TL;DR: Everyone’s debating whether AI will take your job. Almost nobody’s asking whether AI can feed itself. For 45 years, the West has answered “What is GDP for?” with “whatever markets decide” – but the AI buildout demands energy, water, and rare earths at scales that expose a fracture between the financial ledger and the material one.
The reckoning won’t come from regulation, but from the periodic table.
My trusted bakery charges 2.20€ for a croissant now. When I was a kid, the equivalent pastry was a rounding error.
The croissant is the true golden ratio. The most honest inflation indicator on the planet. Forget CPI. Count your pastries.
What if the croissant problem isn’t just monetary? What if we’ve been measuring the wrong thing entirely?
A recent conversation on The Great Simplification cracked this open for me. In 90 minutes, two guests laid out a picture most of the finance world won’t look at.
The short version: the AI economy isn’t a cloud, but a metabolism.
Two Sets of Books
Every economy keeps two sets of books. The one your broker shows you, and the one that actually matters.
There’s the financial ledger: Bloomberg, GDP reports, your brokerage app. Prices, returns, vibes.
Then there’s the material ledger: Tons of copper. Liters of water. Megawatt-hours. Kilograms of dysprosium. (I had to learn to spell that this week, you’re welcome.)
Ledger #2 doesn’t care about your DCF model. It doesn’t negotiate. It just asks: do you have the stuff, or don’t you?
For 45 years, the West has balanced one book and shredded the other. “Markets will work it out.” GDP goes up, therefore we’re fine. The financial ledger became the map and the territory. We forgot there was ground underneath.
Craig Tindale, a private investor who’s spent four decades mapping East-West supply chains, puts it sharper than I can: “if you haven’t got materiality built in, your paper-on-paper economy eventually destroys your currency and your entire system.”
The same argument I made in Everything is a Ponzi Scheme – just from the currency side rather than the commodity side.
The financial illusion: your portfolio is going up, but your money is going down.
The material illusion: you think you’re building the future, but you don’t own the ingredients.
The Heaviest Metabolism
We talk about AI like it’s weightless. Digital. A question goes in, an answer comes out. Magic.
Then you look at the electricity.
Meta’s planned facility in Louisiana will consume more than twice the electricity of New Orleans. The buildout is global, and every rack in every data center is a small mining operation’s worth of refined metals, assembled.
This isn’t software eating the world. This is software demanding the world feed it – at a pace that makes the green energy transition look like a light snack.
Tindale called it “probably the heaviest metabolism in the industrial system.” It snuck up on us because the cloud metaphor is such a good anaesthetic.
Nobody panics about a cloud.
But a cloud that can’t get its parts? That panics.
The Golden Screw
During COVID, Tindale couldn’t drive his car for nearly 12 months. One part. Unavailable. Everything else worked fine. Didn’t matter.
He calls this the “golden screw” problem: in a complex system, one missing component breaks the whole machine. The more complex the system, the more screws it hides.
AI’s supply chain is full of them. And almost every one runs through the same country.
China doesn’t just mine rare earths. It refines them. It controls the off-take contracts. Even when the mine sits in Australia or Canada, the ore ships to Chinese refineries because the West doesn’t have processing capacity and can’t build it profitably under its own rules. Canada killed a copper refinery because the ESG cost alone hit 9% of capital; Western Australia canceled another because it simply wasn’t profitable.
We’re not losing a trade war. We’re forfeiting one.
China is willing to produce at a loss. Indefinitely. The goal isn’t profit, but control. Meanwhile, Western investors want 20-30% returns, regulators add cost, and communities block refineries over arsenic.
Michael Every, global strategist at Rabobank (and overall British big brain), put it bluntly: wars are won with bullets, not profits. And you can’t build the bullets if someone else controls the metal.
This is why roughly half my portfolio is in AI infrastructure. Not the models. Not the apps. The substrate: power, land, permits. The unglamorous plumbing that has to exist before a single model gets trained. I’d rather bet on atoms than abstractions. The financial ledger says the AI trade is crowded. The material ledger says it hasn’t even started.
The Vertical Trap
This is where the spreadsheet hits the species.
Ask an economist about rare earth supply chains, and they’ll talk about price signals. Ask a general about climate modeling and watch the subject change. Everyone’s an expert. Nobody sees the board.
We’ve spent decades building vertical specializations – each one deeper, narrower, more credentialed – and never once integrated them horizontally. The economist doesn’t talk to the geologist. The geologist doesn’t talk to the general. None of them talk to each other because their jobs depend on not understanding each other’s domains.
Every articulated this better than I’ve seen anywhere: “a narrow-boundary argument will almost always win a debate against a wide-boundary perspective.”
Complexity loses every debate it enters. “Markets will sort it out” wins the room.
“Yes, but it depends on material constraints, energy access, demographics, and who controls the refining” gets a polite nod and a “we’ll get back to you.”
The Wrong Game
For 45 years, the West has answered the GDP question with: “whatever markets decide.” Efficient allocation. Comparative advantage. Let the invisible hand sort it out.
