Copper Just Hit a Record High — and It's the Metal the AI Buildout Can't Live Without

Copper Just Hit a Record High — and It's the Metal the AI Buildout Can't Live Without

Let me tell you something that's been on my mind since Dan Yergin said it on CNBC yesterday. The man wrote the book on energy — literally, "The Prize" — and he looked straight into the camera and called copper the best-poised commodity of the AI boom. Then he dropped the number that should have every data center operator in the world sitting up straight: the metal is up 50 percent, driven by geopolitics, policy, and demand.

I've been running hosting infrastructure for over a decade. I've watched this industry obsess over GPUs, over power, over cooling, over memory chips, over optical transceivers. And the whole time, the thing that actually carries the current from the transformer to the rack — the thing every single one of those components depends on — was quietly becoming the hardest wall of all.

Copper just hit a record high. And it's not a blip. It's the loudest warning the AI buildout has given us yet.

The 50 Percent Surge Nobody in Tech Is Talking About

Let me give you the numbers, because this is one of those stories where the market moved before the industry noticed. On Thursday, U.S. copper futures on the Comex touched a record $6.90 a pound. In London, metal pushed toward $14,000 a ton — the highest level in two months, on course for a record close. The rally has been relentless: Yergin called it a 50 percent surge, and the drivers he named are exactly the ones that should worry anyone building anything right now.

Geopolitics. Policy. Demand. Three forces hitting one metal at the same time.

The demand side is the part everyone understands — sort of. AI data centers are copper-eating monsters. A conventional data center uses somewhere between 5,000 and 15,000 tonnes of copper across its power distribution, busbars, transformers, cooling systems, and connectivity. A hyperscale AI campus? Up to 50,000 tonnes per facility. That's roughly ten times the copper footprint of legacy infrastructure. A single gigawatt AI campus has the copper appetite of half a million cars.

Now do the math on the buildout. The industry is planning gigawatts of this stuff, all at once, all over the world. S&P Global's study — the one Yergin chaired — projects copper demand will jump 50 percent by 2040, driven by AI and defense, hitting 42 million tonnes a year. And here's the kicker: nearly a quarter of that demand could go unmet.

What a Hyperscale Campus Actually Eats — 50,000 Tonnes of Copper

Most people still think of data centers as server sheds. They're not. They're copper mines above ground.

Think about everything that has to happen to deliver power to a GPU. The high-voltage line comes in, hits the transformers, and from there the copper takes over. Busbars the thickness of your arm. Cable trays running the length of the building. The entire power distribution chain, from the switchgear to the PDU to the rack — that's all copper. Then the cooling system: the chillers, the pumps, the coils — more copper. Then the grounding, the bonding, the connectors — more copper. By the time a hyperscale campus is built, you've buried tens of thousands of tonnes of the stuff in the walls.

The analysts who track this are blunt about what it means. One estimate says 13 to 14 U.S. hyperscale campuses alone need 650,000 to 700,000 tonnes of copper. Another puts copper intensity at 47 tonnes per megawatt by 2026. Forbes projects data centers could gobble half a million tons of copper a year by 2030. Half a million tons. A year. Just for the buildings where we stick GPUs.

And the supply side? That's where it gets ugly.

The Secondary Bottleneck — It's Not the Mine, It's the Transformer

Here's the pattern I've watched this industry repeat for two years now. Every time the market thinks it has found the bottleneck, the bottleneck moves. First it was GPUs. Then memory. Then power. Then cooling. Then the grid. Now it's copper — and copper isn't even the end of the chain, because the thing that actually turns copper into a working data center is the transformer, and transformers are the true choke point.

Large power transformers now have lead times of two to five years. That's not a typo. Two to five years, from order to delivery, for the single most important piece of electrical equipment in the building. Switchgear and generators are stretched the same way — 120-plus weeks in some cases. Bloomberg reported in April that more than half of the U.S. data centers planned for this year would be delayed or canceled — not for lack of money, not for lack of land, but because there simply aren't enough transformers and switchgear to bring them online.

Wood Mackenzie's numbers tell the same story: of roughly 12 gigawatts of U.S. data center capacity expected to come online in 2026, only about 5 gigawatts is actually under construction. Close to half the planned pipeline is delayed or canceled, driven primarily by the unavailability of transformers, switchgear, and batteries.

Half. The pipeline. Delayed. And the raw material at the bottom of all of it — the copper that goes into every transformer, every busbar, every switch — just hit an all-time high. Hitachi Energy is throwing $6 billion at new transformer capacity, and it's still not close to enough.

The Tariff Arbitrage That's Sucking Metal Out of the Market

Now layer on the part that makes this cycle different from every copper cycle I've seen: Washington.

