Amazon Is Buying Copper, Not Chips

This past January, Amazon didn't sign a deal with a chip designer, a cloud provider, or a software company to feed its AI ambitions.

It signed a two-year agreement with Rio Tinto, one of the largest mining companies on the planet, to lock up copper for its data centers.

The supply comes from a restarted Arizona mine, the first new domestic copper U.S. tech has secured in over a decade.

The same reddish metal humans have pulled out of the ground since the Bronze Age is now the tightest chokepoint in the most ambitious technological buildout in history.

A single hyperscale AI data center, the kind built to house NVIDIA's HGX systems, runs on roughly 50,000 tons of copper, according to the Copper Development Association.

Wiring, busbars, transformers, switchgear, cooling loops, interconnects. Every rack, every server, every cable tray. Copper runs through all of it.

The United States mines about 1.1 million tons of copper a year, but smelts only 585,000 tons.

The rest ships overseas for processing, to Mexico, Canada, Japan, and yes, China. Set that refining ceiling against one hyperscale center's 50,000-ton appetite, and a single facility swallows 8.6% of everything the country can refine in a year.

Take every refined ounce, give none of it to homes, cars, the grid, or the military, and the most you could wire is roughly eleven hyperscale data centers a year. The US is planning hundreds.

The binding constraint on AI is now dirt, and the dirt is running out.

Jensen Huang: AI Needs 1,000 Times More Energy

At Stanford earlier this year, in the university's Frontier Systems series, Jensen Huang put a number on it. The energy computing will need, he said, is "probably 1,000 times more than we currently have."

The man building the engines of the AI revolution is telling you, in public, that the power to run what's coming doesn't exist yet. The guy selling the shovels is saying the mine isn't deep enough.

Hyperscaler data center capex in the U.S. this year is running somewhere between $400 billion and $700 billion. McKinsey puts cumulative worldwide data center investment at roughly $6.7 trillion by 2030.

By some estimates, AI spending drove close to half of last year's U.S. GDP growth.

Wall Street keeps glossing over a very simple fact. AI is digital at the point of use and physical at the point of scale.

Every one of those data centers needs copper for wiring, aluminum for cooling and cladding, gallium and germanium for the chips, rare earths for the drives, and silver, gold, tin, and platinum across the boards.

The demand curve isn't linear, either.

As power density and cooling complexity climb, so does the copper per megawatt. S&P Global puts data center copper intensity at 30 to 40 tons per megawatt.

At gigawatt scale, which is the size hyperscalers now build, that lands back at tens of thousands of tons per campus. You can't code your way around physics.

The Easy Copper Is Already Gone

S&P Global projects global copper production peaking in 2030 near 33 million tons, with the supply gap widening from there. The market is already pricing the squeeze.

Copper hit an all-time high above $14,500 a ton on the London Metal Exchange this past January, and even after cooling off it trades in the low $13,000s, up about 30% over the past year.

The average ore grade at the largest U.S. copper mines is 4 times lower than at the top 10 international mines, so American miners move 4 times as much rock for the same amount of metal.

Costs rise, yields fall, and the easy copper left the building a long time ago.

In South America, which dominates global supply, ore grades have dropped 44% since 2000, from 1.3% to 0.7%.

Miners now pull less than a ton of copper out of every 100 tons of rock. Geology sets that ceiling, and no earnings call talks it back up.

The Trump administration has layered tariffs onto copper imports, and the U.S. still buys about half its copper abroad, roughly 720,000 tons a year from Chile, Canada, and Peru.

Tariffs on that flow raise the landed price and reduce the available supply.

S&P Global estimates the marginal cost of mine production climbed 37% between 2019 and 2025. Sulfuric acid, essential for copper leaching, is in its own crunch after the Strait of Hormuz blockade and China's export restrictions.

What's building is a structural deficit, and structural deficits don't fix themselves on earnings calls.

China Refines 70% of the Minerals AI Needs

Growing up as a first-generation American, I learned something early that most people learn too late. The people who control the raw inputs always have leverage over the people who only consume the outputs.

My father understood it instinctively. He never needed an MBA to know that whoever owns the thing everyone needs gets to name the price.

That lesson scales up to nations.

China refines around 70% of the world's 20 most critical minerals and more than 90% of rare earths. It owns the processing technology, the IP, and the deepest bench of mining expertise on earth.

