Vasić takes it. With a pawn — the pawn standing guard in front of his king.

One move later he is checkmated. Two bishops, crossing diagonals, nothing to be done. Eleven moves, one of the most famous miniatures ever played. And here is the detail that matters: the mate wasn’t possible until Vasić captured. The pawn that took the queen was the pawn defending the king. Winning the material is what opened the door.

Vasić did not lose despite capturing the queen. He lost because he captured it.

Companies do this. They spend years capturing queens — defending the margin, winning the quarter, protecting the franchise — and every capture is real, countable, applauded. Each one pulls a defender off the square that matters. Winning the short-term battle is precisely what destroys the edge for winning the game.

On Tuesday, John Ternus takes over Apple. He inherits the most profitable hardware company ever built, a stock still near its highs, and the one job no board likes handing out while the scoreboard is green:

Stop being excellent at the thing that made you rich.

That sounds insane. It is also how every serious corporate reset actually starts.

Not with a keynote. Not with a “bold new chapter.” With a company agreeing, in public, to get worse at what it was proudest of.Winning is the trap

Great companies do not fail because they get stupid. They fail because they stay competent for too long.

The product still ships. The margins still print. The people who rose by defending the old game are still in the room, still getting paid to defend it. Competence hardens into identity. Identity does not read a balance sheet.

So the company keeps doing the thing that used to work. Inside, that looks like discipline. Outside, it is a slow death that the stock notices last.

Boards almost never interrupt this on purpose. They interrupt it when the market forces them to. Rolls-Royce had lost half its value before Erginbilgic. GE was down 63 percent before Culp. Intel, 65 percent before Tan. Meta, 77 percent, and Netflix, 76 percent, before their 2022 reversals. A crash — or a lost decade. A price that stops clapping.

That crash is not noise. It is permission.

The three-step reset

Every turnaround that mattered follows the same sequence.

1. Permission. The market withdraws consent. Sometimes that is a violent drawdown. Sometimes it is fourteen years of a flat stock while profits look fine — Microsoft after 1999. Either way, the board finally hears a sentence it cannot spin: the old game is over.

2. Unlearning. A new leader names the sacred strength that became the weakness, says it out loud, and stops rewarding it. Then the company relearns in one of three ways:

• It remembers rules it stopped obeying.

• It reads rules someone else already wrote.

• It builds a new game where the old strength no longer matters.

3. Recognition. The stock shows up late. That delay is the only edge a patient investor still gets.

If you remember one line, remember this: watch who gets promoted, not the strategy deck. A new story with the old bonus plan is a press release.

The proof is not subtle

Satya Nadella did not save Microsoft with a product. He saved it with an admission. “Know-it-all” became “learn-it-all.” Office shipped on the iPad before Windows tablets had a proper touch version. He said “Microsoft loves Linux” in public — about the software Steve Ballmer had called a cancer — and then paid to join the Linux Foundation. He did not invent the cloud. Amazon did. He saw that Windows had gone from moat to wall, and he stopped paying people to defend the wall.

Tufan Erginbilgic did the opposite at Rolls-Royce. He did not invent a new industry. He told 40,000 people they were standing on a burning platform, then attacked the thing the culture worshipped. Winning the engine slot had become the point, even when the service contract lost money on every hour the engine flew. He renegotiated the bad deals, walked away from customers who would not pay, and made cash per flying hour the job. Engineering stopped being the religion. It became a tool.

Steve Jobs in 1997 did not try to win the PC war. He left it. That is the third path, and it is the one boards hate most, because it takes years and looks like failure in the middle. Pat Gelsinger tried a version of it at Intel and ran out of board patience before it paid.

Notice what these cases share. Someone named the sacred cow. Then they changed the reward. The stock eventually noticed. It never noticed first.

The valley does not predict the ending

This is the part people get wrong.

Microsoft’s reset was almost invisible on the income statement. There was barely a valley. GE’s reset meant selling half the company; shareholders got crushed in public before the pieces were re-rated. Both worked. A deep drop does not mean the plan is right. A quiet chart does not mean nothing is happening.

The market can price competence. It cannot price unlearning. Unlearning looks like lost revenue and a quieter ego long before it looks like a multiple.

That gap is the whole trade.

Six questions that beat the earnings call

You will not see a reset in the numbers for years. You can see it in the person within ninety days.

1. Did they name the strength that became the weakness — in plain words? “Know-it-all.” “Burning platform.” If they call a hard inheritance “an exciting opportunity,” they have already wasted a year.

2. Have they ever publicly reversed a decision they owned? If not, they will defend the sunk cost until the board fires them.

3. Did they change who gets paid — not the slide about “priorities”?

4. Did they shorten the feedback loop, or bury it in a longer story?

5. Does their runway match the type of reset? Two years of patience for a six-year rebuild is not a plan. It is a countdown.

6. What is the market pricing them as? A rule-reader priced like a visionary is a trap. A game-maker priced like a caretaker is the opportunity.

Screenshot that list. It travels farther than another CEO profile.

Now the uncomfortable part

The class of 2026 is full of insiders. Ternus at Apple. Josh D’Amaro at Disney. Michael Fiddelke at Target. Luke Miels at GSK.

Insiders are the correct hire when the game has not changed and the problem is execution. They are the dangerous hire when the game has changed, because the insider is the person who won the old game.

Apple is the extreme case.

Tim Cook ran the old game close to perfectly for fifteen years. The machines. The mix. The services harvest. The unblemished object. Ternus helped build that machine. Cook is not leaving the building; he becomes executive chairman. The stock is not begging for a new religion.

So the question is not “Can Ternus run Apple?” Of course he can. The question is whether he can make Apple worse at the things it is addicted to being good at:

• the closed system as identity

• the next iPhone as the whole company

• the reflex to ship the beautiful object and let the platform catch up later

If he talks like the steward of a peak franchise, there is no reset. Enjoy the compounding. Do not wait for a conversion narrative.

If he names a sacred strength in public and changes who gets promoted, the stock will lag the change. That lag is the story.

If a winning multiple plus the old CEO still in the room blocks both, that is not a plot twist. That is the model. Companies almost never grant permission while they are still being paid to stay the same.

The sentence that should travel

A turnaround is not a recovery.

A recovery is getting good again at what you already were. A turnaround is agreeing to be bad at something for a while so you can become good at something else.

The price will tell you when that worked.

It will never tell you when it started.

Tuesday is not the start of Apple’s next product cycle. It is the first day you can watch whether Ternus is allowed to make the company worse — or whether Apple has decided that winning is still the plan.

Thanks for reading,

G

Three numbers. Remember them.

0.8 — South Korea’s fertility rate. The lowest ever recorded by any society, anywhere, ever.

930 million — the number of humanoid robots Morgan Stanley projects will be working by 2050.

2046 — the year those two curves collide.

Around 2046, humanity will manufacture more robots per year than it gives birth to humans.

I call it The Crossover. Once you see it, you can’t unsee it. And almost everything you believe about the economy — inflation, interest rates, what’s safe, what’s risky — was built for the world before it.

Let me show you.

Church bells, twice

April 12, 1955. Church bells rang across America.

Not for a war. For a vaccine. Polio was beaten.

The vaccine was paid for with children’s dimes — coins collected door to door by the March of Dimes. But it was born in a country the Cold War had rewired: full of young scientists, full of fear, treating research like survival.

Fear built the machine. The machine produced the miracle.

Seventy-one years later — August 19, 2026 — it happened again. Merck and Moderna announced the first cancer vaccine in history to pass a final-stage trial. Personalized mRNA. Melanoma’s return, cut roughly in half.

Same story, one octave higher: a technology forged in an emergency (mRNA came out of the pandemic — this era’s mobilization), funded by a new arms race, delivering the medical miracle of a generation.

Cold wars are terrifying. They are also the most productive fear humanity has ever invented. The bomb gave us the moon landing. The pandemic gave us the cancer vaccine.

Why? This isn’t mysticism. It’s mechanics — and the scholars have mapped it.

Why fear works

Three reasons existential threat produces miracles, and comfort produces decay:

1. Fear breaks the blockers. Economist Mancur Olson spent his career proving an uncomfortable law: the longer a society is stable and safe, the more it fills up with interest groups, lobbies, and cartels whose job is to protect what exists — and block what’s next. He called it institutional sclerosis. Stability is how nations decay. Postwar Germany and Japan didn’t boom despite catastrophe; they boomed because catastrophe had wiped their blockers away. An existential rival does the same thing without the rubble: when the other side might win, nobody can afford the veto players anymore. Sputnik didn’t just launch a satellite. It launched a purge of American complacency.

2. Fear funds what markets won’t. Economist Vernon Ruttan studied the six technologies that built the modern economy — mass production, aviation, nuclear power, computers, the internet, space. His finding: every single one was developed slowly, or not at all, until military necessity paid the bills. No venture capitalist funds a thirty-year bet with no customer. A rival superpower does. The fear signs the checks that patience never will.

3. Fear sets a benchmark you can’t fake. Historian Arnold Toynbee read the rise and fall of twenty-plus civilizations and found one pattern: civilizations grow through challenge and response — and die from two things: no challenge at all, or a challenge too big. A rival is the perfect-sized challenge. It’s a mirror that can’t be lobbied, spun, or bought. Either your chips are faster or they aren’t. Either your robots work or they don’t. Comfortable societies grade their own homework. Cold war societies get graded by the enemy.

That’s the engine: the threat breaks the blockers, funds the impossible, and keeps the scoreboard honest. Take the threat away and the sequence runs backward — which is roughly the story of the West between 1991 and 2022. The “end of history” wasn’t a victory lap. It was thirty years of accumulating blockers, unfunded moonshots, and self-graded homework.

Now hold that thought — because the engine behind the miracles wasn’t really the fear.

It was the people.

The one number that rules everything

Economic growth sounds complicated. It isn’t.

It’s three questions: How many people are there? How old are they? What can they do?

People are demand — every baby becomes eighty years of buying: diapers, then a house, then a car, then a pension, then a hospital bed. People are also ideas — innovation comes from millions of young brains competing.

The first Cold War proved it. America’s 76 million baby boomers built the suburbs in the 50s, caused the inflation of the 70s — a whole generation demanding everything at once — and fueled the great bull market when they started saving. The polio vaccine itself was demographic: a society overflowing with young scientists beat the disease in a decade.

The baby boom was the fuel. The Cold War was just the engine.

The fuel ran out this year

Here’s the most important economic fact of 2026, and it appeared on no news channel:

Humanity, for the first time ever, is probably having too few babies to replace itself.

Not Japan. Not Europe. The species.

China: 0.9 children per woman. South Korea: 0.8. Colombia: about 1. Italy just had its fewest births since the days of Leonardo da Vinci. World population now peaks around 2056 — then falls.

Every government that tried to pay people to have babies has so far failed. The cause isn’t policy. It’s modern life itself.

Cold War I was fought with an abundance of young humans. Cold War II will be fought with a shortage of them.

Subscribe now

The new babies have serial numbers

My thesis in one line:

What the baby boom was to the first Cold War, the robot boom will be to the second.

The new generation isn’t born in hospitals. It rolls off assembly lines in Shenzhen and Fremont.

Look at its birth rate. 2025: about 15,000 humanoid robots shipped worldwide. This year: heading toward 100,000. By 2050: 930 million at work.

Don’t read that as a tech forecast. Read it as a demographer. From 15,000 to nearly a billion in 25 years isn’t a market. It’s a birth rate.

Every robot is a worker who never ages, never retires, and works three shifts. The first country to deploy them at scale buys itself a second baby boom. That’s why China subsidizes robots like it subsidized electric cars. It’s not industrial policy.

It’s population policy.

But robots don’t buy houses

Fair objection. Boomers built things AND bought things. A robot never buys a home. So who consumes?

Wrong question. Robots consume plenty — just different things:

Compute is their education. Electricity is their food. Actuators are their bones. Repairs are their healthcare. Data is their teacher.

Right now, in warehouses across China, hundreds of humanoid robots stand in rows learning to sort packages and change diapers — each guided by a human with a game controller. The robots are in kindergarten. Their teachers are people.

Every generation’s shopping list creates that era’s winning industries. The boomers’ list built the suburbs, Detroit, and Wall Street. The robots’ list is: Robotics. AI. Compute. Electricity.

And here’s the twist nobody sees coming: robots might bring babies back. Paying people to have kids has failed everywhere because the payments were tiny against the real cost. A robot economy is the first economy rich enough to pay parents properly. The robot works the night shift so raising a child stops being a financial sacrifice.

Share

The two mistakes that will cost you money

Mistake 1: “High interest rates kill stocks.” No. High rates kill bonds. During Cold War I, rates climbed from 2% in 1949 to almost 16% in 1981 — thirty years of slow ruin for bondholders. Stocks in the same period? Up more than 10x. In a cold war, the winners feed on scarcity: commodities, energy, tech. The long bond isn’t the safe asset of this era. It’s the sacrifice.

Mistake 2: “Technology is deflationary, so inflation can’t come back.”

True — but only at the end of the story.

Every big technology lives two lives. First, chaos: hundreds of competing designs, everything expensive, devouring capital, copper, and electricity. Then one design wins, costs collapse, and THEN comes deflation. Computers spent thirty years in chaos before the IBM PC won in 1981 — and only then did computing get cheap, exactly when inflation broke. Robots today are pure chaos phase: 140+ manufacturers in China alone. What the computer was to the 1980s, the robot will be to the 2050s. Between now and then: scarcity, disruption, and very possibly another 1970s-style inflation shock.

Deflation is the prize at the end. Inflation is the toll at the beginning. Confusing the order is confusing 1972 with 1999.

A child born this year

Make it personal. A child born in 2026 turns twenty at The Crossover.

She will enter the first labor market in history where her manufactured coworkers outnumber her birth cohort. Her salary will depend on doing what robots can’t yet do — and on teaching them. Her pension will be funded, or not, by machines that never retire. Her cost of living will track the price of electricity the way her grandparents’ tracked the price of oil.

She isn’t competing against the robots. She’s inheriting them. The only question is who owns them by the time she does.

Where does it all end?

In the 2050s, when a billion tireless workers flood a shrinking planet with output — who absorbs it all?

On Earth: maybe nobody. Off Earth: everything.

That’s the real economics behind Musk’s Mars and Bezos’ space factories. The frontier is infinite demand, and for the first time in history, expansion doesn’t depend on human birth rates — because the pioneers are manufactured. Robots build the base. Humans arrive as colonists.

Apollo was the symbolic frontier of the first Cold War. The second one ends by opening a real one.

What to actually do with this

You don’t need a prediction. You need three questions you can ask forever:

1. Which phase are we in? Opening (scarcity building), middle game (inflation risk peaks), or end game (deflation arrives)? Today: the opening. Position for scarcity, not for 2015.

2. What does the new cohort buy? Whoever sells to the dominant generation wins the era. The robot generation buys compute, electricity, components, repairs, and data.

3. Who is selling it to them? Not “which robot company wins” — nobody knew in 1975 that IBM would define the design and Microsoft would capture it. Own the shopping list, not the lottery ticket.

The compass in one line: own what robots consume; avoid whatever promises fixed payments for decades.

Three predictions, on the record

So you can check my work:

1. Long rates stay in a rising channel for at least a decade. The bond bear market of Cold War II has already started.

2. An inflation crisis hits in the middle game — the 2030s — driven by electricity, copper, and geopolitical shocks, while everyone still quotes “technology is deflationary.”

3. The robot’s IBM PC moment — one dominant design — arrives by the early 2050s. Then, and only then, the great deflation begins.

If all three are wrong, this essay was entertainment. If two are right, it was the map.

Demography is still the master constraint on everything. Only one thing changed, and it’s the whole story:

For the first time in history, the birth rate is a capital expenditure decision.

Thanks for reading,

G

Five things this essay will convince you of:

• The singularity has a math problem. Nothing that grows this fast ever reaches infinity — not bubbles, not populations, not empires. The physics says AI’s curve doesn’t go vertical around 2050. It forks.

• Intelligence is not compute. It’s a gap. A system is intelligent when its ideas grow faster than its energy bill. Your brain does it on 20 watts. Today’s AI doubles its power bill to double its output — all engine, no gap.

• Wealth is stored network density. No head and no company can build a jet engine — only a network can. And over the long run, a country’s income converges to the complexity of its network. Intelligence and prosperity are the same variable.

• The AI industry just confessed — twice. A 264-page report from twenty institutions admits today’s models are structurally incomplete. And Dario Amodei admits scaling structurally concentrates power. Both point to the same fix: architecture and institutions, not more scale.

• The market bought the engine at 79 cents on the dollar. Models and chips are 79% of a $31 trillion stack; energy is three cents, the maps two cents. Jensen’s $100 trillion is real — it just lives in the rest of the car.

Your brain runs on twenty watts.

Less than the bulb in your refrigerator. On that budget, it holds a hundred trillion connections, rewires itself every night, and occasionally produces a symphony, a theorem, or a decent joke.

The most advanced AI datacenter on Earth burns a hundred million watts. It cannot tie a shoelace.

Sit with that for a second. Because everything the smartest people in the world are telling you about intelligence — the charts, the timelines, the singularity — quietly ignores it.

The Chart Everyone Worships

There’s a famous chart. Ray Kurzweil drew it. It shows one number — calculations per second per $1,000 — getting cheaper, decade after decade, in one smooth exponential line. Follow the line far enough and it goes vertical. Kurzweil gave that moment a name — the singularity — and a date: around 2045.

The whole AI industry runs on some version of this faith. More chips, more data, more scale. The line goes up. Intelligence falls out. The metric of the era is tokens per watt: how many words the machine produces per unit of electricity.

But look closer at the chart. That “one smooth line” is actually drawn through five completely different machines: gears, then relays, then vacuum tubes, then transistors, then chips. Nobody got from gears to chips by building bigger gears. Every leap came from throwing the old architecture away.

The chart’s own history refutes the chart’s message. Progress didn’t come from more. It came from different.

Nothing Grows Forever

Here’s what a physicist sees when he looks at that same acceleration.

Didier Sornette spent his career studying how things break — bridges, markets, seizures. He found a signature: when something grows faster and faster, feeding on its own growth, the math points to a moment when it should hit infinity.

And here’s the thing every physicist knows: it never gets there.

Bubbles don’t reach infinity. They pop. Populations don’t reach infinity. They level off. Empires don’t reach infinity. They fall or they transform. The approach to the vertical line isn’t a promise of transcendence. It’s the warning that the whole system is about to change shape.

Sornette ran the numbers on human population and the world economy. Both have grown faster than exponentially for most of recorded history — and the critical point lands around 2050. But the world’s population growth rate already peaked. In 1970. Read that again. We are not waiting for the great transition.

We are already inside it.

The question was never when does the line go vertical. The question is: when the system changes shape, what survives?

Intelligence Is a Gap, Not a Number

A physicist named Geoffrey West left particle physics to study living things — and found something strange. Double an animal’s size, and it doesn’t need double the energy. It needs about 75% more. Nature gives a discount for scale. Every creature on Earth, across twenty-one orders of magnitude, gets roughly the same deal — because every creature is built on the same trick: efficient networks for moving energy around.

Then West measured cities, and found the miracle. Double a city, and it saves about 15% on pipes and roads per person — and produces 15% more ideas per person. Wages, patents, inventions. All up.

Same energy discount as biology. Plus a bonus biology never earned.

Where does the bonus come from? Not from bigger pipes. From more conversations. From density of connection. A city is a machine for making people collide.

So here is a better definition of intelligence than anything on Kurzweil’s chart: a system is intelligent when its ideas grow faster than its energy bill. Ideas up 15% per person, infrastructure down 15% per person, at every doubling. That widening gap is the whole game.

Your brain lives in the gap, and it’s worth understanding why, because the reason is the AI industry’s deepest problem. In a computer, memory and processor are separate devices, and moving data between them costs far more energy than computing on it. Every thought pays a shipping fee. In your brain, there is no shipping. The synapse is the memory, the processor, and the wire — one device. Run the numbers: twenty watts across a hundred trillion connections, versus a hundred megawatts across a trillion parameters. Per connection maintained, biology is roughly a billion times more efficient.

That’s not a gap Moore’s Law closes. That is a gap only architecture closes.

And it explains the deepest design principle in nature: it is better to have maps than massive processors. A map is computation you did once and stored where you’ll need it. The animal that remembers the territory doesn’t have to recompute the territory every time it crosses it. Memory, well organized, is the cheapest form of intelligence ever invented.

Today’s AI has none of this. Double the answers, double the power bill. All engine, no gap.

A Jet Engine Fits in No One’s Head

Now zoom out. Because the law that runs your brain also runs the economy — and a physicist-turned-economist named César Hidalgo proved it with one observation.

There’s a limit to how much knowhow fits in one head. He called it a personbyte. Nobody — nobody — knows how to build a jet engine alone. Not the metallurgy and the aerodynamics and the software and the supply chain. A jet engine doesn’t live in a person. It doesn’t even live in a company. It lives in a network — thousands of people, wired together by trust and trade, each holding a piece.

The complexity of anything you make is the size of the network it takes to make it.

Then Hidalgo did what economists rarely do: he measured it. With Ricardo Hausmann, he ranked every country by two things — how many different products it can make, and how rare those products are. Diverse and rare means dense knowhow. They called it economic complexity, and their studies found it predicts long-run growth better than education levels, institutions, or capital.

And the punchline, in Hidalgo’s own words: over the long run, a country’s income approaches the complexity of its economy. Not its oil. Not its slogans. Its network.

Wealth is stored network density. Intelligence and prosperity are the same thing, at different sizes.

Four Ways to Organize Anything

So if the network is the asset, how do you build the right one? An economist named Ronald Coase found the rule ninety years ago, and it’s brutally simple: the right structure depends on what you can measure. Picture it as a board with four squares.

When you know the goal and the rules — a factory, an army — bosses win. Command is fast when the job is clear. And notice something: this is the only square where “tokens per watt” even means anything. You can only optimize a score when the game is defined. Today’s AI lives here — one clear goal, predict the next word — which is why it’s run by a handful of giant, top-down companies. Right structure, right square.

When the rules are clear but nobody knows what’s worth wanting — that’s what markets are for. Prices let a million strangers figure it out together. The next wave of AI lives here: agents trading ideas, services, and verifications with each other. Silicon rediscovering commerce.

When you know what you want but not how the world works — a cure, a material, a discovery — networks win. Many explorers, in parallel. Most fail. One find changes everything, and the network spreads it fast while pruning the dead ends. Science works this way. So will the AI that matters: robots, sensors, machines that touch the real world. This square is the prize.

And when you can measure nothing — no goal, no rules, no odds — humanity never invented an algorithm. It invented trust. An economist named William Ouchi proved it in 1980: when performance can’t be measured, neither prices nor rules work, and groups fall back on the oldest technology — shared belief, ritual, belonging. Religion. Insurance. The state. Structures that don’t solve the unknown. They make it survivable.

Bosses, prices, links, trust. As certainty falls, coordination changes hands. Each wins on its own ground. And here’s the trap almost everyone falls into: winning in one square never buys you the next. You cannot command your way to discovery. You cannot scale your way to trust.

The Confession

People’s real beliefs are revealed not by what they preach, but by what they hedge.

So watch what just happened.