Great strategy… unless the other team isn’t playing that game. Unless they’re using state capitalism to wall up every choke point while you’re still reading Adam Smith and hoping the price mechanism will fix a geological bottleneck.
The West built an atomized, hyper-efficient system that excels at exactly one thing: consumption. We sit on the couch, watch Netflix, and expect the world to deliver stuff because we’re Western. (Every’s words, not mine.)
So what is AI for?
If the answer is “replacing 80% of white-collar jobs in 18 months,” congratulations – you’ve launched the worst marketing campaign in history.
Telling everyone they’re about to become redundant before you’ve proven the technology works at enterprise scale. Bold.
But there’s a better answer. Feed a single system the commodity flows, the defense budgets, the water tables, and the shipping manifests – all at once. Not to predict the future. Just to see the present clearly for the first time. AI as the thing that finally connects the verticals. Not a consumer gimmick, but a lens.
The irony is that building the lens requires the very materials we can’t secure.
(Who says the universe doesn’t have a sense of humor?)
Renaissance Man & Feudal Man
Michael Every made a point near the end that’s been rattling around my skull since.
What we need, he said, isn’t just the intellectually broad thinker who can see across disciplines. We need Renaissance Man and Feudal Man. People who understand systems theory and know how to grow a field of vegetables. The philosopher who can hew wood and carry water.
And then, with perfect comic timing: “I increasingly think most of our population is already feudal man. We just don’t know it yet.”
Tindale’s response was better: “We haven’t forgotten. Look at our ancestors. Look at how many people had to die for us to get here. They knew all that stuff. So it can’t be that hard to learn.”
Between those two lines is a whole philosophy of the next 20 years. We’re going to need people who can read a balance sheet and can a jar of tomatoes. The average MBA program teaches neither. The average trade school teaches one.
And we have maybe one generation to teach people how to think before AI does the thinking for them. After that, the question isn’t whether the tool works.
It’s whether anyone left knows how to check.
Grow Something
After 90 minutes of supply-chain warfare, geopolitical chess, and existential resource constraints, Tindale closed with this:
“Be kind to yourself. You can’t predict what’s gonna happen. So take yourself out of the future and bring yourself to the present. Walk among the trees. Grow some vegetables.”
He grows food. He builds things. He cans goods and tells his kids to act like zoologists studying their own species: humans have specific needs, specific inputs, and you ignore them at your own expense.
I wasn’t expecting tenderness from a man who can map the world’s rare earth choke points from memory. But there it was.
The greatest philosophy was written in hard times. The worst economics was done in good times.
The biggest threat to AI isn’t regulation. It isn’t the alignment problem.
It’s the periodic table. And you can’t lobby the periodic table.
Inspired by Craig Tindale, Michael Every & Nate Hagens on The Great Simplification. Listen to it.
On the weight of climate and what it teaches: Cold Introspection
Frequently Asked Questions
What does “atoms vs abstractions” mean?
It’s the idea that the digital economy – AI, software, finance – runs on a physical substrate it mostly ignores. “Abstractions” are the models, the code, the financial instruments. “Atoms” are the copper, lithium, rare earths, water, and electricity that make all of it possible. When you bet on technology without securing the materials, you’re building on sand.
What is the two-ledger problem?
Every economy runs two sets of books. The financial ledger tracks prices, GDP, and returns – the numbers on your brokerage app. The material ledger tracks physical inputs: tons of copper, liters of water, megawatt-hours. For 45 years, the West optimized the financial ledger and largely ignored the material one. The AI buildout is forcing a reckoning because you can’t train a model without electricity, and you can’t generate electricity without materials someone else controls.
How much energy does AI actually use?
A lot more than the “cloud” metaphor suggests. The IEA projects data center electricity demand could approach 945 TWh by 2030 – roughly Japan’s entire consumption. Meta’s planned facility in Louisiana alone will consume more than twice the electricity of New Orleans. Each GPU server rack is a small mining operation’s worth of refined metals, assembled and powered continuously. The heaviest metabolism in the industrial system, as investor Craig Tindale put it.
Why does China control rare earth supply chains?
Refining, not mining. Even when ore comes from Australia or Canada, it ships to Chinese refineries because the West lacks processing capacity and can’t build it profitably under current regulatory frameworks. China produces at a loss indefinitely because the goal is strategic control, not profit. Western investors want 20-30% returns. China wants leverage.
What is the “golden screw” problem?
In a complex system, one missing component breaks the whole machine. Craig Tindale couldn’t drive his car for nearly 12 months during COVID because of a single unavailable part. Everything else worked. Didn’t matter. AI’s supply chain has these single points of failure everywhere, and most run through China’s refining infrastructure.
What is GDP actually for?
That’s what the article asks. For 45 years, the West’s answer has been “whatever markets decide” - price signals and comparative advantage. The problem: that assumes everyone is playing the same game. State capitalism doesn’t optimize for market efficiency. It optimizes for strategic control. GDP that measures consumption without tracking whether you own the means of production is a number that flatters you right up until it can’t.
Last updated: May 2026.