While the U.S. debates tariffs and permitting timelines, China is building one of the largest data center complexes on the planet, China Telecom's Inner Mongolia Information Park, at nearly one million square meters.

The U.S. holds a modest lead in frontier AI models.

China holds a structural advantage in the physical materials AI needs to scale: gallium-nitride for power electronics, germanium for semiconductors, scandium for fuel cells, rare earths for everything from drives to defense systems.

The U.S. consumes critical minerals. China supplies them. In my experience, the supplier always sets the terms.

Defense, EVs, and Quantum Want the Same Dirt

AI data centers aren't the only buyer. Defense, electric vehicles, robotics, grid electrification, and renewables all draw from the same finite pool of copper, aluminum, rare earths, gallium, and germanium.

An electric vehicle needs nearly three times as much copper as a conventional car.

Defense systems, from autonomous weapons to secure comms to quantum hardware, are packed with critical minerals. And quantum is its own draw.

Google's Willow chip ran a benchmark in under five minutes that would take the fastest classical supercomputer an estimated 10 septillion years, longer than the age of the universe.

Machines like that depend on tiny quantities of specialized minerals. Without them, quantum scaling stalls too.

Every frontier of the 21st century converges on the same bottleneck: the physical materials that make it work. The fourth industrial revolution assumes those minerals will be there.

That assumption is wrong, and when assumptions meet geology, geology wins every time.

What This Means for Your Money

The AI trade is a materials trade, and the investors who internalize that are the ones who will profit from it.

Wall Street still prices AI as a purely digital phenomenon, as though all that matters is who has the best model or the fastest chip. The real constraint sits one layer down, in the copper, the grid, and the dirt.

After three decades of underinvestment, the mining sector still trades cheap.

Gold miners change hands around seven to eight times EV/EBITDA while the S&P 500 sits in the high teens. They carry less debt, pay fatter dividends, and run margins well above the index.

The companies digging these materials out of the ground trade at a fraction of the tech names that can't run without them.

The market is handing you that gap.

The one thing worth doing this week: audit your portfolio for exposure to critical minerals and the companies that mine them.

If your book is 100% tech and zero materials, you're betting the physical world will cooperate with the digital world's ambitions.

History says it won't.

Look at copper miners. Look at high-grade deposits in friendly jurisdictions, the ones with real infrastructure, affordable power, and governments that want them to succeed.

Look at the suppliers of scandium for fuel cells, rare earths for magnets, aluminum for cooling.

Everything I just handed you is free. The thesis, the bottleneck math, the corner of the market nobody's pricing yet, the nudge to go audit your own book this week.

Act on all of it without paying me a dime. I hope you do.

The free reader gets the thesis. The Premium member gets the trade.

I've been writing this materials story to Moonshot Minute Premium for over a year. Those readers got more than the argument.

They got the specific names, the entries, the position sizing, and the exact moment to pull money off the table.

The track record is why I can write a section like this without flinching.

In our first year, every position we closed was in the green. Not one realized loss.

Several picks more than doubled, and when they did, we sold enough to pull our entire original stake back out and let the rest run on house money. Those free rides are still going.

Our silver position is up more than 160%. Our gold miners are over 100%. Our memory-chip pick sits north of 150%. One of our AI hardware names is up more than 300%.

I'd rather earn your trust than your applause, so here's the other side: Some of our open positions are down, the worst by about 30%. I plan for that.

Our winners run several times larger than our losers, which is the whole reason you buy asymmetry and cut what stops working.

A newsletter that only shows you green screenshots is hiding the part that makes the green believable.

The thesis you just read is the same one driving those winners, and I think we're still early. The copper deficit hasn't peaked. The grid hasn't been rebuilt.

The next names positioned for exactly what this essay describes go only to Premium members and that’s why you should seriously consider joining Premium today.

The Bottom Line

The AI revolution is real. The models keep getting smarter, the chips faster, the capital flowing like nothing I've seen in decades of watching markets.

But every digital dream runs on a physical foundation. Every algorithm needs electricity, every data center needs copper, every chip needs gallium.

And every one of those materials comes out of a hole in the ground, dug by people, in places with their own politics, geology, and constraints.

The biggest tech boom in history could stall on a chunk of copper.

Make sure you own the copper.

Double D

🔓 Premium Content Begins Here 🔒