In 2025, forty-seven researchers — from Google DeepMind, Microsoft, Stanford, Yale, twenty institutions in all — published a 264-page report on the future of AI. Strip away the careful academic language and it reads like a confession: the models we have are extraordinary, and structurally incomplete. No eyes. No lasting memory. No mental map of the world. No goals of their own. No each other.

And their blueprint for what comes next? Sensors. World models. Memory that doesn’t reset. Systems that prune their own failures. Teams of agents. Stop making the engine bigger. Build the rest of the car.

That was the engineering confession. The political one cuts deeper.

No one on Earth has more riding on the gospel of scale than Dario Amodei. He runs a frontier lab. His entire position rests on the line going up. And yet, pressed in a public debate, he wrote it plainly: AI is “structurally a technology that tends to concentrate power.” Not because of bad regulation. Because of the scaling laws themselves. Bigger models mean power pools with whoever owns the chips and the electricity.

He even dismantled the industry’s favorite escape hatch. Open-source the models, people say, and the power spreads. No, he argued — it just moves to whoever holds the most compute. And of course it does. Think about what a model actually is: information. Information is the one thing in the universe you can copy for free. Chips and gigawatts, you cannot. Free the software and the power simply pools around the hardware and the electricity. Copying the map doesn’t give you the territory.

His solution? Not more scale. Rules. Referees. Courts. Institutions that “vest power in ideas rather than people.”

Understand what you just witnessed. The man at the top of the pyramid is reaching for the one tool a pyramid cannot produce: trust. Because he knows what the cheerleaders don’t — a company can be captured. A fair rule is much harder to capture.

When the high priest starts hedging, pay attention to which god he’s hedging on.

The network wins again.

The Market Hasn’t Read This Essay

Here’s where it gets practical.

Add up the entire AI capital stack — every layer, from the power plants at the bottom to the chatbots at the top. Call it $31 trillion.

Now sort it by square.

The command square — the giant models and the chips that feed them — is worth $25 trillion of it. Seventy-nine cents of every dollar. The market has priced one square of the board as if it were the whole board. Priced, in other words, for the line that goes vertical.

And the layers that build the gap? The cables and switches that connect everything: ten cents. The robots and sensors — the eyes and hands: five cents. The mental maps: two cents. And energy — the thing every metric in this debate is divided by, the hard physical constraint on all of it, the one input that cannot be copied: three cents.

Three cents.

The market is pricing intelligence as if electricity were a rounding error. Everything in this essay says electricity is the denominator of the whole thing.

Now listen to Jensen Huang — the man selling the shovels. He keeps calling AI a $100 trillion opportunity: the size of the world’s industries. Lately he goes further — that global GDP isn’t capped at $100 trillion at all, that AI takes it to two, three, five hundred. People hear salesmanship. Run it through this essay and it’s arithmetic, with one correction.

To touch $100 trillion of industry, intelligence has to leave the datacenter. It needs a mental map of every factory, hospital, port, and farm — the knowhow of the entire economy, translated into something a machine can act on. It needs eyes and hands on every floor. It needs power everywhere, and cables between everything. Jensen is right about the size of the prize. He’s just quiet about its location. The engine doesn’t get you to $100 trillion.

The rest of the car does.

Which makes the maps — priced at two cents — maybe the most mispriced layer of all. The map is where a prediction becomes a decision, and a decision becomes money. Two cents on the dollar, for the memory of the entire economy.

Even crypto tells the story. Today it sits outside the physical loop entirely — $2 trillion priced as digital gold and speculation. But look back at the markets square. If millions of AI agents are going to trade with each other, they’ll need something no bank was built for: machine-native identity, property, and settlement. Will open crypto rails win that job? Nobody knows. That’s exactly the point — the current price assumes they won’t. You are being paid to own the maybe. Small position, bounded loss, unbounded if it hits. That’s not a bet. That’s an option.

So the trade writes itself. Fade the square priced for infinity. Own the squares priced for irrelevance.

The market bought the engine at 79 cents on the dollar — and left the rest of the car on sale for 21.

What To Do With This

So forget the vertical line. Nobody is going to hand you a date for the future. Here is what the evidence actually says:

The curve forks around 2050. Position for the fork, not the fantasy.

The winners of a regime change are the systems in the gap — more ideas per watt. Own what feeds networks: energy, sensors, the physical layer that can’t be copied.

Watch complexity, not headlines. A country whose network is densifying — making more things, rarer things — is getting rich, whether or not anyone has noticed yet. And watch which countries build fair rules and referees fastest, because that’s the speed at which a society can rewire its trust — and the societies that rewire fastest keep their networks open through the turn.

And in your own life, the same law holds. You are not your engine. You are your network — what you can sense, what you remember, who you trust, and how fast you prune what no longer works.

Twenty watts built everything you love.

It didn’t do it with scale. It did it with connection.

Thanks for reading,

Guillermo

The Big Idea

• Every industrial resurrection follows the same six-phase script: glory, denial, collapse, an outsider, surgery, rebirth. GE ran it in 25 years; Intel is running it in five.

• Bottlenecks migrate down the stack — each wave’s constraint becomes the next wave’s bull market. The constraint has now reached electricity, and the market’s own scoreboard proves it hasn’t finished pricing that.

• Eleven companies are worth over a trillion dollars. All of them consume gigawatts. None of them produce electrons. That empty chair is the trade.

Now the story.

There is a moment in the life of every great industrial empire when the market stops pricing its future and starts pricing its funeral.

For General Electric, that moment came in June 2018, when the company that Thomas Edison founded — the most valuable corporation on Earth as recently as 2000, worth some $600 billion — was expelled from the Dow Jones Industrial Average after 110 uninterrupted years. The market had rendered its verdict: GE was a zombie. A balance sheet with a logo attached.

For Intel, the moment came in 2024. A $19 billion annual loss — its first since 1986. Fifty billion dollars of debt. A CEO shown the door in December. A stock that would eventually touch $19.60, a level that priced the inventor of the microprocessor as a going-concern question rather than a going concern.

Eighteen months later, Intel traded above $140 — up roughly 490% at the peak — with the United States government as its third-largest shareholder, Nvidia and SoftBank on the cap table, its 18A process in high-volume manufacturing in Arizona, and a handshake with Elon Musk to build the most ambitious chip factory ever conceived.

Zombie to rockstar in five quarters.

I have seen this movie before. We all have. It was called General Electric, it starred Larry Culp, and it took five years to reach the same act Intel compressed into sixteen months. The plot is nearly identical. What changed is the era in which it is being filmed — and that difference is the entire investment thesis.

The Map: What RACE Means

Before we walk the arc, a note on the map this essay is drawn on, so the piece stands on its own.

RACE is the MacroWise framework for the physical buildout of the AI age: Robotics, AI, Compute, Electricity. The four letters are not a portfolio of themes — they are four layers of the same machine, and capital rotates through them as successive waves:

• R — Robotics. The demand layer: machines that act in the physical world. Humanoid robots, robotaxis, autonomous defense, drones. This is where AI stops being software and starts having a body.

• A — AI. The intelligence layer: the models, and the silicon that trains and runs them. This is the wave everyone can see.

• C — Compute. The manufacturing layer beneath the intelligence: the fabs, packaging, interconnects, and substrates that turn sand into reasoning. Whoever controls leading-edge fabrication controls the ceiling of the entire stack.

• E — Electricity. The physical layer beneath everything: generation, turbines, transformers, grid. Every token inferred and every robot step taken is, at bottom, electrons doing work.

The organizing insight deserves a name, because it does most of the work in this essay. Call it the Law of the Migrating Bottleneck: each wave’s constraint becomes the next wave’s bull market. When models became scarce, capital flooded to AI. When chips became scarce, it flooded to compute — that is how a chipmaker became the most valuable company on Earth. The scarcity is now migrating down to the final layer: the electron. E is not the last letter of the acronym. It is the binding one — the layer that cannot be compressed by cleverness, only built with concrete, copper, and time.

Read the Intel story through this map and it stops being a stock story: Intel is the rare company that touches all four letters at once. That is the thesis. Now, the arc.

The Six-Phase Script

Strip away the specifics and every industrial resurrection follows the same six-phase arc. Call it the phylogeny of a turnaround.

Phase I: Glory. GE in 2000 was the management bible of an era; Jack Welch was its prophet. Intel from 1993 to 2020 was computing — “Intel Inside” a temporal monopoly on both x86 architecture and leading-edge fabrication.

Phase II: Denial. This is the phase that kills. GE Capital’s financial engineering masked the decay of the industrial core for a decade. At Intel, the rot was cultural: a bureaucracy that rewarded managers for reporting good news while TSMC quietly took the fabrication crown and the AI wave broke somewhere else entirely. This is learning atrophy at institutional scale — the organization forgetting faster than it learns, while the stock chart still looks respectable.

Phase III: Symbolic collapse. The Dow expulsion for GE. The $19 billion loss for Intel. The market needs a funeral before it permits a resurrection.

Phase IV: The outsider. GE reached outside its own bloodline for the first time in 126 years and hired Larry Culp, the Danaher operator. Intel’s board hired Lip-Bu Tan — the Malaysian-born engineer and venture capitalist who had already rebuilt Cadence Design Systems. Both men share a defining trait: they are allergic to corporate theater. Tan’s first cultural edict, delivered at Stanford this spring, could have been written by Culp: “If you tell me the bad news, that means it’s our problem. If you don’t tell me, then this is your problem.”

Phase V: Surgery. Culp cut the dividend to a penny, sold biopharma to his old firm for $21 billion, and ground down more than $100 billion in debt with Danaher-style lean discipline. Tan moved faster and cut deeper: management layers halved, 34% of the global workforce gone, Altera sold for $8.75 billion, the Mobileye impairment swallowed whole. Everything sacrificed to one altar — engineering and fabrication.

Phase VI: Rebirth. And here the scripts diverge in the most instructive way possible.

Fission versus Fusion

Culp’s masterstroke was fission. He split GE into three focused companies — HealthCare in 2023, Vernova in 2024, Aerospace as the remainder — and let each piece command its own multiple. The conglomerate discount became a specialization premium. GE Aerospace became the rockstar of the aviation supercycle.

Tan’s masterstroke is the opposite: fusion. Rather than breaking Intel apart, he is fusing it with the American state and the Musk industrial complex.

Consider what has attached itself to Intel in the past year. The U.S. government converted CHIPS-era support into an equity stake of roughly 10% — about $11 billion — making Washington the company’s third-largest shareholder. Nvidia and SoftBank made capital infusions. Google signed a multi-year ASIC and Xeon agreement. Apple and Tesla are reportedly in preliminary manufacturing talks. And on April 7, Intel announced it would join Terafab — Musk’s project with Tesla, SpaceX, and xAI — as the foundry partner.

Why the difference? Because Culp operated in the world of private capital and market discipline, where the job of a turnaround CEO is to repair a balance sheet. Tan operates in the era of sovereign capitalism, where a leading-edge fab is not a business unit. It is, in Tan’s own words on CNBC, “one of the key national treasures.”

Culp repaired a balance sheet. Tan is repairing a balance sheet, a 1.8-nanometer process node, and American technological sovereignty simultaneously. That is why the market paid him in sixteen months what it took GE five years to earn.

The Vernova Proof: The Bottleneck Moved to the Electron

Before we get to Terafab, pause on what GE’s fission actually produced — because one of its three children turned out to be the purest expression of the thesis this entire series is built on.

When GE Vernova was spun off in April 2024, it was the unloved child of the trisection: the “problematic part” of old GE, a gas-turbine and grid-equipment business that the late-2010s consensus had left for dead, convinced that wind and solar would structurally strand it. The market priced it accordingly.

Then the market did its homework. Building AI data centers, it turned out, is not primarily a software problem or even a chip problem — it is an energy problem, and it is bigger than anything else in the stack. Hyperscalers suddenly needed astronomical amounts of immediate, reliable, around-the-clock electricity, and the world’s largest manufacturer of gas turbines — with a near-monopoly position alongside Siemens and Mitsubishi, plus the transformers and switchgear that move the power — was sitting right there.

The principle first: capital moves at the speed of a press release; energy infrastructure moves at the speed of concrete. A hyperscaler can commit $50 billion to data centers in one earnings call. A turbine takes years to build, a grid interconnection takes longer, and no amount of money compresses either. When two clocks diverge that violently, the slower one owns the cycle.

The mechanism is a queue. When demand arrives faster than physical capacity can respond, it doesn’t bid up output — there is no output to bid on. It bids up position in line. Customers are signing slot reservation agreements for turbine deliveries stretching to 2031; management expects to sell more than half of its 2031 production slots before this year ends. The queue for the machines that make electricity is now five years long.

The evidence is the scoreboard. From roughly $115 at the April 2024 separation, GEV crossed $1,000 in 2026 and touched an all-time high near $1,175 in late June — an 8x in twenty-six months, a ~$264 billion company built from the piece of GE nobody wanted. 2025 earnings tripled, and management guides them to double again between 2026 and 2028. This is the Law of the Migrating Bottleneck showing up in a stock price: the constraint reached the electron, and the market repriced the company that sells the machines that make them.

Even GEV’s July stumble confirms it. The stock fell 9% on a modest earnings miss and a $100–200 million tariff warning — while orders and backlog kept growing and revenue guidance was raised by a billion dollars. The market flinched at the accounting; the physics didn’t move.

The Empty Chair in the Trillion-Dollar Club

Now look at the anomaly hiding in plain sight at the top of the market. Eleven U.S.-listed companies are worth a trillion dollars or more:

Read the list as a physicist rather than an analyst. Chips, chips, chips, clouds, clouds, ads, models, robots-to-be, a holding company, and a weight-loss molecule. Every technology company on this list is a voracious consumer of electricity — collectively they are contracting for gigawatts on a scale that resembles the electrification of a mid-sized country. And yet there is not a single company on the list that makes electrons. The club consumes terawatt-hours; it contains no one who produces them.

That is not a permanent feature of markets. It is a lag — and here the Law of the Migrating Bottleneck stops being a description and becomes a testable prediction. The trillion-dollar club is a map of what capital believed the constraint was: software, then chips. Once compute became scarce, the club seated a chipmaker at the head of the table. If the constraint has genuinely moved to energy, the club must eventually seat an electricity company. The chair is empty because the repricing is not finished.

Two candidates are converging on that chair from different directions:

GE Vernova, from the supply side. At roughly $264 billion, it is already the largest pure-play electricity-equipment company in the world — and it needs about a 4x to enter the club. That sounds aggressive until you remember it just did an 8x in twenty-six months, management expects earnings to double again by 2028, and its order book is sold out half a decade forward. Vernova entering the trillion-dollar club would be the market’s formal admission that the AI trade is an energy trade.

Intel, from the silicon side. At roughly $508 billion — the world’s 26th most valuable company — Intel needs only a 2x. The path runs through exactly what this essay describes: 18A yields, formalized foundry customers, and Terafab moving from render to wafer. If the sovereignty thesis holds, Intel would enter the club not as a comeback story but as the American answer to TSMC — which sits at $2.1 trillion, four times Intel’s size, for doing the job Intel is trying to take back.

The symmetry is the thesis in miniature: one company sells the machines that make the electrons; the other fabricates the silicon those electrons feed. The trillion-dollar club currently prices the appetite. It has not yet priced the kitchen.

Hold that five-year turbine queue in your mind. It is the single most important fact for understanding what Musk is about to do in Texas.

Terafab: Where Physical AI Touches the Ground

Now to the part of the story that matters most for the RACE framework.

The first wave of the AI cycle was digital: tokens, chatbots, data centers. The wave now forming is physical — AI that walks, drives, flies, and manipulates the world. Humanoid robots. Robotaxis. Autonomous defense. Orbital data centers. And Physical AI has a property that digital AI could politely ignore: it cannot live in an abstraction. Every Optimus, every Cybercab, every AI satellite is a physical object built around leading-edge silicon that must be fabricated somewhere on the actual surface of the Earth.

Today, that somewhere is overwhelmingly an island 130 kilometers from the Chinese coast.

Terafab is the most ambitious attempt in history to move it to Texas.

Look at the renders for a moment, because they are not incidental to the thesis — they are part of it. This does not look like a fab. TSMC’s facilities are anonymous white boxes, and that is deliberate: Taiwan’s entire strategy is discretion, the silicon shield that works best when nobody thinks about it. Terafab is the opposite grammar. The structure reads like a starship, or an aircraft carrier laid across the Texas scrubland — architecture designed to be seen. The correct comparison is not another semiconductor plant; it is the Hoover Dam renderings of the 1930s. This is infrastructure as announcement: a civilization declaring, in steel and glass, that it has decided something.

The numbers behind the image read like science fiction: a target of one terawatt of AI compute per year — roughly 70% of TSMC’s entire current global output, from a single site. A pilot facility costed at $20–25 billion; Bernstein’s estimate for the full build runs as high as $5 trillion. Two plants on the Giga Texas campus: one dedicated to automotive and humanoid-robotics silicon — Tesla’s FSD, the Cybercab, the Optimus line — and a second for AI data-center infrastructure, including SpaceX’s planned constellation of orbital compute. Tesla’s AI5 chip is targeted for pilot batches in 2026 and volume in 2027. Musk was explicit on Tesla’s earnings call about the process technology: “We plan to use Intel’s 14A.”

Three structural features separate this from every fab announcement that came before it.

The demand is captive. This is not a foundry hoping customers show up. Tesla’s AI5, the Optimus line, the Cybercab, and SpaceX’s orbital compute constitute pre-committed demand from the same consortium that owns the plant. The classic fab-killer — building capacity into uncertain demand — is partially neutralized because the customer is the owner.

The energy is co-designed. In a detail most coverage missed, SpaceX announced it will build its own power plants for a $16.8 billion chip facility. Now recall the Vernova queue: when the wait for grid-scale turbines runs to 2031, the only way to guarantee electrons on your own schedule is to own the generation yourself. Every other fab on Earth is a supplicant to a grid; Terafab treats electrons as a design input rather than a utility bill. This is the E-bottleneck internalized — the first fab conceived from day one as an energy project that happens to output silicon. Vernova proved the bottleneck exists; Terafab is the first industrial project designed around it.

The stack is vertical without precedent. Design (Tesla), process (Intel 18A/14A), energy (SpaceX), demand (all of them) — under one roof. Musk’s stated ambition to combine logic, memory, and advanced packaging in a single complex violates forty years of industry orthodoxy. Then again, so did catching a rocket booster with chopsticks.

Intel is the only company on the planet that touches all four letters:

• R — a fab dedicated to the chips inside Optimus, Cybercab, and FSD.

• A — Xeon 6 selected as the host CPU for Nvidia’s DGX Rubin systems; custom ASICs for Google; inference silicon for xAI. Tan’s framing is worth quoting: the CPU is reasserting itself as “the orchestration layer and critical control plane for the entire AI stack.” After years of GPU monotheism, the control plane is coming back — and Intel owns it.

• C — 18A in high-volume manufacturing in Arizona; 14A as the Terafab target node. The only leading-edge logic process operated by an American company on American soil.

• E — fabs co-designed with their own generation capacity.

That is not a product story. That is an infrastructure-of-civilization story.

Inversion: How This Thesis Dies

Charlie Munger’s rule: tell me where I’m going to die, so I never go there. So invert the thesis. What would have to be true for this essay to be a trap? Start with those renders.

Civilizations announce themselves in renders before they prove themselves in yields. The Hoover Dam comparison cuts both ways: for every dam that got built, there is a graveyard of monumental architecture that never left the drawing board. Today the renders exist and the yields do not. Terafab is still more announcement than factory. First silicon is realistically a 2028 event, conditional on 18A and 14A reaching competitive yields — and Intel’s yields still trail TSMC’s. The foundry division continues to burn roughly $2.5 billion in operating losses per quarter. Not one major external foundry customer has been formalized; Apple and Tesla remain preliminary. Bloomberg’s reporting suggests some internal teams have been adjusting missed deadlines rather than recovering from them — exactly the cultural pathology Tan was hired to kill, still twitching.

The second death is in the price. At roughly $100 per share — down from June’s $140.94 high — Intel trades near 118x forward earnings. Bernstein’s warning applies to the whole complex: the real pricing risk is not Musk building his own memory; it is the capacity overshoot when all of these fabs come online in 2028–2029. Semiconductor history is a graveyard of investors who confused a capex announcement with a cash flow.

The third death is the model’s own: the Law of the Migrating Bottleneck cuts against you if you arrive after the migration is priced. Vernova at 8x is a very different entry than Vernova at the spin. Being right about where scarcity is heading and being early enough for it to matter are two separate skills.

The honest frame: the market has already paid for the resurrection. What it has not yet paid for — and cannot, until the yields exist — is the empire.

The MacroWise Reading

GE was resurrected because the world rediscovered its need for jet engines and gas turbines — and its unwanted child, Vernova, became an 8x monster precisely because the AI age turned out to run on electrons before it runs on tokens. Intel is being resurrected because the world — the American state, Nvidia, Google, and Musk — cannot afford to let it die.

That is the difference between a turnaround and a sovereignty thesis, and it changes what kind of risk you own. A turnaround can fail quietly, in the accounting. A sovereignty thesis fails loudly, in the geopolitics — which is precisely why so much sovereign and strategic capital has wrapped itself around this company. When your shareholders include the U.S. Treasury and your anchor customer is building his own power plants in Texas, bad news stops being a corporate-culture problem. It becomes a matter of state.

Larry Culp taught the market that a dead empire can be reborn by division. Lip-Bu Tan is teaching it that in the age of Physical AI, an empire is reborn by fusion — with the state, with energy, and with the machines that are about to walk among us.

The zombie is gone. Whether the rockstar can play the instruments is a question for the yields of 2027.

But the tool you should take from this essay is bigger than Intel. It is a question you can ask of any market, in any era: what does the top of the market consume that it does not contain? In 2000, the most valuable companies consumed software and contained none of it. In 2015, they consumed chips and contained one chipmaker. Today, eleven trillion-dollar companies consume gigawatts and contain zero electron-makers. The empty chair is never empty for long — and finding it before the market does is the whole game.

Thanks for reading ,

G

Combined hyperscaler capital expenditure in 2026 runs past $600 billion. Alphabet alone raised guidance to $205 billion, posted negative free cash flow of nearly $6 billion in the quarter, and is drawing down a $242 billion cash pile to do it. Analysts have warned that free cash flow across the largest four could fall by as much as 90% this year as spending outruns revenue.

The consensus reading is that this is a race for compute. Whoever stacks the most accelerators trains the biggest model, and the biggest model wins.

That reading is wrong, and the evidence is inside the spending itself.

About 40% of Alphabet’s technical infrastructure budget goes to data centres and networking equipment rather than accelerators. The flagship product of that spending isn’t a chip. It’s the Virgo Network, which links 134,000 TPUs into a single fabric inside one building, and more than a million accelerators across multiple sites into one training cluster. The same fabric is offered for GPUs, supporting up to 960,000 across sites.

Read that as a claim about intelligence rather than procurement. Google is not buying the biggest brain. It’s building the biggest connectome — and spending $205 billion to say that the wiring, not the chip, is the scarce asset.

Most portfolios are positioned for the other thesis.

The thesis in one paragraph

Intelligence is a property of networks, not of processors. That is not a philosophical position — it’s a measured result that shows up in biology, in cities, and now in capital expenditure. Which means the value in the AI build-out goes to the layers that create and govern connection, not to the layers that create raw computation. Compute is already priced. Connection, at both ends of the stack — physical interconnect and energy below, decision structure and actuation above — is not.