Traders have been betting on U.S. tariffs on refined copper, and the arbitrage is doing exactly what arbitrage does — it's pulling metal out of the global market and pointing it at America. The spread between Comex and London prices hit $621 a ton in early August. Front-month Comex contracts traded more than $500 a ton above London cash. One trader put it plainly: all tons are being directed to the U.S. Imports are running at historically elevated rates of 150,000 to 200,000 tons a month.

Meanwhile, visible inventories are at scary lows. Combined LME, Comex, and Shanghai exchange inventories stand at about 1.12 million tonnes — roughly 15 days of global demand. Fifteen days. That's not a buffer, that's a heartbeat. Any supply shock — a mine outage, a port delay, a strike — and the price doesn't just move, it gaps.

Analysts are calling it a new situation for Dr. Copper, and for once the old adage might be wrong. Copper is supposed to be the metal that predicts the economy. This cycle, it's the metal that predicts the AI buildout — and it's saying the buildout is about to get very expensive.

Washington Finally Noticed — Three Billion Dollars and a Roundtable

Here's the part that would be funny if it weren't so expensive. The same week copper hits a record high, the administration holds a roundtable with mining executives at the State Department and announces $3 billion in critical minerals and battery projects. The Pentagon is putting $1.4 billion into a battery facility and $400 million into scandium production. The message is right — America needs domestic mining, desperately — but three billion dollars against a problem measured in trillions of dollars of infrastructure and decades of underinvestment is not a fix. It's a down payment.

And here's the uncomfortable truth the politicians won't say out loud: even if every mine in America got approved tomorrow, it wouldn't matter for this cycle. A new copper mine takes 10 to 15 years from discovery to production. The mines that will supply the 2030s are mostly already in the ground or in the permitting queue today. S&P Global projects global copper production peaks in 2030 at 33 million tonnes — and demand keeps climbing past it. That's the definition of a structural gap, not a cyclical one.

China understands this. Its grid investment rose 13 percent year over year in the first half of 2026, and Beijing announced a roughly $574 billion power grid upgrade plan. They're not waiting for the market to sort itself out. They're building the copper supply chain of the future while we're still holding roundtables.

What This Actually Means for Independent Hosting Providers

Alright. Enough doom. Here's what you do about it, because there is stuff you can do.

First, lock your electrical equipment orders now. If you're planning any capacity expansion — even a modest one — get your transformer and switchgear orders in today. The lead times are two to five years and they are not coming down. Every month you wait is a month of queue you're giving to someone else.

Second, price copper into your colocation contracts. If you're buying colo space or building your own, assume electrical infrastructure costs keep rising. The hyperscalers will absorb it and pass it to their cloud prices. You need to build the same escalation into your rates — not as gouging, but as math.

Third, treat copper as a strategic material, not a commodity. If you're designing a power chain, look hard at where you can cut copper intensity — higher-voltage distribution, aluminum busbar where code allows, tighter cable runs. Every kilogram you don't need is a kilogram you don't have to fight for.

Fourth, watch the Comex-LME spread as your early-warning signal. When the spread widens past $500 a ton, metal is leaving the global market for the U.S., and every electrical component you need gets more expensive and slower. When it narrows, the pressure is easing. That spread is now a better leading indicator for your business than any GPU roadmap.

Fifth, don't break ground without a transformer reservation. A data center without a transformer is a very expensive warehouse. The projects that survive this cycle will be the ones that secured electrical equipment before they started pouring concrete — not the ones that poured first and prayed.

The Structural Reality — This Shortage Has a Decade Left to Run

Let me be clear about what this is and isn't. This is not a crash. This is not the AI bubble popping. This is the physical world finally enforcing its constraints on a digital boom that thought it could outrun them.

The buildout isn't going to stop. The capital is too big, the demand is too real, and the competitive pressure to build first is too intense. But it is going to get slower, it is going to get more expensive, and it is going to get more selective. The projects with real power, real transformers, and real supply chains will win. The ones that were never more than a press release will quietly die.

You can already see the sorting happening. Half the 2026 pipeline delayed. Five gigawatts under construction out of twelve planned. The survivors are the ones who understood that the bottleneck isn't in the code — it's in the metal.

The Bottom Line

Here's the truth I keep coming back to: for two years, this industry has treated every bottleneck like it was temporary. GPU shortages, memory shortages, transformer shortages, and now a copper shortage — each one was supposed to be a speed bump on the road to infinite compute.

They're not speed bumps. They're the road.

The AI buildout is now running on a metal that costs more every single day, in quantities the mining industry cannot produce, on timelines that stretch past a decade. Yergin called copper the best-poised commodity of the AI boom, and he's right — but "best-poised" cuts both ways. It means the metal is going up. It also means the people who need it are going to pay.

If you're an independent hosting provider, that's not a reason to panic. It's a reason to plan. Lock your equipment. Price your contracts honestly. Watch the spread. And never, ever start a build without the transformer.

Because the copper doesn't care how big your vision is. It just wants to know if you ordered it early. Buh trust me on that one.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.