The Copper Squeeze

The previous research piece identified power transformers as the binding constraint on AI's $690 billion capex cycle. But transformers themselves are made of copper — thousands of tonnes of it. So are the transmission lines that connect data centers to the grid, the busbars inside the data centers, the cables in every EV, and the wiring in every solar installation. Follow any electrification story deep enough and you arrive at the same place: copper.

The framing of the energy transition is about end products — solar panels, batteries, electric vehicles, AI chips. The function requires a metal that humans have been mining for 10,000 years and are now running short of.

S&P Global published an updated study in January 2026 warning of a "substantial shortfall" in copper supply. The numbers: global copper demand is projected to reach 42 million metric tonnes by 2040, a 50% increase from current levels. But global production is expected to peak at 33 million metric tonnes in 2030 and then decline as existing mines deplete. Only 70% of projected 2035 demand can be met from currently planned supply. The gap isn't a temporary blip — it's structural.

In 2026 specifically, the market is shifting from a slight surplus in 2025 to a deficit estimated between 150,000 and 400,000 tonnes, depending on whose models you trust. UBS forecasts the larger number. The deficit is driven by converging demand from AI data centers, electric vehicles, grid modernization, defense spending, and industrial reshoring — all competing for the same metal simultaneously.

The AI demand alone is staggering. A single 100-megawatt data center campus requires several thousand tonnes of copper — roughly 27 to 33 tonnes per megawatt of installed capacity. That copper goes into power distribution, cooling systems, networking, and the transformers and switchgear that connect the facility to the grid. AI data centers could consume over 500,000 tonnes of copper annually by 2030. For context, total global copper production is about 22 million tonnes per year. AI alone would claim over 2% of world supply.

Electric vehicles add another layer. An EV requires 80-100 kilograms of copper, roughly 3-4 times more than a conventional car. As EV adoption accelerates, automotive copper demand scales with it. And every EV needs charging infrastructure — more copper in cables, transformers, and grid connections.

The supply side can't respond quickly. Opening a new copper mine takes 20-30 years from discovery to production. In 2026, major unplanned disruptions at the Grasberg mine in Indonesia and Kamoa-Kakula in the DRC are further limiting output. Even the mines currently under development won't meaningfully add supply before 2028-2029. You cannot build your way out of a copper deficit on any timeline that matters for the current investment cycle.

This creates a structural price floor. J.P. Morgan expects copper to average roughly $12,075 per tonne in 2026. Bank of America forecasts $11,313 per tonne in 2026 rising to $13,501 in 2027. Goldman Sachs sees copper reaching $15,000 per tonne by 2035. These forecasts reflect a simple reality: demand is growing faster than supply, and supply cannot be created quickly.

The connection to the transformer bottleneck is direct. The previous piece showed that power transformers are the binding constraint on data center construction, with lead times stretching to 3-5 years. Those transformers are wound with copper. If copper supply tightens further, transformer production slows further, and the entire AI infrastructure buildout slows with it. The bottleneck compounds.

What to do with this — data as of 2026-04-10

BUY: Freeport-McMoRan (NYSE: FCX) — ~$66, market cap ~$95B. The largest copper producer in the Americas, responsible for 70% of US refined copper production. Produced 3.5 billion pounds of copper in 2025. Analyst consensus: Buy, average 12-month target ~$67 with a high estimate of $81. 52-week performance: up 115%. Pure-play copper exposure with significant gold and molybdenum byproducts that subsidize copper production costs. The most direct way to bet on rising copper prices.

BUY: Southern Copper (NYSE: SCCO) — ~$178, market cap ~$137B. The fifth-largest copper producer globally with the lowest production costs in the industry and the largest copper reserves of any publicly traded company. Has approved expansion projects adding 156,000 tonnes of annual output by 2027. Low costs mean higher margins as copper prices rise, and the reserve base means decades of production ahead. Premium valuation reflects the quality, but the cost advantage is real.

BUY: Global X Copper Miners ETF (COPX) — ~$82, AUM ~$6.9B. Diversified exposure to copper miners globally, tracking the Solactive Global Copper Miners index. Up 144% over the past year. This is the simplest way to get broad copper exposure without picking individual miners. Holds Freeport, Southern Copper, and dozens of smaller miners. If you want one position instead of three, this is it.

HOLD/WATCH: BHP Group (NYSE: BHP) — ~$77, market cap ~$186B. The world's largest mining company, actively shifting strategic focus toward copper. But copper is only one segment alongside iron ore and coal, so the stock won't move purely on copper prices. Better as a diversified resources holding than a copper bet. Buy if you want broader commodity exposure with copper upside.

AVOID: Copper futures or leveraged copper ETFs for long-term positions. Goldman Sachs warns that short-term copper prices may pull back from record highs in 2026 before the structural deficit fully materializes. The thesis is about a multi-year supply gap, not a short-term price spike. Buy miners who benefit from sustained high prices, not instruments that punish you for timing volatility.

The pattern from the first piece extends: AI appears to be about software. It functions as an electrical infrastructure buildout. And electrical infrastructure functions as a demand shock on industrial metals — above all, copper. Each layer you go deeper, the investments get more boring and the supply constraints get harder to solve. Buy the boring thing that everything else depends on.