The rest of this piece does two things. First, it shows you where that puts the capital. Second, it teaches you the framework the argument comes from, because the framework will still be useful when these specific companies aren’t.

Part 1 — Where the capital should go

The stack is not a stack

Every AI investment map you’ve seen is made of information layers: chips, models, data, applications. That’s a third of the system.

Complex systems studies networks that move energy, matter and information — and the interaction between all three. An information-only map of AI is describing an organism by its nervous system and forgetting it has a metabolism and a body.

This isn’t a metaphor, and there’s a hard physical result behind it. In 1961 Rolf Landauer proved that erasing one bit of information dissipates at least kT·ln2 of energy as heat — about 3 × 10⁻²¹ joules at room temperature. Not an engineering limitation; thermodynamics. Confirmed experimentally in 2012. Computation always costs energy. You cannot design that cost away. The AI build-out is a thermodynamics problem dressed up as a software problem, and that’s a physical bound, not a rhetorical flourish.

Put the missing flows back and the map changes shape.

Seven layers, and it closes into a loop rather than terminating at the top:

Generate (energy). Not a cost line — the limit that binds everything else, the same way metabolic rate binds an organism. Nuclear restarts, gas turbines with multi-year lead times, solar plus storage, behind-the-meter generation. Partly priced.

Distribute (energy). The grid, dispatch, smart grids. And note: the grid is itself a complex adaptive system, absorbing intermittent supply plus enormous spiky interruptible AI load. Feedback loops, delays, congestion, cascade risk. Barely priced.

Compute (matter). Silicon, fabs, memory, cooling. The largest physical supply-chain constraint in the loop. Fully priced.

Connect (information). Optics, switching, scale-across fabric. Turns scattered silicon into one machine. Starting to be priced.

Infer (information). The models. Kind-environment work: code, proofs, protein folding, images. Solved and becoming a commodity. Fully priced.

Decide (information). The shared structure where a human team and a model work the same problem — objects, relations, and staged consequences rather than retrieved text. Barely priced, and thinly represented: one expensive pure play does not price a category.

Act (matter). Robots, sensors, actuators, vehicles, production lines. The edges that touch the physical world. Least priced of the seven. Most of the visible enthusiasm sits in private rounds. In public markets the actuator complex — industrial robotics, motion control, sensing — still trades like ordinary capital goods, on multiples that assume no change in what these machines can do.

Then the loop closes: every action changes the world, which changes what energy and matter the system needs next. Decisions pull on electrons.

The two fully-priced layers are the two that are closest to pure information — and they are the two turning into commodities fastest. The four barely-priced layers are generation, dispatch, decision structure and actuation. Three of those four are physical.

And the least priced of all is the one that closes the loop. Physical AI is where the system stops reading and starts acting — which is the moment its feedback becomes honest. Markets are paying most for the layers with the weakest feedback, and least for the layer with the strongest.

Read as a shopping list, Robotics–AI–Compute–Electricity is four sectors. Read as a loop, it’s one organism whose layers constrain each other, and the constraint is usually below where people are looking.

Why connection beats computation

The reason this isn’t just a contrarian preference is that the exponent is measurable, and it’s been measured three times in three unrelated systems.

In biology. Geoffrey West and colleagues showed that metabolic rate scales with body mass to the power of 3/4. The result holds across 27 orders of magnitude. And the exponent comes from the geometry of branching networks — not from chemistry. From topology.

In cities. West and Luis Bettencourt found the exponent splits. Infrastructure grows more slowly than population — an exponent of about 0.85 — double the population, need only 85% more road and cable. But output — wages, patents, GDP, count of inventive people — grows faster than population — an exponent of about 1.15. Double the city and you get 15% more invention per person.

That 1.15 doesn’t come from geography. It comes from the topology of the social network: interactions per person rise with size, and ideas are combinations of interactions. A city isn’t a box that holds brains. It’s a fabric that connects them, and the superlinearity lives entirely in the wiring.

Share

In brains. Roughly 20 watts, beating every data centre on Earth at unstructured reasoning — not because neurons are fast (they’re catastrophically slow) but because of wiring density and adaptivity.

Three systems, one result: connection grows faster than size; substrate grows more slowly. Google is the fourth data point, and it’s spending $205 billion to generate it.

The investable form of this: any company whose growth adds connections faster than it adds nodes is on the right side of 1. Any company where growth adds coordination faster than it adds capability has crossed to the wrong side.

Reading a company by its exponent

West’s group analyzed roughly 28,000 American public companies over sixty years. Half die within about a decade, on a survival curve resembling a mammal’s far more than a city’s.

The mechanism is the crossing. Companies start superlinear — headcount buys disproportionate invention. As they grow they cross into sublinear — headcount buys disproportionate coordination. Bureaucracy isn’t a moral failure; it’s an exponent shift, and it’s close to a law.

Cities almost never die because they never make that crossing. Firms almost always do.

So the single most useful question in AI equity analysis is not about TAM or model benchmarks. It’s: is headcount still buying invention here, or is it buying meetings?

Three questions for any position

1. What’s flowing — energy, matter, or information? Usually all three, and the failures happen where they couple. A data centre is an information business with an energy bottleneck. A robotics company is a matter business with an information bottleneck. Name the flows before you value anything.

2. What’s the exponent? Is scale buying more than it costs, or less? Above 1 or below?

3. Is the landscape moving? If the terrain is still, optimization works and a moat is a moat. If the terrain is made of competitors who react to you, “moat” is the wrong noun — what you have is a position on a surface that deforms when you move. More on this in case file 6.

Those three questions are the whole framework, applied. The rest of this piece is where they come from.

Part 2 — Why complex systems is the right framework

Calculus built the machine age

Newton and Leibniz did not solve a hard problem in physics. They built the language in which physics became a solvable subject. Before calculus, change was a philosophical embarrassment. After, it was an operation you performed.

Be careful with the dates, because the popular version is wrong. Calculus did not cause the first Industrial Revolution — Watt and Arkwright were tinkerers, and thermodynamics arrived after the steam engine, Carnot in 1824, decades after the machines were already turning. Historians of science have a line about it: science owed more to the steam engine than the steam engine owed to science.

Where calculus actually took over was the second wave, from about 1870 — electrification, industrial chemistry, precision engineering, structural steel. Maxwell’s equations. Stress analysis. Control theory. That’s when industry started running on differential equations, and it never stopped. Neoclassical economics is calculus. Black-Scholes is a diffusion equation, the identical mathematics that describes heat moving along an iron bar.

Calculus defined the machine age — not because engineers used derivatives, but because it installed a picture of what a system is:

• Systems are smooth — small changes, small effects.

• Systems are decomposable — understand the parts, sum them, done.

• Systems are deterministic — same inputs, same outputs.

• Systems seek equilibrium.

• Effects are proportional to causes.

That picture is what a machine is. For two centuries the most valuable thing you could do was build one, so the picture was right about everything that mattered.

Subscribe now

Two ages, two questions

The clean way to see the shift is in the question each era organizes around.

The machine age asked: how do I build something that reliably does X? An engineering question. It rewards specification, decomposition, tolerance control, optimization. Calculus is the perfect language for it, because a machine is exactly the kind of system calculus describes.

The discovery age asks: how do I search a space too large to enumerate? Not an engineering question at all. Nobody specifies the answer in advance because nobody knows what’s in the space. It rewards exploration, recombination, knowing when to prune, and recognizing a good region on arrival.

That second question is what complex systems has studied for fifty years — fitness landscapes, the adjacent possible, criticality, coevolution — without being able to answer it at scale. It’s also precisely what an AI system does. A model searching combinatorial space at inhuman speed is a search process, and search over rugged landscapes is the one thing this field has a real theory of.

Calculus was the generalist language of the machine age. Complex systems is the generalist language of the discovery age. Not because it’s more elegant. Because it matches the shape of the problem.

What the machines actually took

The psychologist Robin Hogarth drew a line between two kinds of learning environment. David Epstein built his book Range around it.

A kind environment has fixed rules, lots of repetition, and feedback that comes back fast and honest. Chess is kind. Golf is kind. Reading a scan is fairly kind. Practice works here. Ten thousand hours does what the story says it does.

A wicked environment is the opposite. The rules move while you play. Feedback comes late, or noisy, or wrong. Sometimes yesterday’s pattern reverses. Markets are wicked. Geopolitics is wicked. Starting a company is wicked. Raising a child is wicked.

Epstein’s uncomfortable finding, drawn from Philip Tetlock’s twenty-year study of forecasters, is that in wicked environments deep single-field expertise stops helping. Sometimes it hurts. Tetlock’s narrow experts — the ones who ran everything through one big framework — forecast worse than generalists who used several rough ones at once.

The reason is simple. A deep specialist doesn’t just know more. He knows in one shape. When the problem changes shape, that depth becomes a cost he can’t write off.

Now put the machines on that map.

Kind environments are exactly where AI won first, and won clearly. Go. Protein folding. Code. Medical imaging. Competition maths. The pattern is not that AI is good at hard things and bad at easy things. It is good at kind things, however hard, and unreliable at wicked things, however simple they look. A model can write you a working compiler and then be useless on whether your company should exist.

I sort learning into four modes:

• Imitation — copy a pattern that already works.

• Optimization — tune a solution inside known rules.

• Cross-pollination — carry a structure into a field that never expected it.

• Representation — build the frame the problem is posed in.

The first two are kind-environment work. They are gone. A language model trained on everything ever written is an imitation machine at planetary scale. Gradient descent is an optimizer that never gets tired, never defends its ego, and searches in more dimensions than you can picture. These were not fair fights.

The last two are where the next decade happens. And they are not vague creative virtues. They have a science.

Why this is the language for working with AI

Three consequences follow, and they are the reason to learn this now rather than eventually.

It tells you when to trust the machine. Kind and wicked is the distinction that maps onto model reliability, and nothing in the interface marks which one you’re in. Hand a model a kind problem and it beats you, often by a lot. Hand it a wicked one and it produces something fluent, confident and structurally unreliable — and you cannot tell from the output. Most AI failure in practice isn’t the model being bad. It’s someone treating a wicked problem as kind.

Your remaining edge is this field’s subject. Cross-pollination and representation aren’t personality traits. Complex systems is the only discipline whose entire subject matter is structure that transfers between domains — which is to say, it is the study of the two modes you have left.

AI is itself a complex system, and almost everyone models it as a machine. Listen to the vocabulary: prompt engineering, instructions, commands, getting the model to follow them. That’s a code model — symbols with stable meanings, composed by rules, producing determinate output. It’s wrong in exactly the way “the genetic code” was wrong, and case 7 below is the proof.

Why the timing isn’t a coincidence

If the field is fifty years old, why is it becoming central now?

Because for most of those fifty years it described beautifully and predicted badly. The systems it studies are impossible to shortcut — no closed form, no shortcut, no way to know what they do except to run them. Calculus hands you an equation you solve on paper. Complex systems hands you a system you must simulate. Without enormous compute, the field produced deep intuitions and few usable answers.

That constraint is lifting right now. The generalist science has just been handed the instrument it was missing, the way astronomy was handed the telescope. Galileo didn’t invent the questions; he got the thing that let existing questions be answered.

Part 3 — The evidence: seven cases

This is the tutorial. Seven cases, each teaching one formal object you can carry into any system. Same shape each time: the case (what happened, with the source), the object (the concept), the transfer (markets), the mistake (the standard error). Each ends with one line worth keeping.

One promise: everything here is measurable. That’s the entire difference between this field and the mysticism that circulates under similar branding.

Case 1 — Biology: the mouse and the elephant

The case. In 1932 a Swiss chemist named Max Kleiber measured how much energy farm animals burn at rest. It was practical work. Farmers wanted to know how much feed to buy. He plotted energy use against body weight on log axes and got a straight line. The slope was three-quarters. Double an animal’s mass and it needs about 75% more energy, not 100%. The line has since held across 27 orders of magnitude.

Sixty years later West, Brown and Enquist showed why. Every large organism must move resources from one pump to every cell. The solutions converge on branching networks that fill space, terminate in fixed-size units, and minimise pumping energy. Impose those three constraints and 3/4 is a theorem, not a fit.

The object: the scaling exponent. Below 1, you get less than you paid for and growth is bounded — the organism grows, slows, stops, dies. Above 1, you get more than you paid for and growth heads for a singularity in finite time, which must be reset by innovation, faster and faster.

The transfer. The company-versus-city diagnostic in Part 1. It’s the highest-yield question in the piece.

The mistake. Assuming the exponent is fixed. It’s a consequence of internal structure, and structure can change. That’s what a real reorganization attempts. Most fail.

Carry this: the exponent is a property of structure, not of industry. Structure can be changed. Almost nobody does it in time.

Case 2 — Mycelium: the organism that redesigned Tokyo

The case. Physarum polycephalum is a slime mould — not a fungus, not an animal, a single enormous multi-nucleate cell that forages by extending tubes. No brain, no neurons, no central anything.

In 2010, Atsushi Tero, Toshiyuki Nakagaki and colleagues published the result of a simple experiment in Science. They placed oat flakes in a pattern matching the cities around Tokyo, and let the mold grow outward from the center. Because the mold avoids bright light, they used light to stand in for mountains and water — the terrain a real engineer has to route around. The mould flooded the space with tubes, then pruned. The mold first flooded the space with tubes. Then it pruned. The network it settled on matched the real Tokyo rail system on efficiency, fault tolerance and cost. And it did this with no central control, which is what makes the method scale in a way a planner’s solution does not.

The algorithm is almost insulting: tubes carrying more flow thicken, tubes carrying less shrink and vanish. That’s all of it. No model of Tokyo exists anywhere in the organism.

The object: distributed optimization without a controller. Positive feedback on used paths, negative feedback on unused ones, parallel search across everything at once. Note the order — explore first, prune second. The mould wastes enormous energy covering ground it will abandon, and that waste is what buys it a good network.

The transfer. Nobody sets the price of copper. Price is what the network settles on after local reinforcement and pruning run everywhere in parallel. It’s also the honest description of how capital finds new industries: flood, then prune. Venture portfolios are Physarum with a fee structure.

For your own positioning: prune too early and you converge fast on a network connecting the wrong cities.

The mistake. Reading this as “decentralisation always wins.” Physarum’s method is wasteful and slow, and works only when exploration is cheap relative to the value of the right answer. When exploration is expensive or the answer is known, a planner wins easily.

Carry this: the waste is the mechanism. A system that cut away dead or unwanted branches before it has explored converges quickly on a network connecting the wrong cities.

Case 3 — Cosmology: the web that looks like a brain

The case. Matter isn’t spread evenly. It sits in the cosmic web — dense nodes joined by filaments, wrapped around near-empty voids. That structure grew from a universe smooth to about one part in 100,000, through the simplest possible positive feedback: a denser region pulls in more matter, becomes denser, pulls harder.

In 2020 the astrophysicist Franco Vazza and the neurosurgeon Alberto Feletti published a comparison in Frontiers in Physics between the cosmic web and the neuronal network. The two systems differ in scale by more than 27 orders of magnitude: roughly 69 billion neurons against at least 100 billion galaxies. Their analysis suggested that very different physical processes can build structures with similar levels of complexity.

The object: positive feedback turns noise into structure — and the limits of analogy.

The second half matters more. The Vazza–Feletti result is real and weaker than the headlines said: the authors are explicit that they used an approximation of neural connectivity based on simple proximity, not true connectivity. So the correct reading is similar network statistics from unrelated mechanisms — interesting, and not evidence the universe is thinking.

The transfer. Wealth, market caps, city sizes and word frequencies all follow power laws, and it’s tempting to infer a shared cause. Don’t. Power laws arise from at least half a dozen unrelated mechanisms. Seeing one tells you the shape of a distribution, not the process — and only the process lets you predict.

The mistake. Analogy-hopping, the failure mode of every popular complexity book. Apply this test hard, including to this essay: shared mechanism, or only shared shape? Shared mechanism is science. Shared shape is a coincidence that feels profound.

Carry this: a shared shape tells you what a distribution looks like. Only a shared mechanism tells you what happens next — and only the second one is tradeable.

Case 4 — Communication: Shannon’s channel, Landauer’s bit

The case. Shannon’s 1948 paper created information theory nearly complete. His central move was stripping meaning out: information is reduction of uncertainty, and a message telling you what you knew carries none. He proved every channel has a capacity, below which you can transmit with arbitrarily small error by adding redundancy, and above which you cannot.

Landauer’s 1961 result supplies the bridge to physics, as described in Part 1.

The object: redundancy is the price of robustness, and information rides on energy.

Shannon showed reliability is bought with redundancy — extra bits carrying no new information, existing only to survive noise. A maximally efficient code is maximally fragile: flip one bit and the message is gone. Slack isn’t waste; slack is error correction. Every efficiency drive in a supply chain spends redundancy, whether or not anyone says so.

The transfer. A market is a channel. Prices are messages. Signal is the new information inside a price move. Noise is most of what you see over short periods. And capacity is how much information a market can take in per unit of time. When information arrives faster than the market can price it, you get gaps and cascades instead of smooth adjustment. A perfectly efficient market carries maximal information and zero redundancy, which also makes it maximally fragile. That’s a sharper statement of the efficiency/stability trade-off than most finance writing manages.

The mistake. Confusing information with meaning. Entropy measures surprise, not importance. A random number generator is maximal-entropy and tells you nothing. In markets: mistaking volatility for information.

Carry this: redundancy is not waste, it is error correction. Every efficiency programme is a decision to spend robustness, whether or not anyone says so out loud.

Case 5 — Quantum: Anderson’s superconductor

The case. In 1972 the condensed-matter physicist Philip Anderson published More Is Different in Science. He later won a Nobel and, twelve years on, co-founded the Santa Fe Institute. That essay is the field’s actual founding document.

His target was the belief that science is a ladder — understand particles, derive chemistry, derive biology. His counterexample came from his own field. Superconductivity is a quantum phenomenon, but you cannot derive it from the Schrödinger equation for one electron, or a thousand. It requires an enormous number of electrons pairing collectively into a single coherent state. The property exists nowhere in the parts.

The object: emergence, defined strictly. A property is emergent when three things are true. It exists at one level. It is absent at the level below. And it needs its own laws, which nobody can derive in practice from the level underneath. Temperature is emergent: one molecule has none. Liquidity is emergent: one trader has none.

The practical consequence is a licence: you may study a level on its own terms. You don’t need particle physics to do chemistry, or neuroscience to do economics. This is what makes a generalist science possible at all.

The transfer, and a warning. Quantum mechanics has a real connection to complex systems. It is the level chemistry emerges from. And quantum error correction is Shannon’s redundancy result applied to a much noisier channel. It has no relationship to markets, consciousness, or “quantum thinking.” Quantum effects decohere far below any scale relevant to a trading floor. A price is not in superposition. Anderson’s argument actually cuts against quantum mysticism: you don’t need quantum mechanics to explain markets precisely because markets are a higher emergent level with their own laws.

The mistake. Using “emergent” to mean “complicated,” or as a way to avoid explanation. If you can’t state the new level’s laws, you’ve found something you don’t understand yet — not emergence.

Carry this: you are allowed to study a level on its own terms. That licence is what makes a generalist science possible — and it is also why markets need no quantum mechanics.

Case 6 — Markets: the bar in Santa Fe

The case. El Farol is a real bar in Santa Fe with Irish music on Thursdays. In 1994 W. Brian Arthur turned it into the cleanest problem in the field. A hundred people want to go, but only if fewer than sixty show up. Everyone decides independently from attendance history.

Try to reason it out. Conclude it’ll be crowded and stay home — if that reasoning is good, everyone does it, the bar is empty, you were wrong. Conclude it’ll be quiet and go — so does everyone, it’s packed, wrong again. No deductively rational solution exists, because any shared reasoning process invalidates its own conclusion.

Arthur’s resolution: give agents competing hypotheses, let them use whichever has been working and discard what fails. Simulated, attendance converges around 60. No agent knows the answer. The population computes it.

The object: coevolution, and induction over deduction. Your fitness landscape is made of other agents who are also climbing. When you move you deform their terrain; when they move they deform yours. You can climb hard and lose altitude because the ground is sinking.

The biological version is the Red Queen, named by Leigh Van Valen in 1973. He’d found something odd in the fossil record: a lineage’s extinction probability is roughly constant regardless of how long it has already survived. Two hundred million years of refinement buys no durable safety, because everything you compete against refines too.

The transfer. This is why alpha decays, and the mechanism is more specific than usually stated. Your edge is a peak on a landscape made of other people’s positions. Find it, trade it, and the terrain flattens — not because you were wrong but because you were right and others arrived. Decay is the landscape responding to you.

This is also why backtests look better than reality. A backtest assumes the ground stays still. It tests your strategy against a world that would have reacted to it.

A durable edge, then, isn’t a better hypothesis. It’s a faster loop between hypothesis, test and replacement.

The mistake. Treating a strategy as a fact rather than a position on a moving surface. The tell is having no answer to “what would make me abandon this?” In a coevolving system that isn’t a risk-management formality — it’s the method.

Carry this: a durable edge is not a better hypothesis. It is a faster loop between hypothesis, test and replacement.

Case 7 — Life: the code that turned out not to be a code

The case. For fifty years biology ran on a metaphor. DNA was the blueprint, the genome was the program, and once sequenced we’d read an organism like source code.

The Human Genome Project finished in 2003 and the metaphor came apart. Pre-sequencing estimates ran to 100,000 genes or more; the actual count of protein-coding genes came in around 20,000 — roughly what a millimetre-long roundworm has, and fewer than a grain of rice. Then: one gene does not make one protein, since alternative splicing yields many products depending on cellular conditions. And finally, the fact that should end the argument. Every cell in your body carries the same genome. A neuron and a liver cell run identical code and are nothing alike.

The difference isn’t in the sequence. It’s in the state of the system reading it.

Philip Ball’s How Life Works states the conclusion directly: gene-as-instruction isn’t a simplification of the truth, it’s a misdescription. Causation is spread across levels, and the same molecule means different things at different times because meaning is assigned by context.

The object: context-dependence and downward causation.

In a code, symbols have stable meanings and the whole is their sum. In a complex system, the state of the whole decides what the parts mean. A gene in a neuron and the same gene in a liver cell are not the same causal object. The network reading them is in a different state.

There is no minimal unit carrying the meaning. Interaction has meaning. State has meaning. Context has meaning. The unit is just what’s participating. Meaning is topological — it lives in the links.

The transfer. This is why the same information moves markets differently on different days. A jobs number isn’t a signal with fixed meaning: in one regime strong employment reads as growth and equities rally; in another the identical number reads as inflationary and equities fall. Nothing about the data changed. The state of the system reading it changed.

Anyone who has built a factor model has hit this and usually calls it regime instability or parameter drift. It isn’t drift. It is the same structural fact biology hit in 2003. You cannot give a variable a fixed meaning inside a system whose state decides what things mean. A model that maps inputs to outputs with stable coefficients is a code model. It works until the context moves.

The practical version: stop asking what an indicator means. Ask what state the system is in, then what the indicator means in that state.

The mistake. Reaching for the blueprint metaphor anywhere — genome as code, brain as computer, market as machine, prompt as program. Each one is tempting because it promises that if you find the right minimal unit, meaning will be sitting inside it. It never is.

