Every hyperscale data center being built right now needs copper for every meter of power cable, every transformer, every busbar, and every cooling loop. The GPU gets the headlines. Copper does the work.

That gap between what gets attention and what actually makes the system run is where the investment opportunity lives. The AI trade everyone talks about is silicon — NVIDIA, AMD, the fabless chip designers. The AI trade almost nobody talks about is the metal that carries electrons from the substation to the GPU rack. Copper has quietly become the most strategically important industrial metal of the decade, and the supply side is not ready for what’s coming.

This is not a cyclical story. Copper has always had booms and busts tied to Chinese industrial demand and global GDP growth. What’s happening now is different. There is a new, structurally permanent demand vector — AI data centers — layered on top of electrification, grid modernization, and defense spending. And the supply side is structurally constrained by falling ore grades, a decade-long permitting cycle, and declining production from the world’s largest mines.

What AI data centers actually need from copper

The connection between AI and copper is not abstract. It is physical and measurable.

A single large-scale AI data center requires between 27 and 33 tonnes of copper per megawatt of installed power capacity, according to industry estimates. That covers wiring, busbars, transformers, cooling systems, and power distribution equipment. A 150-megawatt hyperscaler campus consumes roughly 4,500 tonnes of copper. For reference, the average American home uses about 200 kilograms. One data center equals roughly 22,500 homes in copper intensity.

The numbers scale fast. Macquarie analysis estimates AI data centers will consume 330,000 to 420,000 tonnes of copper annually by 2030, with demand peaking above 570,000 tonnes in 2028. The International Copper Study Group has already revised its 2026 global balance from a surplus of 200,000 tonnes to a deficit of 150,000 tonnes — a swing of 350,000 tonnes driven almost entirely by the AI buildout.

This is not optional copper. You cannot run a GPU cluster without it. Fiber carries data. Copper carries power. Every $100 billion committed to AI infrastructure is effectively a purchase order for the red metal. The hyperscalers — Microsoft, Google, Amazon, Meta — have committed well over $500 billion in cumulative data center capex through 2030. Each dollar of that has a copper content baked in.

The broader picture: S&P Global’s 2026 study “Copper in the Age of AI” projects global copper demand rising from 28 million metric tons in 2025 to 42 million metric tons by 2040 — a 50 percent increase. The study identifies four demand vectors: core economic growth, energy transition (renewables, EVs, grid expansion), AI and data centers, and defense modernization. AI alone represents roughly 2 million metric tons of additional demand between 2025 and 2040.

The supply side is broken

Here is the problem. Demand is accelerating. Supply is not.

S&P Global projects global copper production will peak around 2030 at 33 million metric tons, then decline to 22 million metric tons by 2040 if no new mines are developed. That is lower than current production. The world needs approximately 80 new sizable copper mines by 2040 to meet projected demand. The average new copper mine takes 10 to 15 years from discovery to production in the best regulatory jurisdictions, and longer in difficult ones.

Ore grades are falling across the board. Chile’s Codelco, the world’s largest copper producer, has seen average ore grades decline from 1.02 percent to 0.66 percent over the past decade. Lower grades mean more rock must be moved per tonne of copper, which means higher energy costs, more water, and more waste. The easy copper has already been mined.

The supply disruptions are compounding. The Grasberg mine in Indonesia, one of the world’s largest copper-gold operations, suffered a force majeure event after a mudslide that could remove 500,000 tonnes of copper supply over 12 to 15 months. Seismic disruptions in the Democratic Republic of Congo cut 90,000 tonnes. These are not anomalies — they are the new baseline in an industry where the remaining deposits are deeper, lower grade, and in more difficult jurisdictions.

The result is a projected supply shortfall of 10 million metric tons by 2040 — roughly 25 percent of projected demand. That deficit exists even after assuming recycled copper scrap more than doubles from 4 million to 10 million metric tons annually. Recycling cannot close the gap. Primary mined supply remains essential, and the industry is not building enough of it.

The AI copper demand timeline

The timeline matters because the supply-demand imbalance is back-loaded. The next three years are where the tightness really shows up.

2026: The ICSG projects a 150,000-tonne refined copper deficit. JP Morgan is at 330,000 tonnes. Copper hit an all-time high above $13,300 per ton in January 2026, up roughly 60 percent from April 2025. Goldman Sachs has a $13,735 target. Citi targets $15,000. These targets are moving higher, not lower.

2028: The year AI data center copper demand is projected to peak at over 570,000 tonnes annually. This is also roughly when the hyperscaler capex commitments from 2024-2025 start translating into physical construction. The concrete-and-copper phase of the AI buildout is still ahead of us.

2030: Data centers could account for 14 percent of total US electricity demand, up from 5 percent in 2025, according to S&P Global. Every percentage point of that growth requires copper-intensive grid upgrades — new substations, new transmission lines, new distribution transformers. The grid interconnection queue in the US hit 2.6 terawatts by the end of 2024, double the entire existing generation fleet. Each of those projects needs copper.

2040: The 10 million metric ton shortfall crystallizes unless major new mines have been built. Given the 10- to 15-year development timeline for new mines, the decisions that determine whether that deficit is real or not are being made right now. The industry is not making them.

