The AI Buildout Is Hitting a Copper Wall

Taseko Mines poured its first copper cathodes from the Florence Copper operation in Arizona last week. That matters because it is the first new greenfield copper production in the United States since 2008 — and it took 15 years to reach this point. The mine was permitted, built, and commissioned during a period when nobody was seriously worried about copper supply. That period is over.

The International Energy Agency confirmed in March that copper prices blew past $13,000 per metric ton. ING Group puts the 2026 refined copper deficit at 600,000 tons — the widest gap in two decades. Smelter processing fees have fallen to zero, which means refiners are now working for free just to keep the lights on. Negative treatment charges are forcing further smelter cuts, tightening refined supply into a market that is already short. Meanwhile, high-density AI compute clusters require between 30 and 47 tons of copper per megawatt for specialized liquid cooling and high-density power distribution. A mid-sized traditional data center build consumes 5,000 to 15,000 tons of copper. A high-density AI campus can require as much as 50,000 tons. The benchmark lease size for new data centers has risen from 100 megawatts a few years ago to over 1,000 megawatts today. A single large AI campus can require roughly 27,000 tons of copper — about what a mid-sized mine produces in a year.

In 2026 alone, AI infrastructure is projected to add 110,000 tons of new copper demand, an 80 percent jump from two years ago. The global copper market is currently operating in a structural deficit estimated at 330,000 to 400,000 metric tons for 2026, according to industry analysis by Wood Mackenzie cited by Tom's Hardware in December 2025. S&P Global projects global copper demand will rise 50 percent by 2040, from 28 million metric tons today to 42 million, while production peaks around 2030 at roughly 33 million metric tons — creating a shortfall of 10 million metric tons by 2040 without significant new supply investment.

Copper prices reflect the squeeze. The metal hit a record $6 per pound in January 2026 and was trading near $5.61 per pound as of April, according to Manufacturing Dive and Omdia. High prices are incentive for exploration and feasibility studies. They are not creating mines. A mine that receives its final permit in 2028 begins production no earlier than the mid-2030s.

The permitting process for new mines in the United States is itself a bottleneck. Ambler Metals submitted a Clean Water Act Section 404 permit application to the U.S. Army Corps of Engineers on April 20, initiating federal permitting for the Arctic copper project in northwestern Alaska. The deposit was first identified in the 1970s. At roughly 5 percent copper equivalent, it ranks among the highest-grade undeveloped copper deposits globally. The permitting clock started last month. First production, optimistically, is a decade away.

The Senate voted 50-49 on April 16 to overturn a Biden-era ban on copper and nickel mining in the Superior National Forest near the Boundary Waters. Twin Metals Minnesota, a subsidiary of the Chilean mining giant Antofagasta, has been trying to develop the deposit near Ely for years. The April 16 vote clears the procedural route for Twin Metals to reapply. First production, assuming permits survive expected legal challenges: the early 2030s at the earliest. A new copper mine takes roughly 17 years on average from discovery to first production, according to S&P Global. An AI company that signs a power purchase agreement next quarter can have servers online in 18 months.

Elsewhere, the supply response is marginal. Glencore is restarting the Bajo de la Alumbrera copper mine in Argentina with equipment delivery booked in Q1 2026. BHP submitted a permit application for a new concentrator at Escondida, the world's largest copper mine. Taseko's Florence Copper — the first greenfield copper production in the U.S. since 2008 — is the exception that proves the rule: it took 15 years to build.

The companies that locked copper supply agreements early are insulated from spot volatility. The rest are exposed to a commodity whose price floor is set by geology and permitting timelines, not by market cycles. Intel shares surged 28 percent in premarket trading on April 24 after reporting stronger-than-expected demand for its processors from AI service providers — demand so strong it sold chips it had previously written off. The compute buildout is real, it is accelerating, and the physical infrastructure to support it is running on a geological timeline that does not care about quarterly earnings.

If AI infrastructure plans proceed at announced scale, copper demand from data centers alone could establish a structurally higher price floor for the metal. If those plans compress or shift toward more efficient architectures, the demand picture weakens. Mine supply will not respond either way within a decade. The physical constraint is not a talking point. It is the time between a Senate vote and a working mine.