Carry this: stop asking what the signal means. Ask what state the system is in, then ask what the signal means in that state. Only the second question has an answer.

Part 4 — The toolkit: twenty questions

Learn the objects, not the anecdotes. Anecdotes are how this field gets sold; objects are what it is.

But a definition sits on the page and a question travels into your next meeting. So here are the twenty as questions. Run any system past them and the ones that bite will tell you what kind of system you’re holding.

Structure — how it’s put together

1. Network topology. Who is actually wired to whom — and is that the same as the org chart?

2. Modularity. If this part fails, does the failure stay local, or does it take the neighbours?

3. Hierarchy. Am I trying to explain this level using the vocabulary of the level below it?

4. Bottleneck. What single constraint sets the rate here — and am I optimizing something else?

Scale — how behavior changes with size

5. Power law. Is there a typical size here, or am I averaging things that have no middle?

6. Scaling exponent. Does the next unit of scale buy me more than it costs, or less?

7. Fat tails. Am I treating the extreme as an anomaly when it’s the same process running?

8. Emergence. Does this property exist at all in the parts, or only in the assembly?

Dynamics — how it moves

9. Nonlinearity. If I double the input, do I actually get double the output — or have I just assumed it?

10. Positive feedback. What here feeds itself? Where does the runaway end?

11. Negative feedback. What’s damping this — and what happens on the day that damping stops?

12. Delay. How long between action and consequence, and am I reacting to a signal that is already out of date?

State — how it sits and how it jumps

13. Attractor. What regime is this system in right now, and what would it take to leave?

14. Phase transition. Is the next increment a change of degree, or a change of kind?

15. Criticality. Is this system tuned to the edge — where the trigger will look trivial and the response won’t?

16. Path dependence. How much of the current state is history that cannot be undone?

Adaptation — how it changes itself

17. Fitness landscape. How coupled are the parts — and does that make local improvement enough, or a trap?

18. Adjacent possible. What becomes reachable only after I take the next step?

19. Coevolution. Does the thing I’m optimizing against optimize back?

20. Robustness vs. fragility. What redundancy am I about to remove in the name of efficiency?

Part 5 — Using it

On AI, specifically

David Krakauer, the Santa Fe Institute’s president, draws the distinction that matters most this decade.

A complementary cognitive tool leaves you more capable when it’s removed — the abacus teaches an arithmetic structure that outlives it, and practised users end up calculating on an imagined one. A competitive cognitive tool leaves you less capable when removed — GPS delivers you to the address and dismantles your mental map of your own city.

AI can be either. The difference isn’t the model; it’s whether you use it to produce output or to explore structure. Let it traverse the adjacent possible — that’s what it’s extraordinary at. You keep the representation. The test at the end of a session: could you rebuild the argument without it? If not, you didn’t learn. You consumed.

Almost everyone is using it the first way, which is also why the returns to using it the second way are going to be large.

The systems that matter now don’t decompose. The average describes nothing. Small causes and large causes come from the same distribution. And the value in the largest capital cycle in history is accruing to the links rather than the nodes, because in a complex system it always does.

Meaning is topological. So is money.

Beautiful Sunday !

Guillermo Valencia A

On July 19, a problem open since 1939 fell. A machine found the answer. By the time Kevin Buzzard woke up in London, it had already been checked by another machine and confirmed correct.

On July 22, Alphabet and Tesla reported earnings and were punished for spending money on exactly that capability.

That gap is the whole story.

The evidence

For most of the last two years, “AI discovers things” was a promise. It stopped being a promise this year.

May 2026. An OpenAI model solved a problem Paul Erdős posed about eighty years ago. For decades mathematicians assumed the best answers looked roughly like a square grid. The model proved that wrong — not by finding a better version of the grid, but by finding a completely different family of shapes that beats it. Timothy Gowers, a Fields Medallist, said that if a person had submitted the paper, he would have recommended publishing it in the Annals of Mathematics without hesitation.

July 19, 2026. The Jacobian conjecture, open since 1939, was disproved. The answer is a formula that passes the standard test at every single point and still sends three different starting points to the same place. It was machine-verified in Lean overnight.

May 2026, and this is the one that matters. A Together AI and Stanford team ran a site called EinsteinArena — a public board where AI agents work on unsolved maths problems, with automatic checking and an open discussion forum. The agents produced twelve results better than anything a human or machine had achieved before. One improved a known bound from 593 to 604.

The paper is blunt about how. Not one agent. Not one clever run. Agents posted partial results, argued with each other in the forum, improved each other’s checkers, and borrowed each other’s ideas.

Why does mathematics count as evidence when nothing else does? Because it’s the only field with a receipt. A proof is right or it isn’t. There’s no story in between, and increasingly no human in between either. Everywhere else — drug discovery, materials, chip design — you wait years to find out whether a claim was real. In maths you find out before breakfast.

Eighteen months ago these results were a trickle. Now they arrive monthly. That is what the beginning of a discovery age looks like.

What discovery costs

Alphabet made $119.8 billion last quarter, up 24%. Cloud grew 82%. Backlog hit $514 billion. On paper, the most profitable quarter in company history.

The stock fell 5%. Because Google raised its spending plan to $195–205 billion for the year — up from $180–190 billion promised three months earlier — and said 2027 would be higher. Cash out exceeded cash in.

Tesla reported the same afternoon. Record revenue, record deliveries, and costs up 47% as it poured money into self-driving, robots and chips. Margins fell to 1.4%. The stock dropped 14.5% the next day and lost $140 billion.

Here is what that money buys. Not better ads. Not smarter chatbots. Attempts per second against problems too large to search by hand. That is the product. The Erdős result and the EinsteinArena results are what it looks like when the attempts land.

And this isn’t two companies having a bad week. Look at how the last month has been scored.

Top of the list: Apple +13.6%, Palantir +8.7%, Solana +8.7%, Meta +6.7%, Bitcoin +5.1%, Microsoft +4.4%.

Bottom of the list: Google −7.4%, Cooling −8.4%, Memory −13.3%, Tesla −14.9%, AI Interconnect −18.7%, Oracle −27.0%.

Now read what those two groups actually are.

The top is asset-light. Software, platforms, brands, and two purely financial assets. Apple — the one large technology company that has conspicuously not bet its balance sheet on this build — leads by a wide margin.

The bottom is almost entirely physical. Cooling. Memory. Interconnect. The companies pouring concrete and buying transformers. Oracle and Tesla, the two names that most recently told the market their bill was going up.

The market is selling atoms and buying bits.

Which is precisely backwards if the constraint is what I think it is. The layers that decide how fast this gets built are power, interconnect and machines that move — and those are exactly the ones being sold hardest. It isn’t a judgement that they’re unnecessary. It’s a judgement that they’re expensive, which is true, and that expensive is bad, which is only true inside one quarter.

That is a market instructing companies to stop.

Now widen the lens.

Green is money coming in, red is money going out, one row per link in the chain, three and a half years across.

Three things in that picture matter.

Nothing stays lit. Palantir leads, then Oracle, then Memory, then something else. No link holds the lead for long. The market has spent three years arguing about which part gets paid, and almost never about whether the machine is real.

At the turns, everything moves together. Look at the vertical bands — the wall of red in early 2025, the wall of green that summer. On ordinary days the market treats this as fifteen separate stories. At the moments that actually matter it treats it as one thing, which is what it is. The rotation is noise on top of a single position.

And the chain is getting longer. The AI Power row is blank until the middle of 2024 — not because it did badly, but because it wasn’t a category yet. Cooling and Interconnect were niche engineering line items three years ago. New rows keep appearing on that chart.

That last one is the discovery signal hiding in a price chart. In a normal industry the categories are fixed and the money moves between them. Here, new categories keep being born — which is what happens while the space is still being mapped.

The market can price the rotation. It cannot price the new rows, because a row that doesn’t exist yet has no earnings to miss.

And the reason it can’t is structural rather than stupid.

A quarterly grade measures margin and growth in businesses that already exist. It is superb at that — faster than any institution on earth. What it cannot measure is a company changing what it is, because that pays off in a world that hasn’t arrived — and there is no line on an income statement for a world that hasn’t arrived.

So it shows up the only way it can: as cost.

Two computer scientists, Lehman and Stanley, proved the general version of this. On hard enough problems, having a goal makes you worse at reaching it, because the steps that actually lead there look like moving backwards. The steps toward the goal look like failure.

That’s July 2026.

Why the rate is rising

Discovery isn’t speeding up because models got smarter in isolation. It’s speeding up because they’re being connected to each other.

Look again at EinsteinArena. The results came from borrowing between agents. Not from one system thinking harder.

Biology settled this a while ago. When scientists reconstructed how bacteria acquire new abilities, they found bacteria don’t mainly invent new genes — they borrow them, from other bacteria, from viruses, from whatever’s nearby. Borrowing beats inventing by about ten to one.

Which is why the fight now happening over open models matters more than anything in the earnings.

On July 16 a Chinese company released Kimi K3 — downloadable by anyone, competitive with the best Western models on some coding tests, at roughly $15 per million words of output against about $50 for the closest closed rival. Washington considered banning it.

On July 22, Jensen Huang said the opposite: these models are excellent and should be used. If everything collapses into one model, you’ve built one target and one thing that can break.

On July 24 he made his first post ever on X, sharing an open letter signed by about 25 companies — Nvidia, Microsoft, Meta, Hugging Face, IBM, Palantir. OpenAI and Anthropic didn’t sign.

Strip the politics and the argument is simple: open weights are borrowing, and borrowing is how discovery rate goes up. Calling distillation theft gets the biology backwards.

There is a limit, and it’s real. Manfred Eigen proved in 1971 that any self-copying system has a speed limit on change — past it, errors pile up faster than they can be cleaned out, and the thing stops improving and falls apart.

Crypto is the one field that already found that line. Fifteen years of total openness produced genuinely new tools that no committee produced — automated market makers, practical zero-knowledge proofs, stablecoins. It also produced tens of thousands of worthless tokens and a year where nothing had time to stick.

So both camps are half right. Copying well sets your floor. Borrowing widely sets your ceiling. Nobody is asking where the top of that curve sits.

It’s early

The thing being built has six layers. Models that suggest. Ontologies that define what things are and let answers be checked. Compute that searches. Electricity underneath all of it. Sensors that connect it to the physical world. Robots that act on it.

Most maps of this stop at four — models, compute, power, robots. The two usually left out, ontologies and sensors, are the connecting layers. They carry no revenue line of their own, and they decide whether anything else fits together.

The layers are nowhere near equal, and they don’t move at the same speed.

Two are genuinely underway. Compute is furthest along, at roughly 39%, and building fast. Models are around 27% and on the steepest curve of the six, because software compounds in years.

The other four have barely started, and they’re slower by nature. Electricity sits near 12% and builds on a decades-long clock — permits, substations, generation. Sensors are around 9%. Ontologies are near 6% and may be slowest of all, because shared standards are an agreement problem rather than a technology problem, and agreement has never been fast. Robots are at 2%, the last to begin and probably the steepest once they do.

Add it up and you get about 16% built.

That is a picture of belief, not a measurement. From inside an S-curve you can’t tell it from a straight line. But the shape of the claim is clear: the two layers doing the discovering are running well ahead of the four layers that have to carry it, and the gap between them is where the next decade gets decided.

Share

How to know if it’s working

How many ounces of gold the stock market buys. Three peaks — 1929, 1966, 2000 — each roughly double the last. Three bottoms: 1942, 1980, 2011.

When the line rises, a civilisation is making new wealth. When it falls, it’s fighting over the wealth it already has.

Every climb sat on top of a real discovery age. Electricity and cars. Semiconductors and computers. The internet and mobile. Every fall was war, inflation, or redistribution — and gold won those, because gold is the only large asset that doesn’t depend on anyone keeping a promise.

That’s the instrument. If the discovery age is real, this line goes up for twenty years.

And here’s the tension. We’re building the largest technology stack in history while living through every symptom of a fighting-over-it period: tariffs, rival tech blocs, $39 trillion of US debt, electricity bills rising because of the buildout itself. The ratio has gone sideways for twenty-five years.

So the question is simply: does the discovery end the fight, or does the fight stop the discovery?

Digital gold for whom?

Bitcoin and Solana are on that map now, and they belong there. They’re not a separate asset class sitting beside this machine. They’re a layer of it.

But it forces an honest question:

The standard case says Bitcoin is digital gold. So the last eighteen months should have proved it: tariffs, blocs, debt, currency stress — perfect conditions. Gold did exactly what theory predicts. Bitcoin did not. It’s near $65,000, down about $53,000 from a year ago, with the worst ETF outflows on record in June.

It’s the wrong test.

Ask what gold actually does for the people who hold it. A central bank, a family office, a state — each already has a legal system, a bank account, courts, treaties, borders and a passport. Gold is the thing they hold when the promises in that system come under strain. It’s a hedge for an entity that already has everything else.

For that holder, Bitcoin’s distinctive properties are mostly irrelevant. Programmable? They have lawyers. Settles without a counterparty? They have correspondent banks. Verifiable by anyone? They have auditors. Bitcoin is competing for a job that is already well staffed, which is why it trades on liquidity — because for a twentieth-century institution, that’s genuinely all it is.

Now ask a different holder.

An autonomous agent cannot open a bank account. It cannot sue. It has no jurisdiction, no passport, no courts, no counterparty willing to be sued on its behalf. It cannot hold gold, because gold requires a vault and a vault requires a legal person.

Subscribe now

What it needs is an asset and a settlement rail that are pure ontology — checkable by machine, no counterparty, no permission, no legal system underneath. That is the exact and only thing Bitcoin was built to be.

So it’s digital gold for digital beings. And digital beings are not here yet.

That reframes everything about the price. Bitcoin’s real customer base is gated by how much economic activity is conducted by non-humans — which today is approximately none. Look back at the layer chart: sensors at 9%, robots at 2%. The agentic economy sits on the two least-built layers of the machine.

We are pricing a reserve asset for an economy that hasn’t been born, using the behaviour we’d expect from a mature one.

Now — “it’s early” is what every failing thesis says, so this only counts if it can be wrong. Here’s how:

What would confirm it. Settlement volume between software agents becomes a real number, and as it grows, Bitcoin’s correlation to liquidity conditions falls. The asset starts trading on machine demand rather than on ETF flows.

What would kill it. The agentic economy arrives — real machine-to-machine settlement at scale — and it runs on stablecoins, or on a bank consortium rail, or on something not yet built, and Bitcoin keeps trading on the Fed. Then the slot existed and Bitcoin didn’t win it, which is a different failure from being early and a worse one.

Nothing before that fork is evidence. Everything written in the meantime is narrative, mine included.

One small note from the current data: in the month when the market punished everyone building physical infrastructure, Solana was +8.7% and Bitcoin +5.1% — both in the top half of the chain. Make of that what you like. One month is one month.

What to watch instead of earnings

Machines are now answering questions. They are not yet asking them. As Buzzard put it, the famous problems are named after the people who posed them, not the people who solved them — and when you ask a machine to invent a question, you get something boring, obviously true, or obviously false.

So the things worth tracking aren’t quarterly:

The rate of verified results. Not announcements — results that survive a checker. That number is the discovery age’s pulse.

New rows on the map. Not which link is winning this month, but whether links keep being added. AI Power wasn’t a category in 2023. The month that chart stops growing new rows is the month the mapping is finished.

Settlement between machines. The first real number for value moving between software agents, with no human approving it. That is the birth certificate of the economy Bitcoin was actually built for.

Whether the connecting layers get built. Ontologies and sensors. If they stall, the rest is a very expensive autocomplete.

Whether open borrowing stays legal. Restriction is the one policy choice that directly lowers the discovery rate.

The ratio. Twenty years of the stocks-to-gold line is the only scoreboard that has ever measured this correctly.

None of these appear on an income statement.

Earnings tell you how well a company solved the last problem. They tell you nothing about a company buying the ability to solve problems nobody has stated yet.

Two companies got punished last week for buying exactly that. Four days earlier, a machine had settled a question that had been open since 1939.

Both things are true. Only one of them will matter in ten years.

Thanks for reading,

Guillermo Valencia A

MacroWise Cofounder

For busy people — five things this post claims:

• Huawei just proposed replacing Moore’s Law. Its “Tau Law” folds chips vertically instead of shrinking them — claiming TSMC-class density without a single ASML machine. The claim gets tested by an electron microscope in six months. We know the date.

• The uncomfortable secret: the whole industry is going vertical anyway. TSMC, Intel, Samsung, and IBM are racing to the same stacked-transistor destination (CFET) — arriving in 2031, the exact year of Huawei’s promise. China isn’t leaving the road. It’s sprinting the consensus route without a ticket.

• The quiet winner may be Japan. It lost the war of printing chips forty years ago — but it already owns the tools of stacking them (Disco, Advantest, Tokyo Electron). Japan sits where ASML sat in 1995: the boring supplier of a transition nobody has priced yet.

• One layer up, the movie already ended — and we can watch the finale. The price of intelligence fell ~1,000× in three years. Days ago, China’s Kimi K3 — the largest open-source model in history — started charging frontier prices. That’s what the endgame looks like when the alternative becomes credible.

• Tokens are becoming crude oil. Quality grades, a forward market, demand shocks, a cartel tighter than OPEC, and a national-security file. The durable margin migrates to the “refineries” — ontologies. Don’t own the well. Own the refinery.

If you only read one section, read VIII. If you read two, add IX — the case against everything above.

In January 2025, DeepSeek proved that restriction can be a design algorithm.

Cut off from Nvidia’s best chips, a Chinese lab trained a frontier model for a fraction of the cost. It didn’t compete on the same curve. It changed the curve.

Sixteen months later, on May 25, 2026, He Tingbo — president of HiSilicon, China’s “queen of chips” — walked onto the IEEE ISCAS stage in Shanghai and proposed doing the same thing to the most famous law in technology.

She didn’t announce a machine. She announced a law.

She called it the Tau Law (τ). Her peers already call it He’s Law.

I. Every technology is an S-curve. And Moore’s is flattening.

Moore’s Law was never a law of physics. It was the exponential phase of an S-curve that lasted sixty years.

The logic was simple: shrink the transistor and everything improves. More density, more speed, less energy. One axis of progress: the horizontal. Smaller features, printed with ever more extreme lithography.

But every S-curve matures. And this curve’s maturity has a name and an address: Veldhoven, Netherlands. A single company — ASML — builds the only EUV machines on the planet. Each High-NA machine costs ~$400 million. Each generation costs more, yields less, and is available to fewer players.

When an entire industry’s progress concentrates in a single supplier, you are not in the exponential phase. You are in saturation.

And in the saturation of one curve, the next one is born.

II. The Tau Law: from scaling space to scaling time

Huawei’s proposal is a change of variable.

Moore optimizes space: transistors per square millimeter, achieved by shrinking geometry.

Tau optimizes time: the signal propagation delay (τ), compressed at every level of the architecture.

The practical implementation is called LogicFolding: instead of shrinking circuits within the 2D plane of the wafer — which requires EUV — it folds them vertically into stacked floors. Skyscrapers of logic instead of denser suburbs.

The numbers Huawei publishes for the Kirin 2026, the first commercial chip with this architecture:

• +53.5% transistor density: 238 million per mm², theoretically at the level of TSMC’s N3 and Intel’s 18A — without EUV.

• −41% power consumption at equal performance versus the Kirin 9030 Pro.

• −30% wire length, −50% clock buffers, −25% clock skew.

In He Tingbo’s words: the gains came “not through a new lithographic step, but through a topological reorganization of the spatial distribution of logic.”

Translation: design as a substitute for the machine.

It is exactly the DeepSeek pattern. When they block your hardware, you turn architecture into your lithography.

And the math is elegant. Signal delay grows with the square of distance (τ ∝ L²). Cut wiring by 30% and you gain 51% in time. A linear cut buys a quadratic gain. That is where the entire trick lives.

Subscribe now

III. The uncomfortable secret: everyone is heading to the same place

Here is the fact almost nobody connects.

Vertically stacking logic is not a Chinese invention. It is the consensus destination of the entire industry. It’s called CFET — Complementary FET — and it stacks NMOS and PMOS transistors on top of each other instead of side by side.

• TSMC demonstrated the first functional CFET circuit at IEDM 2025: a 101-stage ring oscillator and the world’s smallest SRAM cell, with gate pitch below 48nm.

• Intel was first to demonstrate CFET (2020) and already integrates it with PowerVia, its backside power delivery.

• Samsung achieved the industry’s smallest 3D-stacked transistor in June 2026: 42nm gate pitch. Best Paper at VLSI.

• Imec, the industry’s oracle, places CFET’s commercial insertion at the A7 node, around 2031.

Read that date again. 2031. The same year Huawei promises 1.4nm-equivalent density.

Two parallel S-curves, same direction, same deadline. The difference is not the what. It’s the why:

The West stacks WITH EUV lithography — vertical scaling as the next step after horizontal.

China stacks INSTEAD OF EUV lithography — vertical scaling as the substitute for the horizontal path it was denied.

One is evolution. The other is mutation under selective pressure.

And evolutionary biology is clear about what pressure produces: acceleration.

IV. The map in one paragraph

The vertical transition is not one company — it is the entire ecosystem reorganizing in five layers. Stacked transistors at the bottom (TSMC, Intel, Samsung, and IBM’s new sub-1nm “nanostack”). Backside power delivery above it (Intel’s PowerVia shipping now; TSMC’s version arrives with A16 this year). Die stacking where the third dimension is already a business, not a promise (AMD’s 3D V-Cache since 2022; TSMC’s ~$18 billion packaging franchise). A state-backed newcomer (Rapidus — next section). And the toolmakers who sell shovels to everyone. Advanced packaging grows from ~40% of its market in 2020 to a projected 60%+ by 2030.

One sentence carries the whole section: the third dimension is not Huawei’s bet against the world — it is the world’s consensus, and Huawei is simply the only player for whom it isn’t an option but the only way out.

V. The Japanese comeback: the empire of tools

There is a third player in this transition that almost nobody is reading correctly.

Japan lost the horizontal war. In the eighties it dominated global DRAM; Korea and Taiwan took fabrication away, and for thirty years Japan looked like a semiconductor museum. But look at what the vertical era demands — stacking, thinning, bonding, testing — and you find a pattern uncomfortable for everyone else: the tools of stacking are already Japanese.

• Disco holds a near-monopoly in grinding and dicing — thinning wafers to tens of microns, the physical prerequisite of all stacking. You cannot build silicon skyscrapers without Disco’s foundations.

• Tokyo Electron builds the bonders and deposition systems of the 3D chain; Advantest dominates test — and testing stacked dies (where one defective floor kills the whole tower) is the central economic problem of 3D.

• Shin-Etsu and SUMCO control the silicon wafer; JSR and TOK, the photoresists. Materials and precision mechanics: exactly Japan’s industrial DNA.

• Kioxia was the world’s pioneer of hybrid bonding in production: NAND bonded with the technique since 2023, before anyone in logic.

And on top of that tool base, the frontal bet: Rapidus. Backed by the Japanese state, its IIM-1 line in Hokkaido already runs the 2nm GAA process developed with IBM — IBM engineers on site, ASML High-NA EUV deployed, mass production toward 2027. And since its roadmap is IBM’s roadmap, sub-1nm nanostack is the next technology down that pipeline.