Which copper stocks are positioned

The copper stock universe breaks into three tiers, each with a different risk-reward profile.

The large-cap producers. Freeport-McMoRan (FCX) is the closest thing to a pure-play benchmark. It owns the Grasberg mine in Indonesia — still one of the world’s most valuable copper-gold assets — and accounts for roughly 70 percent of US domestic refined copper production. FCX is also pioneering leaching technology that can extract copper from waste rock at a fraction of traditional mining costs, adding an estimated 800 million pounds of incremental volume. At projected 51 percent EPS growth in 2026, FCX offers the most direct operational leverage to higher copper prices.

Southern Copper (SCCO) delivers the highest profit margins in the industry thanks to its low-cost Peruvian and Mexican operations. Q1 2026 revenue hit $4.25 billion, up 36 percent year over year, with net income of $1.58 billion. SCCO always trades at a premium valuation, but the earnings quality supports it.

BHP Group offers the most diversified exposure with a roughly 40 percent earnings contribution from copper, a 5.1 percent dividend yield, and the world’s largest copper mine (Escondida in Chile) as its flagship. BHP is less sensitive to copper price moves than the pure-plays, but it also comes with less single-commodity risk.

The mid-cap growth names. Teck Resources (TECK) has transformed itself from a diversified miner into a copper growth story after selling its steelmaking coal business in 2025. The Quebrada Blanca Phase 2 (QB2) ramp-up in Chile is expected to roughly double Teck’s copper production by 2027. At roughly 15 times earnings, it trades cheaper than the large-cap peers while offering the most aggressive volume growth profile in the sector.

Ivanhoe Mines is a higher-risk, higher-upside development story centered on the Kamoa-Kakula complex in the Democratic Republic of Congo — one of the highest-grade copper deposits in the world. The grade advantage gives Ivanhoe a significant cost edge, but the DRC jurisdiction adds political risk that limits the multiple the market is willing to assign.

The ETF vehicles. For investors who want copper exposure without picking individual stocks, the ETF options have strong track records. The Global X Copper Miners ETF (COPX) holds roughly 40 mining stocks and has returned approximately 69 percent over the past year and over 500 percent over the past decade. The Sprott Junior Copper Miners ETF (COPJ) targets smaller, higher-growth names and returned 109 percent in 2025.

The key trade-off with mining ETFs: they are leveraged to the copper price. When copper rises 10 percent, mining equities tend to rise 25 to 35 percent as margins expand. The same works in reverse. These are not buy-and-forget positions. They are cyclical instruments that amplify the underlying commodity move.

The structural constraints that matter

Three factors make this cycle different from past copper booms.

First, the permitting timeline is broken. A new copper mine in the United States takes 10 to 15 years to permit — longer than in almost any other developed country. The last major US copper mine to reach production was the Resolution Copper project in Arizona, which was first proposed in the 1990s and is still not fully permitted. This means the supply response to higher prices is dramatically slower than in previous cycles. In the 2000s commodity supercycle, high copper prices brought new supply online within three to five years because the regulatory environment was looser and the best deposits were already identified. That is not happening this time.

Second, the quality of remaining deposits is lower. The average copper ore grade globally has fallen by roughly 30 percent over the past 15 years. The deposits that are easiest and cheapest to develop have been developed. What remains is deeper, lower-grade, and in jurisdictions with higher political risk — Chile and Peru facing resource nationalism pressures, the DRC dealing with infrastructure and stability challenges, and Indonesia imposing export restrictions to force domestic processing.

Third, AI demand is not correlated with the industrial cycles that have historically driven copper prices. Previous copper booms were tied to Chinese infrastructure buildout and global manufacturing. Those are cyclical — they slow when GDP slows. AI data center construction is being driven by a technology arms race between hyperscalers that is largely decoupled from GDP growth. Microsoft, Google, Amazon, and Meta are not going to stop building data centers because Chinese PMI data softens. This demand vector has a different, more durable growth profile than anything the copper market has seen before.

Where this leaves you

The copper story is not complicated. Demand is growing at a pace the industry has never seen, driven by a technology transition that has no substitute for the metal. Supply is constrained by geology, regulation, and years of underinvestment. The math produces a persistent deficit that gets worse with every new data center announcement.

The risk is timing and volatility. Copper stocks are not bonds. They swing hard. COPX dropped roughly 3 percent in the past month on growth concerns despite a fundamentally bullish outlook. FCX pulled back from its 52-week high of $71 on lowered guidance related to the Grasberg ramp-up uncertainty. These drawdowns are routine, and they test investor conviction in ways that software stocks rarely do.

But the structural thesis for copper stocks rests on a foundation that strengthens with every hyperscaler capex announcement and every mine disruption. The world needs to add the equivalent of another Chile — the world’s largest copper producer — to meet 2040 demand. That is not happening on the current trajectory. The gap between what is needed and what is being built is the opportunity.

Copper is the metal the AI industry runs on. It cannot be designed around, substituted away, or engineered out of the problem. Every GPU cluster needs power, and every watt of power moves through copper. That is not a thesis. That is a physical constraint. And physical constraints are the most reliable investment theses there are.