source does not decline at the peak of the price cycle.
The supply side experienced problems at the same time from several angles. Weather delays affected production in Chile, which produces around 25% of the world’s mined copper, at a time when the market had little cushion. One of the world’s largest copper resources, the Grasberg mine in Indonesia, had its own operating challenges. Then, at the perfect moment, the Democratic Republic of the Congo, a major supplier of copper concentrate, took action to limit exports of raw material, removing refining feedstock from the international market. In isolation, each of these incidents would have been controllable. When combined, they resulted in a shortfall of hundreds of thousands of tons.
It’s important to remember that other industries are also using this supply, not just AI data centers. Production of electric vehicles, grid modernization initiatives, solar and wind farm installations—all of these have been growing at the same time and demand significant amounts of copper. It seems that the commodity markets misjudged how rapidly these overlapping transitions would converge after spending years predicting demand scenarios based on past industrial patterns. Regarding the particular trends, the models were accurate. They might have been overly hopeful about how quickly supply would react.
On a good day with favorable geology and simple permitting, it takes ten years to open a new copper mine from discovery to significant production. Although it happens more quickly, increasing current capacity is still measured in years. The industry is making more significant investments in scrap recovery and processing, and recycled copper is beneficial. However, it’s difficult to ignore the limitations of the real world that deployment timelines don’t always take into consideration when observing the construction cranes over new data center campuses in Virginia, Texas, and the Arizona desert. Copper is outdated. It’s a relatively new desire. Furthermore, the difference between the two isn’t getting smaller very fast.