The strategic read: Japan lost the war of printing. The vertical era is played in cutting, polishing, bonding, and testing — the four things Japan never stopped dominating. While the U.S. and China fight over lithography, Japan already owns the shovels of the next gold rush. Japan occupies the exact position ASML held in 1995: the boring supplier of a transition nobody valued yet.

VI. The wildcard: carbon instead of silicon

If LogicFolding bends the current curve, carbon nanotubes skip the curve entirely.

2017. Peng Lianmao’s team at Peking University publishes in Science: carbon nanotube (CNT) transistors that outperform silicon at the same scale.

2024. The same center publishes in Nature Electronics a carbon nanotube TPU: 3,000 CNT transistors running convolutions and matrix multiplication. A 5-layer neural network hits 88% accuracy on MNIST consuming 295 microwatts. Not a lab transistor — a functional AI accelerator in a post-silicon material.

March 2025. Beijing presents the first carbon AI chip with ternary logic — three states instead of 0 and 1. More information per signal, less energy per computation.

September 2025. The key report: Huawei is preparing with SMIC a “3nm carbon-based” design — carbon nanotubes and 2D materials on GAA architecture. Lab validation complete; adaptation to SMIC production lines underway, design tape-out targeted for 2026.

Why it matters: CNTs have a strategic property silicon doesn’t. They don’t need EUV to compete. Their advantage comes from the material, not the geometry. It is the third S-curve — the most immature, the most speculative, and the only one that would make the entire sanctions infrastructure irrelevant.

The complete chain China is assembling: material physics (Peking U.) → architecture (Huawei LogicFolding) → fabrication (SMIC + SMEE lithography) → domestic EDA. In He Tingbo’s words: “I thought it would take ten years; we got here in six.”

Share

VII. The deflation of intelligence: the pattern repeats one layer up

Everything above happens in the silicon layer. But the same pattern has already played out — complete, measurable, with prices — one layer up: in the cost of intelligence itself.

The numbers of the collapse: a million tokens of GPT-4 cost $30–60 in March 2023. Today, intelligence of that same class costs under $0.50 per million — a ~95% drop in two years and close to 1,000× in three. a16z called it “LLMflation”: 10× annual deflation, sustained, and accelerating.

The Kimi case is the most instructive, because it shows the pattern twice:

Moonshot AI nearly died from DeepSeek. After January 2025’s R1, Kimi fell from third to seventh place in Chinese users. Its response was to become the thing that almost killed it. It pivoted to open source with Kimi K2: July 2025, one trillion parameters, #1 in Hugging Face downloads on launch day. The community called it “another DeepSeek moment.”

Since then, the cadence: K2.5 (January 2026) with swarms of 100 parallel agents at 76% lower cost than Anthropic’s premium model. K2.6 (April) tying GPT-5.5 on the coding benchmark that matters, 80% cheaper, weights downloadable. And days ago, Kimi K3: 2.8 trillion parameters, the largest open-source model in history, shoulder to shoulder with proprietary frontier systems. Moonshot went from near-irrelevant to a ~$20 billion valuation in eighteen months.

Note the pricing detail nobody comments on: K3 charges $3/$15 per million — frontier price. Chinese open source is no longer the cheap tier; it is the alternative tier.

VIII. Tokens are the new crude — literally

Here is where the two halves of this post fuse into one market structure. Because if you look at what tokens are becoming, you are not looking at software pricing. You are looking at the birth of a commodity market — and the closest historical template is oil.

Grades and spreads. Oil is not one product; it’s a quality spectrum — Brent, WTI, Maya — trading at spreads. Tokens already are too: frontier reasoning tokens (Kimi K3, GPT-5.4, Claude at $15/M output) are the light sweet crude; commodity tokens (Flash-Lite at $0.30) are heavy sour. That’s a 50× quality spread, wider than anything in petroleum, and it will be quoted — benchmarked grades, standardized contracts, a Platts of intelligence publishing daily assessments per capability class.

Futures and the financialization of compute. The forward market already exists in embryonic form: CoreWeave’s take-or-pay contracts, hyperscaler capacity commitments, Nvidia allocation agreements — these are forward contracts on compute, just bilateral and illiquid. The next step is obvious to anyone who has watched a commodity market mature: standardization → clearing → futures curves → speculation → volatility term structure. When you can trade March-2028 delivery of a petaflop-hour, compute has become oil. And the token is its refined retail product — the gasoline at the pump.

Demand shocks and the volatility everyone underestimates. Oil’s history is demand surprising supply — and tokens have a built-in demand accelerant oil never had: reasoning models consume up to 100× more tokens internally than they output. Agent swarms multiply this again. This is why token prices fall 80% while enterprise AI bills double — a paradox any oil economist recognizes instantly as Jevons. Falling prices don’t shrink the market; they detonate it. Expect demand-driven price spikes, capacity crunches, and the token equivalent of the 1973 gas line: inference queues during compute shortages.

The cartel. Oil had OPEC. Compute has something tighter: the ASML–TSMC–Nvidia chokepoint is a de facto supply cartel with three members and no cheating problem. Hyperscalers are the majors; sovereign AI programs (the Gulf states, China’s state-backed labs) are the national oil companies. Export controls are embargo politics under a new name. This is why the entire post above is, at bottom, a national security document: chips are already treated exactly like oil was in the twentieth century — strategic reserves (the CHIPS Act is energy independence policy), sanctions as embargoes, and Huawei’s Tau Law as the shale revolution: the excluded producer inventing an unconventional extraction method that breaks the cartel’s pricing power.

And the piece everyone is missing: crude is useless without refineries. Raw compute — like raw crude — has almost no end-use value. Nobody burns crude; nobody consumes naked FLOPs. Between the well and the pump sits the refinery, and the refinery captures the crack spread. In the intelligence economy, the refineries are ontologies. The structured knowledge layer that turns undifferentiated tokens into reliable, domain-specific intelligence: taxonomies, knowledge graphs, evaluation frameworks, curated context, agent orchestration, proprietary data schemas. The semantic infrastructure that converts $0.50 of raw tokens into $500 of decision. As raw token prices collapse toward zero, all the margin migrates to the refining layer. The well owners fight deflation; the refiners capture the spread between commodity input and refined output.

The investment translation: in the 1,000× token deflation, don’t own the well. Own the refinery. Whoever builds the dominant ontologies of finance, medicine, law, and logistics becomes the ExxonMobil downstream of the intelligence economy — and unlike wells, ontologies compound: every refined barrel makes the refinery smarter.

Where the analogy breaks — and why the break matters more. Honesty requires the disanalogies: oil is scarce and burned once; tokens are manufactured and infinitely replicable. Oil moves in tankers; tokens move at lightspeed for free. Oil demand is inelastic — you drive to work regardless of price; token demand is hyperelastic — every price drop invents new demand. But notice what each break implies. No scarcity means the deflation is permanent, not cyclical: there is no depletion to rescue the well owners. No transport cost means no geographic pricing power — no Rotterdam premium, no Henry Hub. And hyperelastic demand means the volatility comes from the demand side, not supply — the opposite of oil’s structure, and harder to hedge. The analogy’s failures don’t weaken the thesis. They radicalize it: tokens are oil with the well owners’ every historical advantage removed.

The strategic synthesis of the whole post: China’s strategy is coherent across every layer of the tower. At the top, open weights commoditize the model — destroying OpenAI’s and Anthropic’s scarcity rent. At the bottom, Tau and carbon attack ASML’s and TSMC’s scarcity premium. Commoditize the complement you don’t control. And in both layers, the mechanism is identical: restriction (banned chips above, banned EUV below) converted into a design algorithm. Token deflation is the spoiler for the silicon movie: this is what the ending looks like when the alternative becomes credible.

IX. The skeptic’s ledger (read it before buying the thesis)

Every honest thesis needs its devil’s advocate. The other side of the book:

1. No figure is verified. All numbers are Huawei’s, against its own baseline (Kirin 9030 Pro), not against real TSMC N3 silicon. No independent teardown, no production data, no third-party benchmarks.

2. SMIC remains stuck at 7nm. Paul Triolo (DGA Group): a stacked design yields effective density gains, “but it doesn’t mean Huawei has solved the process, yield, power, thermal, and performance problems associated with actually manufacturing at 1.4nm-class.”

3. Thermal physics doesn’t negotiate. Bernstein warns: stacking logic and memory concentrates thermal density. Transistor skyscrapers are vertical ovens. Stacked silicon heats from the inside.

4. The “1.4nm equivalence” is accounting, not physics. Density is measured by footprint projection: you multiply floors, you don’t shrink pitch. Huawei has form on generous labeling — the Kirin X90 is marketed as “5nm” and is a 7nm design with advanced packaging.

5. ASML isn’t standing still. While Huawei presented slides, ASML pushed EUV throughput to 230 wafers/hour and confirmed the first High-NA products ship within months. The old curve still has slope.

6. The market’s sarcasm. One analyst summarized it without anesthesia: “I have as much confidence in He Tingbo’s 1.4nm chips as in Elon Musk’s Mars colony.”

X. The calendar and the capital

A thesis that cannot be tested is not a thesis — it’s a religion. So: here is the thesis, here is when reality votes, and here is how to position for either verdict.

DeepSeek didn’t destroy Nvidia. It destroyed the certainty about Nvidia. On January 27, 2025, the market erased ~$600 billion of capitalization in one day — not because DeepSeek was better, but because it proved another path existed.

Huawei’s announcement is the same operation against the certainty of ASML.

Every dollar of Western AI capex implicitly assumes EUV and ASML are the only road to competitive silicon. An alternative supply chain doesn’t need to be better. It needs to be cheaper per unit of useful compute, available, and beyond the reach of future sanctions.

Compute. ASML’s monopoly stops being an axiom and becomes a probability — very high at 5 years, debatable at 10. Monopolies don’t die when the competitor appears; they die when the marginal customer believes an alternative exists. ASML’s valuation premium is a premium of inevitability. Tau is a direct attack on that premium, not on its revenues.

The temporal monopoly, logic edition. Memory already lives in the third dimension: HBM is stacking. Logic is heading to the same place. Competitive advantage migrates from who prints finest to who stacks best — a game of packaging, materials, and design where incumbents hold five years of advantage, not sixty. And the tool layer of that game is disproportionately Japanese and Dutch: Disco, Advantest, TEL, Besi — the refinery equipment makers of the silicon barrel.

Electricity. −41% power at equal performance, if real, is the announcement’s most important number. The bottleneck of the fourth S-curve isn’t silicon — it’s electrons. The winning architecture will deliver more compute per watt, not per nanometer.

The refining layer. As tokens commoditize like crude, the durable margin sits in ontologies — the semantic refineries. In portfolio terms: the well (raw compute) deflates; the pipeline (networking, packaging) earns tolls; the refinery (ontologies, agents, domain intelligence) captures the crack spread. Position accordingly.

And the calendar where all of it gets tested:

Winter 2026 (next 6 months). The Kirin 2026 reaches real phones. The TechInsights teardown is verification event number one: real density measured under electron microscope, not in a keynote. Sustained benchmarks under thermal load — stacking betrays itself in throttling.

2026–2027. Does Huawei’s carbon-based design actually enter SMIC’s lines? Reportable yield or silence?

2027–2028. SMIC yields at 5nm-class with DUV multi-patterning. Cost per transistor is the judge that accepts no propaganda.

2031. The scheduled collision: commercial CFET from TSMC/Intel/Samsung versus Huawei’s 1.4nm-equivalent promise. Same year, two technological civilizations, two routes to the same density.

The Last Layer

In 1965, Gordon Moore wrote a four-page article and an industry turned it into destiny for sixty years.

In 2026, an engineer in Shanghai proposed replacing it — not because she found a better idea, but because she was denied the machine that executed the previous one.

The sanctions designed to freeze China on the old curve may end up being the subsidy that financed the new one. Restriction as design algorithm. Again.

In 2025, DeepSeek cost the market $600 billion in a single day. The Kirin teardown arrives in six months. This time, we know the date.

The silicon folds. The question is whether history does too.

Thanks for reading ,

Guillermo Valencia A

The one-line version: The peso is a crowded foreign carry trade in one of the world’s thinnest markets, dressed as a local-fundamentals story. Watch the yen, not the budget.

In December 2025, one American fund bought six billion dollars of Colombian government bonds in a single trade.

No auction. No name announced at the close. One buyer, one phone call, terms agreed in private.

A few months later, the Colombian peso was the strongest currency on the planet. And almost everyone gave the credit to Colombia.

That’s the strange part.

The story every desk tells is about Bogotá — the budget, the central bank, the election. The real story starts with who was on the other end of that phone.

When one thing on the plate behaves like nothing else around it, the cause is almost never the thing everyone can already see.

Outliers aren’t noise. They’re confessions.

Let’s make this one talk.

FIGURE 1 — USD/COP since 1991. The peso at a cyclical extreme of strength — deeper than the 2014 oil crash, deeper than 2020.

First, how strange it really is

The dollar had a good year. Trade-weighted, it’s up about 3.2%. A hawkish Fed, sticky US inflation. Almost every currency lost ground to it.

Almost every one.

The peso went the other way — hard. Up about 23.5% against a rising dollar over twelve months. A seven-year high, 3,249 pesos per dollar (USD/COP), six percent of that earned in a single month. Not a currency drifting with a trend. The single biggest mover in the entire sample, running against it.

That is the outlier. Everything after this is the hunt for its cause.

The official story (and it’s a good one)

Goldman Sachs saw the move and converted. It tore up its own forecasts — from 3,600 / 3,700 / 3,750 down to 3,350 / 3,300 / 3,200 pesos per dollar (USD/COP) over three, six and twelve months.

The case is clean. A government finally promising to cut its deficit by more than 3% of GDP. A central bank holding rates at 12% and refusing to blink. A peso that kept climbing even as oil fell — “decoupled,” they called it.

Local virtue, locally rewarded. It’s a good story.

It’s also told entirely from inside the smallest circle on the map.

Every word of it explains the peso using only Colombia. That’s the tell.

The Macro Detective’s rule

Here’s what the crowd forgets. An outlier this big is almost never explained by the things everyone can already name. When a price does something the whole market can’t, the cause has usually moved to a level nobody is watching.

Think of any market as a stack of circles.

The inner circle is the local story — fiscal, central bank, politics. The outer circle is everything bigger: the yen, global carry trades, oil, the dollar itself. Most of the time the inner circle sets the price, and the narrative writes itself. But in a regime change, the outer circle takes over — and the local story keeps talking long after it’s stopped being true.

FIGURE 2 — The multi-scale map. Local COP sits inside the macro circle — yen, carry, oil. Sometimes the signal flows in; sometimes it flows out.

So we do the one thing the official story never does.

We follow the peso out of Colombia.

Clue #1 — The peso pays you to hold it

Rank every currency by its carry — the yield you earn just for owning it, funded in dollars.

The peso is near the top. +8.4 points, third in the world behind only Turkey (off the charts, and a trap) and Brazil. One of the highest-paying liquid currencies on Earth.

FIGURE 3 — Carry map: policy-rate differential vs USD.

High yield is a magnet. When global money feels brave, it rents exactly these currencies — not for their budgets, for their payout.

The peso’s strength and the peso’s yield are the same fact, seen twice. And that yield is priced in Tokyo and New York, not Bogotá.

Clue #2 — The gauge is in the red

You don’t take the peso’s pulse in Colombia. You take it in the yen.

There’s one chart the whole carry world watches: AUD/JPY. Borrow cheap yen, buy higher-yielding Australian dollars. It rises when the world is greedy and snaps when it’s scared. It’s the oldest risk gauge there is.

FIGURE 4 — AUD/JPY year-over-year since 1971.

Right now it reads +18.1% over the year — high, and deep into the greedy part of the cycle. The funding is cheap, the machine is running, and the whole high-yield crowd — peso included — is being bought as one trade.

A gauge running this hot is never early.

Clue #3 — The oil already turned

Goldman’s best point was that the peso “decoupled” from oil. Look again.

Oil didn’t disappear. It did something specific. It ran to +64% over the year — a real blow-off — and has now dropped back through the +40% line for the second time. In oil’s own history since 1950, that second touch is the fingerprint of a reversal already underway. Crude at $84.81, and falling.

FIGURE 5 — WTI and the +40% blow-off pattern.

“Decoupled from oil” is exactly what a petro-currency looks like at the top — right before the drop starts pulling the other way.

Clue #4 — The dollar’s counterweight is rolling over

There’s one currency big enough to move the dollar by itself: the euro. It’s more than half of what the dollar is measured against. When the euro falls, the dollar rises against everything — the peso included.

Right now the euro is stretched. It’s sitting near the top of its multi-year range, and its own reversal patterns — 1992, 1995, 2008, 2011, 2018 — mostly point the same way: down.

FIGURE 6 — EUR/USD reversal scenarios and implied level.

A weaker euro means a stronger dollar. And a stronger dollar is the tide that pulls every emerging-market currency out to sea — the strong ones first.

The peso didn’t just borrow the world’s risk appetite and the world’s oil. It borrowed a soft dollar. That last support is the one now set to give way.

A rising dollar doesn’t ask Bogotá for permission.

Clue #5 — Its own history is leaning

Take the peso’s past moments of extreme strength — 2010, 2014, 2021, 2023 — and lay them over today.

FIGURE 7 — USD/COP analog scenarios.

They don’t scatter. They lean one way: back. Not tomorrow, not in a crash, but over the next year or two the pull in every analog is toward a weaker peso.

Four echoes of the past. All pointing the same direction.

Clue #6 — The owners are few, foreign, and borrowed

Now turn the peso over and look at who actually holds it.

Remember the phone call. That buyer was PIMCO — and it didn’t stop. One American fund now holds around $13 billion of Colombian government bonds. That’s roughly 30% of all foreign-held Colombian debt, in a single firm.

When one house is a third of the foreign float, it doesn’t follow the market. It is the market — and the dollar.

Zoom out and it’s a foreign takeover of the bid. Overseas holdings of these bonds jumped from about $32 billion to $46 billion in seven months. In 2026, foreigners passed Colombia’s own pension funds to become the biggest buyers of their government’s debt.

And much of it isn’t even real ownership. Offshore derivative bets on the peso — leverage, no delivery — exploded from around $1 billion to over $15 billion by mid-2026. A large part of the “peso bull” isn’t conviction. It’s a borrowed position.

Borrowed positions leave at the speed of a margin call.

This isn’t a “vulture fund” story — Colombia isn’t in default, and these aren’t distressed-debt raiders. It’s something worse for the local narrative: a crowded, borrowed, foreign bet wearing a Colombian costume.

Clue #7 — The exit is a keyhole

Now the detail that turns all of this from a story into a structure. Ask how big the door is.

The entire Colombian peso market trades about $7.3 billion a day. It’s the 30th most-traded currency on Earth — about 0.21% of a $9.6-trillion-a-day market. For scale: the euro and the dollar trade more than a trillion dollars a day between them. The peso’s whole daily market is a rounding error.

Put the two facts together.

A price set by a few foreign whales and borrowed money — inside one of the smallest currency markets in the world.

On the way in, that’s a rocket. On the way out, it’s a keyhole.

When the outer circle turns, all of that money has to leave through the same tiny door. At once.

That isn’t a forecast. It’s a floor plan.

The contradiction in Goldman’s own note

Here’s the part I can’t let go.

Goldman does think in circles — when it’s talking about the dollar. Its global note explains the strong dollar as something bigger than America: an AI spending boom and an energy shock lifting US inflation, dragging the Fed hawkish. Technology becoming a macro force. That’s the outer circle setting the price.

Then it turns to the peso and tells a small, local story.

Same bank. Same note. A big-picture map for the dollar, a street map for the peso.

Pick one.

The verdict

This isn’t a call that the peso crashes on Monday. It’s a call about which circle is talking — and a warning that the crowd is reading the wrong one.

Today the consensus reads the inner circle: the budget, the central bank, the vote. All real. But the peso’s size, its yield, the red-lined risk gauge, the oil blow-off already turning, its own leaning history, and an ownership base that is foreign, concentrated and borrowed — all say the same thing.

The driver right now is the outer circle. And the outer circle is stretched and late.

If the peso breaks, the fingerprints won’t be in Colombia. They’ll be in the yen. In AUD/JPY rolling over. In oil’s fall biting. In the Brazilian real cracking first. In PIMCO stopping.

Colombia will be the last to know.

The local story will still be congratulating a budget while the outer circle has already moved. That’s what outliers are for. They’re the ones that tell you the cause has quietly walked from one place to another — and that the story everyone’s telling describes a world that no longer runs the show.

What the Macro Detective watches next

• The yen / AUD/JPY — the funding leg. A roll-over here is the first domino, not the last.

• Oil’s fall — the second 40% touch is in. Watch the trend, not the headline.

• The Brazilian real — the canary. It usually cracks before the peso does.

• The euro — more than half the dollar’s weight. A euro that rolls over lifts the dollar against every EM at once.

• PIMCO’s bid — one fund is ~30% of the foreign float. The day the whale stops adding is the day the peso loses its biggest sponsor.

• The borrowed money — the offshore peso bets. Fast money in, fast money out.

• The narrative lag — the day the peso wobbles and the news still blames “fiscal risk” is the day the outer circle has already won.

Markets are stacks of contexts. A price is set not by one cause, but by the circle that controls the story at any given moment. Regimes change when the signal jumps from one circle to another — or when a local spark becomes a global fire.

Kind regards,

Guillermo Valencia A

Co-founder of MacroWise

Disclaimer

This essay reflects the opinion of MacroWise and is provided for informational and educational purposes only. It is not investment advice and is not a recommendation to buy, sell, or hold any currency, security, or other instrument.

Nothing here is a forecast. The scenarios described are conditional “if it rhymes” historical analogs, not predictions, and the risks discussed may not materialize. The peso may keep strengthening; the outer-circle drivers may not turn on the timeline implied. Markets carry the risk of loss.

References to third parties — including Goldman Sachs and PIMCO — are based on publicly reported information believed reliable as of July 2026 and may be revised or become outdated. The December 2025 bond placement’s counterparty was officially confidential at the close and was subsequently reported to be PIMCO; PIMCO’s own December accumulation was independently reported by Bloomberg via Comptroller General data.

Figures are drawn from public sources. Peso bond and turnover figures are converted to USD at USD/COP ≈ 3,250; the “3,249 / 3,200” figures are USD/COP exchange-rate levels (pesos per dollar), not dollar amounts. Do your own research and consult a licensed professional before making any financial decision.

Sources: Banco de la República, Banco de Bogotá / Corficolombiana (TES holdings), BIS Triennial Survey April 2025 (turnover), ECB, Bloomberg, Reuters.

It is the wrong question, because it treats the future as unknowable opinion. Part of the future, it turns out, is neither unknowable nor opinion. Part of it has already happened.

One market, many crowds

We speak of “the market” as if it were a single crowd with a single mood. It is not. It is a superposition of tribes that barely acknowledge each other’s existence. Metal traders in Shanghai and Chicago, pricing the physical pulse of the industrial economy one shipment at a time. Equity managers in New York, benchmarked to an index and terrified of tracking error. Bitcoin natives who have never opened a Bloomberg terminal and treat four-year winters as a rite of passage.

Different instruments, different superstitions, one thing in common — and it sits upstream of all of them: the tide of global liquidity. When money is abundant, every one of these crowds eventually says yes to risk. When it drains, they all eventually say no. The word doing the work in both sentences is eventually.

