Start with the arithmetic, because it settles most of the argument before the narrative gets going.
The refined copper market runs somewhere around 28 million tonnes a year. Deficit projections for 2026 cluster between 124,000 and 400,000 tonnes depending on whose balance sheet you read, with 150,000 the most commonly cited figure for 2027. Call it one percent, give or take. Within that, the incremental draw from AI data centers relative to a pre-AI baseline is estimated at roughly 110,000 tonnes for 2026.
Data centers consume about 20 to 30 tonnes of copper per megawatt across busbars, power distribution and thermal management. A one-gigawatt campus therefore locks in something like 25,000 tonnes, which is close to a single day of world consumption. The entire announced 2026 hyperscaler buildout does not move the physical balance of the copper market in any way that would prevent a shovel going into the ground.
Copper at $11,000 to $13,000 a tonne is a real cost input and a real equity thesis. It is not a shortage in the sense that anyone is going without. Nobody’s data center is late because there was no copper.
What is actually late
Lead times on a major high-voltage transformer from a tier-one manufacturer have reached sixty months. Substation-class power transformers average well over two years and generator step-up units longer. Medium-voltage switchgear is sold out through 2028 across much of the channel. Against roughly 16 GW of announced 2026 data center capacity, only about 5 GW is physically under construction, and credible industry estimates put 30 to 50 percent of planned openings at risk of delay or cancellation.
That is the wall. It has nothing to do with cathode availability.
Open up a large power transformer and the binding constraints are elsewhere in the bill of materials. Grain-oriented electrical steel for the core has risen 60 to 70 percent in price since 2020, and Cleveland-Cliffs is the only domestic producer, which makes core steel allocation rather than labour the throughput cap at several manufacturers. High-voltage bushings and on-load tap changers are the components most consistently identified as schedule breakers: highly specialised, certified per application, and served by a handful of qualified suppliers worldwide. A delay in either stalls the entire build no matter how complete the rest of the unit is.
Then there is the constraint nobody can capitalise their way out of. Every high-voltage unit must pass impulse voltage withstand and short-circuit testing before it ships. Factory testing bays are physical assets with fixed cycle times, and they cap weekly output regardless of how many shifts the winding floor runs. You can order copper on a spot market. You cannot order a testing bay and have it certified this year.
The policy tell
In April 2026 the White House invoked Section 303 of the Defense Production Act for grid infrastructure, covering transformers, substations and high-voltage circuit breakers, on the explicit finding that domestic industry cannot meet demand in a timely manner.
No equivalent instrument has been aimed at copper. When an administration that has spent two years reaching for emergency authorities across the entire minerals complex declines to reach for one here, and instead reaches for it on fabricated electrical goods, that is a revealed judgment about where the binding constraint sits. It is worth more than any forecast.
Current tightness is two accidents, not depletion
The 2025 to 2026 squeeze is substantially the product of unplanned outages at Grasberg and at Kamoa-Kakula rather than a structural failure of supply. Kamoa is already working through its recovery: guidance of 380,000 to 420,000 tonnes for 2026, rising to 500,000 to 540,000 for 2027, against a maintained medium-term target of 550,000 tonnes, with dewatering advanced and the on-site smelter ramping.
Disruption-driven deficits mean revert. That distinction matters for anyone sizing a position on the assumption that today’s tightness is the beginning of a permanent condition.
Where the copper thesis is genuinely intact
The long-dated case is strong and does not depend on any of the above. Ore grades at legacy operations have fallen roughly 40 percent since 1991. Existing and planned mines are assessed as capable of covering only about 70 percent of projected 2035 demand. Demand itself is modelled to reach around 42 million tonnes by 2040, a 50 percent increase, with data centers and defence together contributing on the order of 4 million tonnes of that. Discovery to first production in developed jurisdictions runs 15 to 29 years.
That gap is real, it is arithmetically difficult to close, and it will not be resolved by high prices alone because high prices do not compress permitting. But it is a 2032 to 2040 problem being traded as a 2026 problem, and those are different durations with different appropriate instruments.
The pattern underneath
This is the third instance of the same error in six months of supply chain commentary, and the shape is consistent enough to be predictive.
In rare earths, the anxiety attached to mined tonnage while the actual chokepoint was separation and sintering equipment. In silicon, the widely quoted 94 percent Chinese share is a polysilicon figure that describes photovoltaics, while the genuine dependency sits in metallurgical silicon smelting and is a question of electricity price. In copper, the deficit headline attaches to the metal while the physical constraint is a certified testing bay in a transformer factory.
The bottleneck is almost never the material. It is the fabricated intermediate one layer downstream, where capacity is lumpy, certification is slow, and no amount of capital compresses the qualification cycle. Markets price materials because materials have tickers and daily settlement. They underprice bushings.
Positioning implication
Through 2028 the scarcity rent accrues to the people who can deliver energised capacity: transformer and switchgear manufacturers with booked order backlogs, and the core steel position upstream of them. Those backlogs are contracted, visible, and effectively price-insensitive on the demand side, because a hyperscaler with a signed lease and no transformer will pay almost anything.
The miners are a different trade with a different clock. They are a duration position on the 2030s gap, and they are being bought on 2026 headlines. That mismatch between the horizon of the thesis and the horizon of the catalyst is where the risk sits.
Watch the testing bay expansions, not the drill results.