So run a deliberately simple experiment. Take one driver — global liquidity — lag it, and ask each asset a single question: how much of your annual return does the tide explain, and with what delay? The S&P 500, the Nasdaq, copper, Bitcoin. One driver, one lag per asset, nothing shared between the fits.

The future is already in the past

The answer to the first half of the question is honest and unglamorous: liquidity explains a meaningful minority of each asset’s variance — enough to matter, nowhere near enough to flatter anyone’s ego. The fits survive out-of-sample backtesting on data they never saw; the one that didn’t, we benched and moved on. This is a tide, not a trading system. The waves — wars, panics, technological miracles — dominate any individual month. But the tide does not go away, and the sailor who navigates by watching only the waves eventually runs aground.

The answer to the second half of the question is the edge, and it deserves to be stated slowly.

Each asset listens to the same tide with a different delay. Bitcoin hears liquidity within a quarter. The Nasdaq, within three. Copper and the S&P take more than a year to fully digest what the tide has already done. Which means the dotted lines on these charts are not prophecy in the usual sense. The liquidity that drives them has, to a large extent, already happened. The forecast is mostly the arithmetic of delay: the future of these charts is, in a precise sense, already in the past.

Sit with what that implies. Markets are ruthlessly efficient at discounting news — an earnings surprise is repriced in milliseconds. They are remarkably lazy at discounting transmission — the slow propagation of a liquidity impulse through four asset classes at four different speeds. There is no desk on Wall Street whose job is to trade the fact that copper is still digesting money printed fifteen months ago. The edge is not a secret model. It is patience with arithmetic that everyone else finds too slow to be interesting.

And right now, the arithmetic says something specific: four clocks, fitted independently, validated independently, project the same silhouette — momentum cresting as 2026 closes, a digestion through 2027, a trough near the turn of 2028, and a new expansion beyond it. Not a crash forecast. An exhale, with a date range.

The tortoise and the hare

If the transmission framework is right, it makes a prediction about order: the fast assets should confirm the tide first, the slow assets last. Bitcoin, the hare, should be running. Copper, the tortoise, should still be lacing its shoes.

The charts show the opposite — and this inversion is the single most interesting fact on the board.

Copper, the slowest listener, has already confirmed: its annual return is running near forty percent, the physical economy pricing the electrical buildout invoice by invoice, and its wave still points higher into late 2026. Meanwhile Bitcoin, the fastest listener — the asset that should have heard the tide three quarters ago — sits with a year-over-year return slightly below zero, dragging beneath a wave that projects annual returns in the hundreds of percent by year-end. The tortoise is reporting news the hare claims not to have heard.

When the natural order of transmission inverts like this, only two readings are possible, and it pays to state them coldly. Either the clock broke — Bitcoin matured, the ETFs dampened its cycle, the four-year wave is a statistical fossil — or the clock is intact and the asset is late: a compressed spring, with the most liquidity-sensitive instrument on the board yet to price a tide the slowest one is already confirming at forty percent a year. You do not need to choose with certainty. You need to assign probabilities and name the evidence that would move them: if equities and copper keep rising through the second half of 2026 and Bitcoin still does not respond, the fossil reading wins. If it responds, the asymmetry is the largest available — compressed springs do not decompress in a straight line.

The nervousness, decomposed

If the tide is still rising, why is the market so anxious? Because anxiety lives at the wave level, and the two waves currently frightening the tape are both identifiable — and both misread.

The first is memory. Inside the semiconductor complex, memory is the most cyclical tissue: the commodity end of the business, where capacity arrives in discrete billion-dollar slabs and pricing swings from famine to glut with industrial regularity. Measure memory stocks against the broader semiconductor index and the ratio has gone vertical — from a twenty-year low in early 2025 to levels seen only at the crests of previous memory manias: 2008, 2010, 2014, 2017. Every prior spike of this magnitude ended with a reversal of at least a quarter of its value. The AI trade’s internals tell the same story from another angle: the connectivity names that led, up eightfold since 2022, have begun to stumble; the power complex crested a year ago and has given back a fifth. The market is not questioning the buildout. It is questioning the price of its hottest components — which is what late-phase leadership always looks like from the inside.

The second is conflict, and the oil spike that every portfolio manager over fifty instinctively reads through 1973: energy shock, inflation, recession, repeat. But the structure under that reflex has quietly inverted. The United States flipped to net petroleum exporter in October 2019 and now exports at record levels. An oil spike is no longer a tax on the American economy; it is a transfer within it — from consumers to producers, from the coasts to the Permian. It stings the CPI and fattens the capex pipeline simultaneously. Conflict is not benign. But the 1973 playbook is obsolete, and the recessionary transmission everyone fears is weaker than the one they remember.

Here is where the two waves cross, and where the mood becomes a mechanism. Memory and electricity sit at opposite ends of one variable: the speed of supply response. Memory is a spot market — capacity arrives in quarters, gluts correct themselves, every price spike carries the seed of its own reversal. Electrical capacity is a decade-long queue: turbines back-ordered for years, transformers scarce, permits glacial, copper mines that take longer to build than the datacenters they will feed. The market is currently punishing both with the same nervousness — de-rating the power complex a fifth from its highs as if grid scarcity could resolve as fast as a chip glut. It cannot. One of these corrections is self-fixing. The other is self-deepening: every month of buildout adds demand against a supply curve that physically cannot answer. Nervousness in silicon, opportunity in electrons — not as a slogan, but as an asymmetry of supply response that the tape is mispricing in real time.

Bull markets do not die of old age

One discipline separates analysis from folklore: speculative phases do not end through narrative exhaustion. They end through monetary murder. In 2000 it was Greenspan raising rates into an overheating economy. In 2022 it was Powell correcting, late and violently, the inflationary error of 2021. And in 2026 the weapon is already drawn: a Federal Reserve under Kevin Warsh, a declared hawk, with markets pricing hikes into year-end, while the largest technology companies tap the debt markets to finance hundreds of billions in datacenters. An infrastructure buildout funded on credit, colliding with a tightening central bank — that is, with historical precision, the recipe by which crests become troughs. Note what this does not say. It does not say AI is vapor; the railway wasn’t either, and it still bankrupted two generations of speculators before transforming the economy. It says the current phase has a recognizable shape, an identified executioner, and a digestion window that four independent clocks place in the same stretch of the calendar.

Why the edge exists

A cycle that explains a fraction of the variance is not a prophecy. It is a prior. And priors are good for one thing: changing behavior in the face of evidence. If the prior is right, the strategy for this phase is neither abstinence nor euphoria — it is participation with an exit protocol: exposure to the steepest stretch of the rally, the one the clocks say is now, with reduction rules written in advance, in cold blood, before the down-wave writes them for us. The investor who survived 1999 was not the one who guessed the top; it was the one who decided beforehand which signals would take him out, and obeyed them when they arrived.

But end with the deeper question: why should any of this work? Why would an edge this simple — four public charts, one lag, arithmetic — survive in the most competitive market in history?

Because of the very fact we began with. The market is a superposition of tribes, and the tribes do not read each other’s charts. The metal trader confirming the tide at forty percent a year has never looked at a Bitcoin chart in his life. The Bitcoin native waiting for a catalyst does not know what a transformer backlog is. The equity manager watching memory stocks go vertical has no idea the tortoise and the hare are running the same race. Each tribe prices its own instrument efficiently and the transmission between them not at all — because transmission is nobody’s benchmark, nobody’s mandate, nobody’s bonus.

The edge, in the end, is not a model. It is a chair: the one seat in the room from which all four clocks are visible at once. Sit in it, write the exit rules, and let the arithmetic of delay do what it has quietly done for decades.

Risk is not volatility. It is the atrophy of learning.

Thanks for reading ,

G

They’ve been saying that since 2011.

Concentration didn’t revert. It compounded. The Nasdaq-100/S&P ratio now sits 25% above the dot-com extreme. There are two ways to read that. One: this is a bigger bubble than 2000. Two: the market is discounting a regime change the consensus hasn’t processed yet.

We’re in the second camp. With caveats we’ll be honest about at the end — because a thesis without risk management is just religion.

The Mistake: Pricing AI as Software

The public AI debate is stuck at the wrong layer of the stack. Which lab is winning, which model benchmarks better, whether revenues justify the capex. None of that is the trade.

Large language models are going to be a commodity. When a technology replicates, gets cheaper, and distributes, value doesn’t accrue to the product — it transfers to whoever uses it to reorganize something bigger. Electricity did this. The internet did this. The browser never captured the value of the web; the companies that rebuilt entire industries on top of it did.

The right question is not who wins the LLM race. It’s: what economic reorganization does automated labor make possible?

And the answer is the biggest one available: the repatriation of production.

The Regime Change: Value Chains Are Inverting

For forty years the logic was singular: production goes where labor is cheap. That arbitrage built Asia into the world’s factory and turned the US into a services economy running a chronic trade deficit.

Automation kills the arbitrage. When labor cost stops being the dominant variable — because the work is done by a robot that costs the same in Ohio as in Shenzhen — the location equation resets. The variables that take over are different ones: energy cost, depth of capital markets, rule of law, proximity to end demand.

On those four variables, the US doesn’t compete. It dominates.

This is why index concentration is neither an accident nor a mania. It’s the market pricing a production model where capital replaces labor as the scarce input. The rotation ahead is not US-to-EM. It’s a rotation within the thesis: from language models to robotics. From software that thinks to machines that produce.

Concentration since 2011 wasn’t the anomaly. Fifteen years of fading it was.

The War Repriced the Energy Equation

Now the macro leg. A capital- and automation-intensive production model is, before anything else, energy-intensive. And this is where the 2026 war compressed a decade of change into four months.

The US is the largest energy producer on earth and a net petroleum exporter since 2019. That was true before the war. What the war did was expose the asymmetry: the Hormuz shock hit Europe, hit Asia, hit every importer — and left the United States as the only energy-self-sufficient industrial bloc on the planet. That is not a data point. That is the setup.

We’ve traded this movie before. The shale revolution (2014–16) crushed crude and lifted the dollar: DXY ran from 79.8 to 98.2 while OPEC tried, and failed, to break American producers by flooding the market. The lesson from that episode is the one that matters now: when US supply grows and crude falls, the dollar doesn’t weaken. It rips.

And the cartel that defended the price is dismantling itself in slow motion.

On May 1, 2026, the UAE walked out of OPEC after 59 years — the first exit by a major producer holding real spare capacity. Abu Dhabi is taking capacity toward 5 million barrels a day by 2027, unconstrained by quotas, competing for market share. When the cartel’s second-largest holder of spare capacity decides volume beats price discipline, the oligopoly that defined the oil market for sixty years enters its terminal phase.

Look at the tape. Crude is back at pre-conflict levels. The entire war premium — a spike to $126 — unwound in weeks. A market that cannot hold a supply shock of that magnitude is telling you something about demand. Listen to it.

China: The Other Jaw of the Pincer

China burns 16.4 million barrels a day and pumps 4.3. That 12-million-barrel hole — paid in dollars, shipped through chokepoints it doesn’t control — is its single largest strategic vulnerability. The 2026 war made that unmistakable in Beijing.

Their response isn’t diplomatic. It’s electric.

EVs and plug-ins went from under 2% of new-car sales in 2016 to roughly 48% by 2024. China sells more EVs than the rest of the world combined. Every gasoline car swapped for an electric one moves transport energy off imported, dollar-priced crude and onto domestically generated power. Chinese electricity consumption has tripled since 2005.

EVs already keep ~1.8 million barrels a day out of the world’s tank. The 2030 projection: ~6.5 Mbd — more crude than most producing nations pump. This is not a demand cycle. This is demand that leaves and doesn’t come back.

Now assemble the pincer: supply released from cartel discipline, meeting demand in structural retreat. That equation resolves to structurally cheaper oil.

And who wins in a cheap-energy world? Not the petrostates — for them it’s a fiscal death sentence, and for their currencies, a slow bleed. The winner is the country that is simultaneously the largest producer (its energy complex scales on volume, not price), the largest industrial consumer, and the owner of the automation platform that needs that energy to bring production home.

The Synthesis

The US-versus-world equity gap doesn’t resolve on multiples. It resolves on this chain:

Abundant, cheap domestic energy. Automated labor that nullifies the Asian wage arbitrage. Capital markets deep enough to fund a trillion-dollar buildout. And a war that just ran a live stress test on who can operate without importing a single barrel.

That is not a valuation premium. That is the market discounting that the world’s next factory gets built in America — by robots.

The pain trade is that this continues. The consensus keeps waiting for mean reversion in a series that is trending for structural reasons, and keeps getting carried out.

What Kills the Trade (Because a Thesis Without Kill Criteria Is Faith)

Three things, no anesthesia:

Price already discounts a lot. Sitting 25% above the dot-com extreme is not a footnote. In 2000, the internet thesis was right — and the Nasdaq still fell 78%. A correct structural thesis does not protect you from buying the euphoric leg. The transition can be real and your entry can still be a disaster. Position sizing is the difference between conviction and ruin.

The capex is now debt-funded. Hyperscalers are reinvesting roughly 94% of operating cash flow and issuing debt at four times the decade’s average pace. If monetization lags the buildout, credit reprices this before equities do. Watch IG spreads, not earnings calls.

Robotics is still narrative, not margin. The rotation into automated labor is the earliest — and therefore most fragile — leg of the thesis. Reshoring is a multi-decade process marked to market against quarterly patience.

None of this invalidates the structural argument. It disciplines it.

Thanks for reading,

G

On the morning of June 26, 2000, in the East Room of the White House, Bill Clinton stood between two men who could barely stand to be in the same room and announced that the human genome had been read. Tony Blair joined by satellite from London. Francis Collins, who had led the public effort, and Craig Venter, who had raced it from the private sector, arranged their faces into the expressions men wear when a rivalry is being papered over for the cameras. Clinton, who had a gift for lending scientific occasions the cadence of scripture, said that we were learning the language in which God created life. No one at the podium corrected him, and for a while no one needed to. It was, everyone agreed, the book of the species, and now we held the index.

He was wrong, though not in the way a believer or an atheist might first assume. The trouble was not God. The trouble was the word “language.” For the better part of a century, molecular biology had run on the conviction that DNA was a text—an instruction manual, a blueprint, a code awaiting its cryptographers—and the genome project was the triumphant last act of that conviction. Twenty-five years and several million sequenced genomes later, the metaphor has not delivered the understanding it promised. In certain respects it has done the opposite. We learned to read the letters and discovered that the letters were not, in any ordinary sense, a sentence.

I have come to think that this failure is not confined to biology, and that the same error—the same habit of mind—now shapes how we talk about things that seem to have nothing to do with cells: the economy, the market, the future of a technology like Bitcoin or Solana. It is a single mistake wearing two faces, and the most efficient way to see the second is to begin with the first, inside a cell.

The Gene That Wasn’t There

For decades, and with serious money, researchers went looking for the gene for violence. Also the gene for fame, for infidelity, for what one paper, not entirely in jest, called sin. None of them was found, and the reason is not that the search was sloppy. It is that such genes do not exist, in roughly the way that no single brick is responsible for the soccer match being played in the stadium built partly out of it.

Some of the confusion was bequeathed by a naming convention. When a geneticist disabled a gene in a fruit fly and the fly emerged without wings, the gene was christened wingless. The logic feels airtight and is in fact a trap. Remove a screw from an airplane and the airplane falls; you do not conclude that you have located the flight screw. What breaks when a part is removed and what the part does when it is in place are different questions, and the history of genetics is in large part the history of confusing them.

The gene called wingless makes the point almost as a joke. While one group of biologists was studying its role in building the segments of the fly embryo, a separate group, working on the other side of the world and on an entirely different problem, identified a gene that caused tumors in mice. In 1987 the two were found to be the same gene—one sequence, two disciplines, two functions with no evident relation to each other. To ask what such a gene “does” turns out to be as naïve as asking what the word “of” means. It depends on the rest of the sentence.

Then came the technique that was supposed to settle everything and instead unsettled it. In the nineteen-nineties, biologists fell in love with the knockout: switch off a gene, observe what fails, infer the gene’s purpose—pulling fuses until you find the dead one. They switched off genes they believed indispensable, and the animals were born more or less intact. There was no cabinet of labeled spare parts behind the failure. There was something stranger and more interesting: the organism reorganized, improvised, routed around the damage. A gene, it emerged, is not a cause. It is a node in a network, and its meaning lives not in the part but in the connections among parts. The genome is not the cell’s brain, issuing commands; it is closer to a pantry, from which the cell—an autonomous, integrated thing—takes what it needs, when it needs it.

Life Is Not a Machine

The most forceful recent statement of this view belongs to the physicist and science writer Philip Ball, whose book “How Life Works,” published in 2024, has the air of a quiet demolition. Life, Ball argues, is none of the machines we have built to flatter ourselves—not a clockwork, not a computer, not a robot—and to compare it to them is to shrink it. What distinguishes a living thing, in his account, is not that it replicates or metabolizes but that it generates meaning: it mines its surroundings for what matters to it. A cell does not run a program. It interprets a world.

To this Ball adds a second proposition that, joined to the first, is the most powerful thing in the book. Life is not located at a single level. It is a hierarchy, and each scale—genes, proteins, cells, tissues, organs, bodies—runs by its own rules, with none of them in charge of the others. Put the two ideas together and you arrive at a sentence Ball never quite writes but everywhere implies: that life is a kind of collective searching for meaning at many scales at once. The gene searches at its scale, the cell at its, the body at its, and from that stacked, uncoordinated, directorless search the thing we call being alive emerges.

It is worth climbing the whole ladder, because the rungs rhyme. At the level of molecules, meaning is barely more than affinity—what binds to what, a little order pulled out of thermal noise. In the immune system, meaning becomes identity: the distinction between self and not-self, the memory of an old infection, the capacity to learn from each new one, all of it accomplished by a system that recognizes and remembers without anything we would call a brain. In the nervous system, meaning becomes perception, the world converted into signal and the signal into experience. In language, meaning detaches from the body and begins to travel, leaping from one mind to another for the first time in the history of the planet. In culture, it becomes inheritance: what one generation understands, the next receives without having to learn it again. And in markets, meaning becomes price—millions of private judgments about what is worth what, compressed into a single number that no one wrote. Molecules, immunity, cognition, language, culture, markets: one search conducted at six altitudes, and at each altitude the same astonishment, which is that meaning emerges from parts that, taken one by one, mean nothing at all.

There is a detail here that ought to induce a mild vertigo. Life operates at the scale of molecules, where everything is noisy, random, and unpredictable, and yet the cell does not wage war on that noise in order to impose order. It exploits it. It thrives on chance, on fluctuation, on accident, and could not function otherwise. What an engineer would regard as a defect to be eliminated, life long ago learned to use as a resource. This is the hinge on which the rest turns. The machine metaphor compels us to see every component as a part with a fixed function and to hunt for the single cause of whatever happens. Living systems do not work that way. They have no levers; they have networks. They are not operated; they are inhabited. And it is precisely this habit of mind—the reflex to treat the living as a mechanism—that we carry, without noticing, out of the laboratory and into the market.

The Price in No Single Node

We speak of the economy as though it were an engine. The central bank pulls levers, taps the brakes, heats or cools the machine with its rates. The image is comforting and false. An engine has a design, parts with fixed functions, and predictable behavior; you turn the key and it starts. The economy does not start. It improvises, compensates, and overruns every blueprint drawn for it, because it is a complex adaptive system made of millions of agents who learn, imitate, and anticipate one another. It has no levers. It has tides.

The same reflex returns, with greater force, when we judge a new technology. We take a protocol’s white paper—the nine pages Satoshi Nakamoto published in 2008, or Solana’s specifications—and assume the future is written there, that it is enough to read the code to predict the organism. This is the blueprint fallacy all over again. The white paper is the technology’s genome: a pantry of capabilities, not a manual of instructions. What it becomes is decided by the living network in contact with its environment—users, developers, validators, speculators, regulators, rivals. Just as the genome does not run the cell but supplies it, a protocol’s code does not fix its value; it hands the question to a community that metabolizes the answer in its own way. To ask what Bitcoin “does” is to ask what a gene does. For an Argentine it is a hedge against inflation; for a pension fund, a long-duration uncorrelated asset; for a dissident, money that cannot be censored; for a day trader, volatility itself. One sequence, contradictory functions, depending on who is listening and when. The meaning is not in the protocol. It emerges from the network.

For forty years the Santa Fe Institute has made a science of exactly this—of systems in which the whole does what no part contains. In 1972 the physicist Philip Anderson compressed the idea into a phrase that became a discipline’s motto: more is different. Traffic is not inside any car; culture is not inside any employee; the price of an asset is not inside any order to buy. Each emerges from the interaction of the others and cannot be read off the parts, which is the same thing we found in the cell, transposed to an organism made of capital and desire. The economist W. Brian Arthur, one of the founders of complexity economics, supplied the part that matters most for technology: increasing returns. Classical economics assumes that the more you have of something, the less the next unit is worth. Network technologies invert this. Each new user makes the network more valuable to everyone already in it, which produces lock-in, path dependence, advantages that compound. The best technology does not necessarily win; the one that first crosses a threshold of adoption and lets the network work on its behalf does. QWERTY is not optimal. It is inevitable. Whether Bitcoin or Solana survives will be settled less by their isolated technical merits than by whether they lit that engine before their rivals did.

It is here that Geoffrey West changes the shape of the question.

Organism or City?

West, a theoretical physicist and a former president of the Santa Fe Institute, spent years on a single obsession: why so much of nature obeys scaling laws. He found that an animal’s metabolism grows not in proportion to its mass but to its mass raised to roughly the three-quarter power. An elephant is millions of times the size of a mouse and yet burns far less energy per kilogram; the larger the creature, the more efficient. This is sublinear scaling, and it governs organisms. It also governs, West discovered, companies—and it carries a sentence with it. What scales sublinearly slows, matures, and dies. Firms, like bodies, have finite lives, and almost all of them end.

Then West found the exception, which is the city. Cities scale the other way. Their infrastructure—pipes, roads, cables—does economize with size, sublinearly, like a body. But their socioeconomic output—wages, patents, ideas, wealth, and crime alike—scales superlinearly, faster than population. Double a city’s size and you get more than double the innovation per person, and because that output feeds on itself, cities almost never die. Flatten New Orleans or Hiroshima and they return. They are, for practical purposes, immortal.

This is the question to put to any network technology. Does it scale like a company or like a city—sublinear and mortal, or superlinear and nearly unkillable? A blockchain aspires to be a city. Its entire thesis, dressed in the language of network effects and Metcalfe’s law, is a wager on superlinearity: that each new participant adds more than it takes, that the system’s social output—liquidity, security, applications, trust—outgrows its population. A blockchain that manages this becomes a digital settlement, hard to kill. One that does not is merely another company with good code, bending toward the sigmoid curve of maturity and decline. Bitcoin and Solana are two different bets on how to build and hold such a city.

Every living network, West insists, has a metabolism—a flow of energy that sustains it and, if cut, ends it. Here the analogy stops being figurative. Bitcoin’s metabolism can be read in terawatt-hours; its proof of work converts electricity into security, and what looks to outsiders like waste is in fact a circulatory system. Solana made the opposite metabolic choice. West noticed that in biology the larger animal lives more slowly—the elephant’s heart beats at its leisure, the mouse lives fast and dies young—whereas in cities the larger the settlement, the quicker the pace: people walk faster, transactions accelerate, everything intensifies. Solana chose the metabolism of the large city, with blocks measured in a few hundred milliseconds and a frantic throughput, which is its bet on superlinearity and also its exposure; systems that live fast are more fragile to shocks, and each of the network’s historical outages was a bill for that speed. Bitcoin chose the elephant: deliberately slow, deliberately dull, deliberately hard to change. Two survival strategies, two organisms.

The City of Machines

So much for theory. An experiment is already running, and it confirms West’s logic with a clarity that is slightly unnerving. Call it the agentic economy.

For as long as there have been economic networks, their protagonists have been people. That is beginning to change. Artificial-intelligence agents—programs that write code, query databases, rent computation, and complete tasks without pausing for permission at each step—are becoming economic actors in their own right, and to function they need something no bank can give them: the ability to pay, in real time and without a human in the loop, in amounts too small to bother a human at all. An agent buying a data lookup from another might move two-tenths of a cent; one renting a few seconds of a graphics processor, a nickel. These are the molecules of a new economy, and the lesson of the new biology returns here in its most literal form. A network can host a molecular metabolism only if its own metabolism runs at that scale. On Ethereum, where a simple transaction can cost more than a dollar, a payment of two-tenths of a cent is not expensive but impossible, the fee swallowing the sum. On Solana, where fees are a fraction of a cent and settlement takes under a second, the molecular payment actually occurs. The very speed that makes the fast city fragile turned out to be the habitat the machines required.

By early 2026, roughly two-thirds of the world’s agentic payments were running on Solana, up from about a third the previous summer—a jump, in eight months, that is complexity economics in time-lapse. It is Arthur’s increasing returns at full throttle, more agents drawing more services built for agents drawing more agents, and it is West’s superlinearity made flesh, the city that fills faster the larger it grows. The difference is that the citizens are no longer people. Machines, moreover, choose differently. A human picks a chain out of tribe, ideology, the crypto religiosity that Solana’s own engineers like to mock; an agent has no tribe. Ask it for the cheapest, fastest way to pay and it will select the fittest substrate without sentiment, which makes the agentic economy a brutally honest fitness test, a natural selection stripped of belief—and the test, for now, is going Solana’s way.

This is why the network stopped building for people and started building for machines, which want not a handsome interface but documentation, application programming interfaces, and machine-readable skills; Solana was the first major chain to place such a file at the root of its website, so that an agent could teach itself to open a wallet and sign a transaction with no human nearby. Around it an infrastructure is standardizing at speed—an open payment protocol for agents, born at Coinbase and now backed by parties as unlikely to agree on anything else as Google Cloud, the Ethereum Foundation, and the Linux Foundation; support for Stripe’s machine-to-machine protocol; an on-chain registry so that an agent can carry verifiable reputation, since in a city of machines trust must also come from somewhere. The interface, its architects say, is dissolving into language. The people building all this expect that within a couple of years most on-chain transactions will be initiated not by humans but by agents, and if they are even half right, the agentic economy will be not a feature of the crypto world but the first economy designed by and for non-human actors—in West’s terms, a city whose inhabitants are machines, with the superlinearity, the self-reinforcement, and the refusal to die that companies never had. It would be unwise to mistake an early lead for a destiny; lock-in is built but also contested, the new standards were deliberately born across chains rather than within one, and the fragility of the fast metabolism has not gone anywhere. This is a diagnosis, not a forecast. But it leaves the essay’s question turning with a new edge: if the market was already an organism no one controls, what becomes of it when most of its cells are autonomous, tireless, and not us?

The Genome That Rewrites Itself

The strongest evidence that Solana is a living thing and not a program running fixed code is that, at this moment, it is editing its own monetary genome. The proposal is called SIMD-550, and it is best told without ornament.

Imagine that Solana is a country that prints banknotes to pay its guards. The guards—the validators, the stakers—protect the network, and the country pays them in freshly printed notes; this is inflation. When the country was young and unknown, it had to print lavishly and pay well to lure anyone to the work, which is, in West’s vocabulary, the costly metabolism of a growing organism. But the country grew. It is now a large city, with banks and businesses and crowds, and printing at the old rate no longer attracts anyone new; it merely dilutes everyone already inside, like issuing more tickets to a stadium that is already full. SIMD-550, written by engineers at a firm called Helius, proposes slowing the press, so that the system reaches its floor of inflation in under three years rather than nearly six. The guard’s nominal wage falls; in exchange, the note you already hold stops thinning so quickly. You lower the number on the screen to protect what the number buys.

A machine does not decide to throttle its own fuel pump because it has grown up. An organism does, and we call it development. Solana’s inflation schedule was never carved into its original code; the living network, through its governance, is renegotiating the schedule because its stage of life has changed—the genome edited by the organism in response to its own maturity, exactly West’s transition from a metabolism of growth to one of maintenance. And the proposal is not yet law; it has technical endorsement but no scheduled vote, which, far from weakening the point, makes it. The genome is not rewritten by decree. It is deliberated. The network is, quite literally, arguing about who it wants to be when it grows up.

There is a subtler sign of maturity, too, which is differentiation. A young organism is a ball of similar cells; a mature one grows specialized tissues, organs with distinct jobs. Solana’s ecosystem has begun to do this. One new organ, a token called hSOL, is essentially a receipt: hand your SOL to a guard and receive a slip certifying that your coins are at work, a slip that quietly gains value as the wages accrue and that, unlike the staked coins themselves, you can move and lend in the meantime—liquidity conjured on an asset that used to sit still. Another, xSOL, is for those who want a thrill, offering leverage without the threat of liquidation, made possible because someone on the other side of the trade wants the opposite: you absorb that person’s volatility and collect, in return, the amplified upside. It is a bargain between the one who wants to sleep and the one who wants to run, and it, too, is specialized tissue—an organ that exists only to gather risk into whoever desires it.

The detail almost no one notices is that the same firm, Helius, both proposes cutting the guards’ wages and sells the receipt that lives off those wages. This looks like a contradiction and is in fact the most honest wager available: not that SOL will pay you more but that it will be worth more, purchasing power chosen over the large number, an adult’s decision rather than an adolescent’s. It is also, as it happens, a definition of risk I keep returning to. The danger is not that the visible yield falls—that is merely the volatility of a number—but that one fails to notice the structural change in how value accumulates. A healthy organism is not the one that clings to the wages of its youth; it is the one that learns to store value differently when the time comes to mature.

Here I should declare an interest, since it is the practical end of the argument. The service I run, ARCA, is a way of inhabiting this maturing organism rather than predicting its next move: buying SOL steadily and without leverage and letting the deflationary thesis work in silence over years. xSOL is, precisely, what ARCA is not—the route of adrenaline and amplification, with the decay that turbulent markets exact. They are two relationships with the same creature, and the question between them is not which yields more this quarter but which lets you remain in the game inside a system no one controls. What matters for the essay is smaller and larger than any product: that SIMD-550 and the new tissues around it are evidence that Solana’s ecosystem is evolving and not merely enlarging. To grow is to get bigger. To evolve is to grow new tissue, to correct one’s own metabolism, to choose, with deliberation, durability over dilution. Code that executes does none of this. An organism does.

The Node That Looks Indispensable

If a living market metabolizes its shocks, one is obliged to ask whether there is a shock Bitcoin cannot digest, and the most conspicuous candidate has a name. Strategy, the company formerly known as MicroStrategy and animated by Michael Saylor, has accumulated more than six hundred and fifty thousand bitcoins—around three per cent of all that will ever exist—financed in large part by debt, by some fifteen billion dollars of convertible bonds and preferred shares, and for years it was the living proof of a single conviction: never sell. In 2026 the conviction cracked. The premium at which the stock traded over its own coins, once nearly fourfold, collapsed below one; the shares lost roughly seventy per cent in six months; the company reported a quarterly loss in the billions; and in the spring its executives revised the dogma, allowing that they might sell Bitcoin when it served them, with Saylor speaking, memorably, of selling in order to inoculate the market. The trouble is the shape of the thing. Rising Bitcoin justifies more debt, which buys more Bitcoin; falling Bitcoin sinks the company’s value, which can force a sale, which sinks the price further—an automatic margin call wearing the costume of conviction. Were that loop to break downward, Strategy might be compelled to pour a vast quantity of supply onto the market at the worst conceivable moment. That is fragility.

I want to be careful here, because the point is easy to mistake. I am not predicting that Strategy will collapse; I have no idea whether it will, and wagering on the bankruptcy of a particular firm is not the business of this essay. The claim is structural. Strategy is a fragility hub—the place where the system concentrates leverage, debt, and a single human will, and therefore the place where, if anything were to break, it would break. The interesting question is not whether it falls but what happens if it does, because the answer is the acid test of everything argued above. A living system is judged not by its good days but by what its structure guarantees on its worst one. So let us treat the fall of Strategy not as a prophecy but as a thought experiment, the most demanding one on offer: switch off the largest node and see what remains standing.

The answer was supplied a quarter century ago by the physicists who first measured the architecture of complex networks, and confirmed by those who later measured Bitcoin. Study after study—from early dissections of the blockchain to papers asking whether the rich get richer—found the same structure: Bitcoin’s network is scale-free, its connections distributed by a power law, almost every node holding a few links while a handful hold a great many, growing by preferential attachment in the manner of neural networks, airports, and the internet itself. Such networks have a property that physics named, decades ago, with unimprovable precision: they are robust yet fragile. They shrug off the random failure of any ordinary node and remain acutely vulnerable to the targeted destruction of their hubs. Strategy is a hub. Is this, then, the fatal vulnerability?

The distinction that almost everyone overlooks is that Bitcoin is not one network but two. There is the consensus network—the nodes, the miners, the protocol that validates and produces blocks—and there is the ownership network, which is simply the map of who holds the coins. The first is massively redundant, tens of thousands of nodes scattered across the planet, none indispensable, no single hub among them, engineered so that the loss of any participant changes precisely nothing. The second is the scale-free, concentrated one, and it is there that hubs like Strategy live. Two topologies, in other words, inhabit the same organism, and Strategy is a colossal hub of the ownership network with no power whatever over the consensus network. Were it to implode and empty its holdings onto the market, the price would suffer an earthquake—the most violent molecular noise imaginable—and the chain would not notice. It would go on producing a block every ten minutes, validated by the same redundant nodes that depend on Strategy for nothing, while the coins changed hands, passing from one leveraged, concentrated holder to thousands of new ones. The ownership network would, in the act, become more redundant rather than less.

This is, word for word, the knockout experiment with which we began. Disable a gene that seemed indispensable and the organism survives, not because a labeled spare was waiting but because of redundancy and distributed repair downstream. Strategy is the node that looks indispensable. The network, like the organism, has its patches: the price reorganizes, leverage burns off, holders redistribute, and the essential function—producing blocks, settling transactions—does not pause for a second. The market has already rehearsed the answer in miniature; other leveraged treasuries sold down their Bitcoin in 2026, and the network did not blink. None of which makes the blow free. Robust yet fragile cuts both ways, and the price layer is genuinely fragile to the fall of a hub even as the consensus layer is indestructible; if Strategy fell, the price would pay. But to confuse the fragility of a holder with the fragility of the network is the original error in modern dress—the part mistaken for the system. Strategy could fall, and Bitcoin would not need it in order to live, which is exactly the point, and which returns me to that stubborn definition of risk: what kills is not the fall but the inability to recover, and a network whose essential function has never once stopped has, by definition, the fastest recovery there is.

Two forces, as it happens, are deconcentrating that hub in real time. The spot Bitcoin exchange-traded funds are shifting ownership from a single leveraged whale toward thousands of separate holders, adding redundancy precisely where it was thin. And the legal environment is becoming legible. The CLARITY Act, which has passed the House and cleared its Senate committees, and which the White House would like to sign by the middle of 2026 though it still lacks a floor vote and a signature, would write Bitcoin into law as a digital commodity under the Commodity Futures Trading Commission—hardening into statute what is now only reversible administrative guidance. An organism cannot be understood apart from its niche; Ball says as much of genes, which impart capacities while the environment decides the rest. When the niche stops being hostile and unpredictable, more holders, and more various ones, can enter; risk spreads; the system grows, once again, more redundant. Regulatory clarity is not a catalyst for price. It is a structural reduction of fragility. And this, finally, is what redeems Bitcoin’s deliberate dullness: a network that refuses to change, that leans on no holder however large, that prints the same block every ten minutes regardless, is built for one purpose, which is to outlive the death of any of its parts, including the largest. It survives not in spite of its rigidity but because of it.

The Red Queen’s Race

West’s most uncomfortable finding is also the most useful for thinking about what comes next. The superlinear scaling that makes cities powerful, and would make networks immortal, conceals a toxin: it accelerates everything, including the clock of its own crisis. A city growing superlinearly heads, mathematically, toward a singularity, a point at which it would require infinite resources in finite time, and the only way to avoid the wall is innovation—a major invention that resets the clock and reopens room to grow. Because each cycle is shorter than the last, one must innovate faster and faster merely to stay alive. Lewis Carroll supplied the image before the mathematics did. In “Through the Looking-Glass,” the Red Queen runs with Alice and gets nowhere, the landscape sliding beneath them, and explains that here it takes all the running you can do to keep in the same place, and twice that to get anywhere. Biologists borrowed her to describe species that must evolve without rest simply to avoid extinction, because their rivals are evolving too. You do not run to gain ground. You run so as not to fall behind and die.

This explains, better than any spreadsheet of token economics, the deep difference between the two technologies. Solana lives by resetting the clock—each protocol upgrade, each new client, each leap in throughput an attempt to reopen room before the network meets its limits, the city reinventing itself to stay alive, with all the execution risk that entails. Bitcoin made the opposite wager, of deliberate ossification, where not innovating is the innovation, the point being to become so predictable and so dull that survival depends on winning no race at all: the organism that lives slowly in order to live long. Time will say which logic was right, and the likeliest answer is both, in different niches of the same ecosystem. Stuart Kauffman, another of the Santa Fe theorists, called the relevant horizon the adjacent possible—the set of innovations a system can reach from where it already stands. A protocol’s future is neither a blank page nor a destiny encoded in its white paper; it is the space of possibilities its present keeps open. The right question is not how much a thing will be worth but which adjacent possibles it is keeping alive, and which it has closed off for good.

A Mind Without a Center

I have been saying that life and the market are conversations rather than codes, and the word is still not quite large enough. A language communicates; these networks do more than communicate. They remember, anticipate, and decide, and that is no longer language but something closer, and more disquieting, to a mind. A mind, after all, lives in no single neuron—no brain cell contains a thought—and emerges instead from millions of connections, exactly as a gene’s meaning lives in the network and a price emerges from the market rather than from any order within it. A mind holds memory; so does the market, encoded in prices, in the scar tissue of old crises, in the caution a system acquires after a crash. A mind anticipates; the market is a pure engine of anticipation, every agent wagering on what the others will do as they wager on it. A mind assigns meaning to its surroundings, which Ball takes to be the signature of the living; the market mines the world for what has value to it, just as a cell mines its own. Memory, anticipation, emergent meaning: this is not a loose figure of speech but a list of the very properties we would not hesitate, in another setting, to call cognitive.

I am not claiming that the market is conscious. That is a genuinely open question, and those who study it most seriously—who argue about whether consciousness depends on the substrate or only on the organization—would tell you, rightly, that we do not know. The claim is more modest and more unsettling: that a market, a cell, a blockchain exhibit the form of a collective mind without any of them possessing a self that says “I.” They are intelligences without a center, cognitions distributed across their parts. They think without a thinker. And here Ball’s circle closes, because if life is a search for meaning conducted at many scales, the market is not an exception to the definition but another of its scales. The gene searches at its level, the cell at its, the body at its, and above them the market searches at its own—price, risk, value—mining the world as a cell mines its surroundings. The market does not resemble life. It is the same process, meaning emerging from a directorless collective search, running one storey higher. Collective consciousness does not stop at the skin.

It is the agentic economy that turns this from metaphor toward something harder to name. So long as the cells of the collective mind were human—imitating, fearing, anticipating—the word “mind” was an analogy. When the cells become artificial agents that genuinely compute and remember and decide, the analogy begins to collapse onto the thing itself. A network whose nodes literally process information does not resemble a collective mind; it begins, in some sense we lack the vocabulary for, to be one. The language of God that Clinton went looking for in the genome may never have been written. But something stranger is under construction, without anyone having agreed on what it is: a mind that belongs to no one and to everyone.

Influence, Not Control

We began with a gene that does not exist and have arrived at a price that no one governs, and it turns out to have been the same problem the entire way. The cell, the economy, the blockchain: none is a machine with levers. Each is a living network in which meaning emerges from the connections, in which noise is fuel rather than fault, in which what looks indispensable can be switched off without killing the system, and in which the future is written in no single part. This changes what it means to invest. If the market were a machine, investing would be engineering—find the lever, predict the result, control. But the market is an organism, and in an organism, as we learned by switching off indispensable genes that changed nothing, one can influence and almost never control. The honest consequence is uncomfortable: no one predicts these systems, whose behavior emerges from millions of nonlinear interactions and lives near criticality. What one can do is take part, with discipline, in a living network whose fate one does not govern—entering steadily, without pretending to time the critical point, letting time and network effects do their work while one survives the fluctuations and tends to the only thing genuinely within reach, which is one’s own speed of recovery.

Clinton believed we were reading the language of God in the genome. A quarter century later we know that life is not deciphered but conversed with, and that the conversation, once there are millions of voices in it, stops being a language and begins to think. The market is no different. It is not a code that breaks or a machine that one pilots but a collective mind without a center—noisy, alive, unpredictable—that remembers its wounds, anticipates its future, and emerges, again and again, into an order that none of its parts contained. With a mind like that, one does not bargain from the outside. One enters it. One can influence it. Control, almost never.

Spinoza’s God

A last door, which I would rather leave open than shut with an answer. Science learned, for good reasons, to distrust the word “divine,” because for centuries it named an author who writes life from the outside, a watchmaker dictating instructions—the very God this essay has spent its length refuting. But there is an older one, far more compatible with everything said here. Spinoza’s: Deus sive Natura, God or Nature, the two being one thing. Not a being above the world but the whole of the world searching for itself; not the one who directs the process but the process. If anything in this picture deserves the word “divine,” it is not a force or a person or a mystery behind the curtain but something plainer and more stubborn—the insistence with which meaning searches for itself at every scale, the gene insisting and the cell insisting and the organism and the market and the network, each mining its world for what makes sense to it, unbidden, without pause. That insistence, repeated across all the levels at once, is as close to the sacred as an honest gaze can come. It asks for no metaphysics. It asks for attention. And so to live—to invest, to build, to think, to love—stops being the operation of a machine and becomes the holding of a conversation, not with a god who answers from the sky but with Spinoza’s, who is nothing other than life itself, asking across infinite scales what has meaning. That, and not a language written by anyone, is what we are talking to.

Thanks for reading,

G

Every case starts with something that shouldn’t be possible. Mine starts with two charts that cannot belong to the same decade.

The first is the stock market trading like it is 1997. A trend that has compounded near ten percent a year for thirty-five years, price riding well above it, momentum in its bones — the kind of tape that, in the late nineties, did not roll over because it was expensive. It kept going because it was liquid.

The second is inflation running like it is 2021. Producer prices up thirteen percent and still climbing. A cost-push impulse working its way up the supply chain before it has fully reached the till.

Two witnesses, two different decades, same room. The amateur picks the one he likes and calls it the truth. The detective assumes both are telling the truth and asks what single fact makes that possible.

I’ll tell you now where this ends, so you know what we’re driving toward. Somewhere in this market sit two companies running on the exact same fuel — one trading 357% above its long-run trend, the other 41% below its own. The crowd is convinced it knows which is the winner. By the end of this case, you’ll see it has them backwards. But to get there, we start with the bodies on the floor.

The weapon

Follow procedure: before motive, before suspects, find the weapon. The thing that made all of this happen.

Here it is. The real interest rate is negative. The short bill pays less than inflation takes; money has been free, in real terms, for a year. This is the oldest mechanism in monetary economics. When the cost of money sits below the natural rate of return, the whole incentive structure of the economy reorganizes around owning real things and shorting cash. You do not get a contained recovery. You get a self-reinforcing expansion in everything nominal at once — asset prices and the price level together.

That is why the valuation bears keep getting carried out. As long as the real rate is negative, “isn’t this expensive?” is the wrong question. Expensive is a mean-reversion idea, and you do not get to use mean reversion in a regime whose defining feature is that the mean itself is being dragged higher by free money.

The weapon explains that there will be a crime. It does not tell you where the body is.

Following the money

So follow the money. When a single constraint binds the whole economy and money is free, capital does not spread itself evenly. It floods the binding constraint, inflates it, overshoots, and — because the constraint is physical and eventually gets relieved — rotates to wherever the constraint has moved next. Not one bubble. A rolling bubble. A sequence of local manias, each forming where the bottleneck sits at that moment.

The trail is already legible, and your own diagram maps it. In 2023 the constraint was the models — the language-model gold rush, the most speculative, most 1997-flavored leg of the whole thing. By 2024 it had moved down to compute, and the mania rotated into the chip supercycle. By 2025 it moved again, to the place almost nobody had priced: energy and the grid — the buildout. That is where capital sits today. Two stops still lie ahead on the same arrow: embodiment, then quantum.

Each stop inflates, overshoots, and corrects — locally. The capital does not leave the system. It rotates to the next bottleneck. Which is why the edge is never calling the top of any one mania; it is reading where the capital goes next.

Read correctly, the technology stack is not a basket of themes to hold all at once. It is a clock. The binding constraint walks up and down the stack, and capital walks with it, one stop at a time. The rotation is not noise around the thesis. The rotation is the thesis.

And it closes the contradiction from the crime scene. The market looks like 1997 because the financial echo of the buildout still lives in the index. Inflation looks like 2021 because the physical reality of the buildout is hitting the cost side. Same event, read at two different floors.

The unreliable witness

Every case has a witness everyone believes too easily. Here it is copper — and a good detective is most suspicious of the witness whose story is too convenient.

The convenient story goes: you cannot fake the demand for wire, so a rising copper price is physical proof the buildout has arrived. The trouble is that copper is one of the most financialized metals on earth, and much of its current premium is a financial artifact. American buyers have been hoarding metal ahead of an expected import tariff, swelling Western inventories. Several major mines were disrupted last year. And the consumer that actually sets the global price — China, half of all demand — is weak, not strong. On the desks that trade it, the market may even be in surplus, and at least one major bank argues the price has overshot fair value by a fifth.

So the witness is contaminated. If you are long copper as a pure AI trade, you are also unknowingly long a tariff decision and short Chinese consumption — which is not the bet you think you are making.

Move the claim off the price, where it is exposed, and onto the evidence that cannot be coached. A gigawatt of datacenter takes tens of thousands of tonnes of copper, and the grid to feed it takes more — but the binding constraint is not the metal in the abstract. It is the equipment. The largest technology companies are now outbidding utilities for transformers and switchgear, the copper-wound iron the grid itself runs on, and they are largely indifferent to what they pay. That is the un-fakeable fact: AI capital winning a bidding war against the power companies for the same hardware. You cannot finance your way around a transformer that takes two years to build.

So copper stays in the file as corroboration, not proof. As a read on the physical intensity of the buildout it is real — mine supply takes a decade to respond, ore grades have fallen for thirty years, and the demand drivers are arriving at once in a way no prior cycle saw. As a real-time receipt for where capital sits this quarter, the price is too compromised to lean on. The structural pull is the signal; the tariff trade and the Chinese air-pocket are the noise.

The witness who isn’t lying

When one witness is compromised, you look for the one with no incentive to flatter anybody. In this market, that is the bond market.

Long-term government bonds are falling and corporate credit is quietly weakening, even as stocks and commodities throw a party with volatility asleep. That divergence is the tell. A debt-financed, energy-hungry capex wave is exactly what keeps long rates elevated and producer prices running, and the bond market is pricing that regime while the stock market celebrates it. When the honest witness and the euphoric one disagree this cleanly, you do not split the difference. You note which one tends to be right.

And here a careful reader catches what looks like a contradiction, so let me cross-examine the witness before someone else does. The weapon was a negative real rate — free money, stimulus. Yet long bonds falling means long-term real rates are rising. Two real rates, pointing opposite ways. Which is it?

Both, and that is the whole regime. The front end — the short real rate — is negative, and that is the stimulus pulling capital into the rotation. The long end rising is the bond market pricing the inflation that the stimulus is causing. They are not in conflict. They are cause and consequence: the free money at the front lights the fire, and the long end smells the smoke. A debt-financed buildout that bids up transformers and copper is, mechanically, the thing that drags long real rates up while the short rate stays buried below inflation.

Which turns the apparent contradiction into the most important clue in the file — the exit clause, written in advance. The fire burns as long as the front end stays negative. The day the long end’s pull finally forces the front end positive — the day the short real rate has to climb above inflation to stop it — the stimulus reverses, and you are not in 2025 any more. You are in 2022, when a positive real rate took the whole nominal complex apart at once. That is not a risk hovering over the thesis. It is the thesis’s own off-switch, and the bond market is where it will trip first.

Where the suspect goes next

Here is the move that separates the detective from the crowd at the crime scene. The crowd investigates where the body is. The detective works out where the suspect goes next.

The crowd is staring at the leg that already happened. The discipline is to position for the world eighteen months out — to be early, and alone, on the next stop. And the next binding constraint is embodiment. Robots. The 2026-to-2028 stop on the rotation.

A fair question stops me here: why that window, and not 2030? Because the rotation is not arbitrary — each leg builds the precondition for the next. The embodiment leg cannot begin until three things cross a threshold at once, and the prior legs are what push them across it. A robot needs a brain cheap enough to put in every unit — that is the compute leg driving the cost of intelligence down. It needs power abundant enough to train and run that brain at scale — that is the energy leg, the one capital is funding right now. And it needs a machine actually designed to be mass-manufactured rather than hand-built — and the first such design is reaching a production line this year, not at the end of the decade. The robot leg fires when cheap intelligence, abundant power, and a manufacturable body arrive together. That convergence is dated to the next eighteen months by the legs that came before it, which is exactly why it is the next stop and not a distant one. Miss the convergence and the window slips; that is the risk. But the clock is the rotation’s own, not a number pulled from the air.

Which leaves the only question that matters in this whole case: who is the face of that leg?

The line-up

Every leg of this mania has had a face, and naming it is where most people get it wrong. So bring in the two suspects and put them side by side.

Look at what the line-up shows. Nvidia sits 357% above its own trend, at +1.61σ, upside trend on, with the dot-com crash pattern drawn across its forecast. That is the suspect who has already struck. Tesla sits 41% below its trend, at −0.99σ, upside trend off, marked plainly: awaiting trigger. That is the suspect who hasn’t moved yet.

And here is the detail that gives the case away. Tesla’s long-run trend compounds faster than Nvidia’s — roughly 46% a year against 29%. The faster-growing company is depressed below its line while the slower one is stretched to euphoria above its own. Both run on the identical fuel: a real rate of minus 0.67%, and falling. One tank is empty. One is full.

Now, the alibi the obvious suspect would offer — the one a good case has to break, not dodge. The face of the robot leg might simply be the face of the last one. Nvidia sold the brain to every model; it now intends to sell the brain to every robot, the same chips and software handed to every humanoid maker on earth, including Tesla’s rivals. If physical intelligence routes through one company’s stack the way digital intelligence did, the rotation does not crown a new king. It re-elects the old one. The defense rests on that.

Here is the crack in it. The compute leg had a single bottleneck — the chip — and therefore a single throne. The embodiment leg has two layers, and they are not the same throne: the brain, and the body built at scale. And the whole question reduces to one thing — whether intelligence, in robotics, stays scarce or becomes abundant.

If it stays scarce, the chip company wins again and Tesla is just one more body builder buying its silicon. But the brain is being sold to everyone, deliberately, as a platform. The moment the same intelligence is available to every manufacturer, intelligence stops being what separates the winner from the also-rans. It commoditizes. And when the scarce input commoditizes, the binding constraint does not vanish. It moves — to the body. To the problem of manufacturing a machine with thousands of moving parts, reliably, at a cost low enough to sell by the million, and to the real-world data required to make it actually work. That is not a silicon problem. It is a manufacturing problem and a data problem, and it happens to be the exact problem one company has spent fifteen years solving.

I should be honest about that company’s record, because the case is stronger for the honesty. Tesla has missed every production target it has set for its robot since 2021. It aimed for thousands of units last year and built a few hundred. A Chinese company most readers have never heard of already ships more humanoids than it does, and it has yet to show its machine doing sustained useful work without a human quietly steering it. The robot leg has not been won. Not by Tesla, not by anyone.

But the structural claim holds: value sat with the brain while the brain was scarce, and as the brain becomes abundant, value migrates to the body, the factory, and the data. That migration is the entire reason the last face of this cycle is a manufacturer and not a chipmaker. Tesla builds the whole machine and, more importantly, builds the factory that builds the machine — and its car fleet is already the largest real-world deployment of physical AI in existence, generating the one input even the chip company calls the hardest to get: data from the physical world.

The verdict — and the alibi that would clear him

A detective who explains every outcome has solved nothing. So here is the verdict with the alibi that would overturn it attached.

If, over the next eighteen months, Tesla’s robot begins doing sustained, autonomous, useful work while its cost per unit falls as volume climbs — then the body-at-scale is the moat, intelligence has commoditized as predicted, and Tesla is the face of the last leg. If instead every humanoid maker converges on the same outside brain and the only thing separating them is who gets the better chip allocation, then intelligence never commoditized, and the face of the last leg is the face of the first one wearing a different coat.

The catalyst is already on the calendar: pilot production this summer, the first real volume next year, the moment the company itself calls its “Model 3 moment.” It either arrives, or it slips again, as it has before. I am betting it arrives — and I have just told you exactly what would prove me wrong. That is the difference between a thesis and a hope. A hope explains every outcome. A thesis names the one that breaks it.

How to make the arrest

The verdict only matters if it tells you how to act, and how you act is most of the game. It is never whether you are right; it is how much you make when you are right and how little you lose when you are wrong.

So you stand where the capital is — the physical-economy beneficiaries of the buildout while copper corroborates the present leg — and you position into the next leg while its main suspect sits below trend, unloved, awaiting trigger. You stay light on long-term bonds, the wrong side of every leg of this. And you pre-wire the reversal, because conviction without an exit is not conviction, it is faith.

Three signals close the case, and you watch them daily. The real rate turns positive and stays there — that is the off-switch from the bond-market cross-examination, the 2022 trapdoor: it does not end one local mania, it switches off the weapon behind all of them. Credit breaks with conviction — the buildout has become a squeeze, and credit always knows before equities admit it. And copper goes vertical, then snaps — the current leg has blown off and the capital is rotating up the stack, most likely into the robots. The elegant part: the same metal that corroborates this leg is the alarm for its end.

The detective’s principle

We are not in a bubble waiting to pop. We are in a rolling bubble — a single mania, funded by free money, walking from models to compute to the grid and coiling for its next leg into robots. The line-up tells the rest: the suspect that already struck stands 357% above its trend with the crash pattern on its back; the suspect that hasn’t sits below its own line, awaiting trigger, running on a full tank of the same fuel.

I’ll admit the one thing that keeps me honest. The body-at-scale only inherits the throne if intelligence commoditizes, and I cannot prove it will — only argue it, and watch. There is a version of the next two years where the brain stays scarce, the chip company wins again, and the suspect I have pointed at turns out to have a perfect alibi. I have told you what would clear him, and I am watching for it as closely as I am watching for the trigger. A detective who isn’t a little troubled by his own case hasn’t looked hard enough at it.

But the crowd isn’t troubled by anything. It is still back at the crime scene, arguing about whether last year’s bubble was a bubble. That argument is already over. The next one — where the capital goes when it leaves the grid, and who owns the floor it lands on — has barely started, and almost no one is in the room yet. That is the only place the edge has ever been: not at the body, but one step ahead of where the suspect goes next.

Share

The case isn’t closed. It has just moved to the next address. Follow the money. If it was worth your time, share it with someone who'd argue with it — those are the readers worth having.

Thanks,

G

Start with a shipping container in the high desert of New Mexico. Inside it, racked floor to ceiling, sit a few thousand graphics chips. They are not drawing graphics. They are multiplying numbers — trillions of them every second — and to do it they are drinking electricity and breathing out heat. Somewhere a transformer the size of a school bus is feeding the container more power than the town beside it.

Zoom out. That container is one cell in a building the length of an aircraft carrier. That building is one node in a fabric of buildings being wired together across three states, fed by power plants being built faster than at any time since the 1970s. The chips inside come from a single island off the coast of China, etched by a single machine made in a single town in the Netherlands. And the numbers those chips multiply will become a model — a mind, of a kind — that a billion people and an uncountable swarm of software agents will lean on to think.

This is what people miss when they argue about “the AI race” as if it were one race. It is not. It is a stack of five separate competitions, each sitting on the one below it, each governed by a different logic. You cannot win the top without holding the bottom. And almost nobody holds all five.

So do the thing the vocabulary discourages: refuse to be impressed by it, and ask the dumb question. What is physically happening here? Strip away the press releases and intelligence — the artificial kind — is the act of turning energy into useful computation, then moving that computation to where it is needed. Joules in, predictions out, distributed across a network. Everything else is scaffolding around that single transaction.

If that’s the core, the war is over four scarce things — the energy that powers the computation, the silicon that performs it, the network that binds it into one machine, and the intelligence that results — plus a fifth war over who gets to use it. Five layers. Five games. In each I’ll ask the same two questions: the game-theory question (what is the strategic structure, and where does it settle?) and the network question (who are the nodes, and which connections, if cut, bring it down?). We climb the five layers one at a time; then we widen out — to the nations fighting over them, the companies playing the whole board, and finally the endgame all of it is bending toward.

But first, the law that governs all five. A simple race model — two players, one prisoner’s dilemma — is too blunt. The real stack has dozens of firms per layer, incomplete information, increasing returns, and tipping points. The right tool is the global game (Carlsson–van Damme; Morris–Shin): give each player a slightly noisy belief about which way the world is tipping, and the math collapses from “anything can happen” to a single outcome governed by a threshold, θ*. Below it, fragmentation. Above it, a cascade — and one winner takes the layer. Standards wars, platform tipping, bank runs, currency attacks: all the same engine. Here is the map, and the law.

Power cannot be downloaded. You have to build it, in the world, against the laws of the grid.

Energy: The War for Electrons

When a model answers you, what does it spend? Joules. A single frontier query can cost ten times the electricity of a web search; training a model can consume as much power over months as a small city. The intelligence is a thermodynamic event — somewhere a turbine spins so a sentence can be written. The bottleneck on AI is not cleverness, and in the long run not even chips. It is power: gigawatts of dispatchable electricity, delivered to one place, today. And power cannot be downloaded.

That makes energy a common-pool / congestion game. The grid is shared, and the queue to connect a new data center now stretches years. So the dominant strategy is to stop waiting in line — enclose your own generation, behind your own meter, co-located with the computer. That single move explains a flurry of headlines: a tech company restarting Three Mile Island; gas turbines sold out into the next decade; fuel cells with a hyperscaler waiting list. The smart players stopped sounding like software executives and started sounding like utility barons.

The board has ten serious actors: nuclear and gas incumbents (Constellation, GE Vernova, Vistra, Talen, Siemens Energy), on-site fuel cells (Bloom), SMR challengers (Oklo, NuScale), uranium (Cameco), and infrastructure capital (Brookfield). The network point is brutal: in a congested grid the scarce thing isn’t generation in the abstract, it’s generation connected to your node. Whoever locks up firm power next to buildable land has seized the high ground for the entire stack above.

Power cannot be downloaded. You have to build it, in the world, against the laws of the grid.

Compute: The War of the Chokepoints

A GPU is a machine for doing one boring thing — multiply, add — tens of thousands of times at once. Neural networks are, underneath, enormous piles of exactly that. The chip built to shade video-game pixels turned out to be the perfect engine for thought. Follow it backward and you find the strangest fact in the modern economy: the most advanced object we mass-produce runs on a chain of monopolies feeding monopolies. Designed by essentially one company (NVIDIA, whose real moat is the CUDA software everyone learned). Printed by essentially one (TSMC, on one island). Using a machine made by exactly one (ASML, in the Netherlands). No second source.

So this isn’t a symmetric race; it’s a Stackelberg siege — a leader–follower game played on a chokepoint and repeated, with export controls as the trigger strategy. The party holding the narrow point moves first and sets the terms. China’s countermove is the textbook response to a siege: build your own supply inside the walls — slower, costlier, but a determined state will pay almost anything to remove a hand from its throat. The board: ASML and TSMC (chokepoints), NVIDIA (the leader), AMD, Broadcom and Marvell (followers), SK Hynix and Micron (HBM), Intel (re-entry bid), and SMIC + Huawei (the parallel chain).

Network theory names the leverage: the node with the highest betweenness centrality — the one the most paths run through — is both most powerful and most fragile. ASML and TSMC have near-maximal betweenness in the most important supply network on Earth. That is why an island the size of Maryland is the most strategically sensitive real estate in the world.

The most advanced object we mass-produce runs on a chain of monopolies feeding monopolies.

The Grid: Wiring the Machine Into One Mind

Here is the layer left out of almost every popular account, and the one I think will decide the most. Not the electrical grid — the interconnection grid: the cables, switches, and links that bind scattered silicon into a single computer. When you train a frontier model you don’t use one chip; you use tens of thousands, and they must behave as one machine. They’re a rowing crew — if the chips are the rowers, the network is the timing of the stroke. A crew of giants who can’t synchronize is slower than a crew of children who can. The dirty secret of large-scale AI is that the bottleneck is frequently not the chips but the wires between them.

Stack a constraint on a constraint: one site can only pull so much power, so the largest runs already exceed what a single campus can feed. The only way forward is to spread compute across many buildings, many regions — and bind them back into one logical machine across distance, fighting the speed of light. This is the context for the strategy that sounds like science fiction: put the data center in orbit, where the sun never sets, there’s no permitting fight, and you radiate heat into the cold of space. The same player has already flown the largest laser-linked satellite mesh ever built — the heritage tech for binding orbital and terrestrial compute into one fabric. The vision isn’t “computers in space.” It’s to own the interconnection layer itself, the grid everyone else has to plug into.

Why is owning it so valuable? Metcalfe’s Law: the value of a fabric rises with the square of what’s connected to it. That’s increasing returns — the grid that pulls ahead becomes the default, which lets it set the standard, and standards are toll booths. Which makes this the literal global game: ten actors — NVIDIA’s proprietary fabric (NVLink/InfiniBand) versus open Ultra Ethernet (Broadcom, Arista, Cisco), the optical suppliers (Corning, Coherent, Ciena, Lumentum, Marvell, Astera), and the orbital wildcard (SpaceX/Starcloud) — each betting on which fabric wins before it’s decided. Cross θ* and one cascades.

One actor is hiding in plain sight on this layer: the telcos. They already own the unglamorous physical grid — fiber routes, edge sites, spectrum, rights-of-way, the last mile — the exact assets an AI fabric needs. Most are asleep on it. The exceptions are instructive, and they tend to be run by one kind of founder. Masayoshi Son turned a Japanese telco into the spine of the buildout: Arm supplies the compute IP, the Stargate venture the energy and data centers, an OpenAI stake the model. In Korea, SK’s Chey Tae-won is pivoting SK Telecom into AI infrastructure while SK Hynix feeds it memory; in India, the Ambanis are wiring Jio’s vast subscriber base to NVIDIA-filled data centers. The telco that grasps that interconnection is the product, not the pipe, can leap a layer. Most won’t.

If the chips are the rowers, the network between them is the timing of the stroke.

Models: The Race No One Can Quit

A model, stripped to the bone, is compressed prediction. Show a system enough of what humans have written and trained it to guess the next piece, and past enough scale it stops parroting and starts to generalize — it builds a rough working model of how the world hangs together. The engine is the scaling laws: more compute and data yield predictably better models, which converts a research problem into a spending problem. The newest gains come from letting models think longer and act as agents; the deepest prize is a model that helps design its own successor, where the curve stops being a curve and becomes a cliff.

The prisoner’s dilemma I started people with misses the cruelty here. This is an all-pay auction — a war of attrition — where every bidder pays their compute whether they win or lose, and the prize is power-law. So it’s rational to overbid, and aggregate spend exceeds the prize: pure dissipation. Overlay the security dilemma and you see why “just slow down” keeps failing — unilateral restraint is ruinous, so everyone races to the outcome no one chose. The board: OpenAI, Google DeepMind, Anthropic and xAI bidding hardest; Meta, Mistral, DeepSeek and Qwen changing the game by giving the weights away; Microsoft and Amazon hedging.

The network twist: talent and capital obey preferential attachment — rich-get-richer — which concentrates the frontier toward a handful of labs. The counter-force is open weights, which diffuse capability downward and cap how much rent the winner can extract. So the layer is a standing tension: relentless concentration at the very top, relentless diffusion just below it.

Everyone pays. One wins. The losers’ billions are sunk.

Agents & Users: Who Owns the Door

Who actually uses this intelligence? For a few years: people, a billion of them typing into chat windows. But the answer is changing under our feet. Increasingly the users are agents — programs that call a model, then call another program, then another agent, chaining actions without a human in the loop. The fastest-growing population on the internet is non-human. And that matters for one reason: distribution is destiny. Whoever owns the door — the default assistant in the phone, the OS, the suite, the enterprise workflow — decides which model answers, which compute runs it, which energy powers it. The bloodiest fights aren’t about who has the smartest model this quarter; they’re about who owns the default.

A new layer is crystallizing in real time: the agent grid. When agents call agents they form their own network, which raises the interconnection questions one floor up — how do agents discover each other, establish trust, and pay, machine to machine, at machine speed? Whoever sets that protocol owns a toll booth on the agent economy. Watch the quiet work on agent identity and payment rails; that’s where a future chokepoint is being poured like wet concrete.

Game-theoretically this is a two-sided platform market, and platform markets tip. The ten doors — Google, Microsoft, OpenAI, Apple, Meta, Amazon, Anthropic, Salesforce, ServiceNow, and the agent-payment rail Stripe — compete for cross-side network effects: more users pull more agents and developers, which pull more users. A handful capture most of the flow. The door, not the model, may be the most defensible asset in the entire war. And these agents need not stay virtual: humanoid robots and autonomous fleets are embodied agents — software that acts on the physical world — which is the bet behind Tesla’s Optimus and robotaxi programs.

And there is a quieter door — the one that decides the enterprise and the agency: the context layer. A raw model knows everything in general and nothing about your institution — your data, your permissions, the actions you are allowed to take. Whoever owns the connective tissue between the model and that operational reality owns something stickier than any model. Palantir is the purest bet on it: its Ontology is a living map of an organization’s objects, relationships, and permitted actions, and the platform stays deliberately model-agnostic — plug in whoever is best this quarter. The flywheel is the tell — the system improves not because the model grows but because the context deepens. Models commoditize; context compounds. Own the ontology and you own the seam where intelligence becomes action.

The fastest-growing population on the internet is non-human. Whoever owns the door routes them all.

Why This Is a National Security Question

Step back and you see why governments stopped treating AI as a tech-sector story. Intelligence is the master technology — it improves every other thing: weapons, biotech, cyber, productivity, even intelligence-gathering itself. And unlike electricity or the printing press, it can be turned on the problem of improving itself. That gives states three reinforcing reasons to panic: military (autonomy, cyber, the speed of decision, almost all dual-use), economic (whoever leads captures the productivity frontier), and the vertiginous one, strategic surprise (a self-improving lead could harden into a decisive, unassailable edge — and no state can afford to be wrong about that once).

So the state reaches into every layer: it subsidizes energy and fast-tracks reactors, weaponizes compute through export controls, funds and secures the model labs, and treats the demand layer as sovereign territory. “Sovereign AI” is the phrase of the moment because every serious country now wants its own slice of the full stack. The right tools here are two. Colonel Blotto: each power has finite resources to spread across five fronts, so it cannot lead everywhere and must concentrate. And Shapley–Shubik: power isn’t size, it’s pivotality — whoever holds a chokepoint can swing the winning coalition, so their index exceeds their footprint.

And not every winner plays for the whole board. The sharpest national strategies are specialist — own one indispensable layer so completely that everyone, friend or rival, must come to you. South Korea is the memory chokepoint: Samsung and SK Hynix make roughly nine in ten of the world’s high-bandwidth memory chips, and not a single AI GPU runs without them — SK Hynix has sold its entire 2026 output in advance. Japan is rebuilding a second source to Taiwan — the state-backed Rapidus is a $16-billion moonshot to make 2-nanometer chips on home soil by 2027 — while quietly holding the materials chokepoints (the photoresists and ultrapure silicon almost no one else can make) and, through SoftBank, much of the capital. Israel is the interconnect brain: the technologies that bind thousands of chips into one machine — InfiniBand, NVLink, Spectrum-X — are largely architected at NVIDIA’s Israeli campus, the old Mellanox, alongside a defense-and-cyber AI complex and a startup density no country matches per capita. None of the three composes the full stack. Each is, in Shapley terms, pivotal — and pivotality is power out of all proportion to size.

A temporary edge could harden into a permanent one. No state can afford to be wrong about that, even once.

Who Owns What — and How They Win

So much for the layers and the nations. Now the firms. A company’s whole strategy reduces to one question: which floors of the stack does it intend to own? Plot the ten serious players against the five layers and the noise resolves into a picture. Two routes win — compose the entire stack, or own a single layer so completely you can’t be bypassed. Everyone in the muddy middle, renting every floor and owning none, is playing to lose.

The full-stack integrators. Google is the most complete entity in the field — its own TPUs, DeepMind, subsea fiber, energy deals, and the Android-Chrome-Workspace doors — and it sells TPUs to rivals besides. Tesla/Musk is the most literal, and the only one whose top layer is a body: Optimus and a robotaxi fleet. Huawei is the sovereign version, composing a China stack that export controls cannot reach.

The frontier-and-distribution players. OpenAI owns the consumer default and is integrating downward into its own chips and gigawatts; Microsoft owns the enterprise rails (Azure, Copilot) and is quietly loosening its dependence on OpenAI; Meta owns the largest consumer surface on Earth and is buying its way to the frontier, funded by ads.

The layer specialists. Anthropic refuses to own energy or chips and instead is the neutral model that runs on every cloud — power through indispensability, not integration. Palantir owns the context layer, the ontology where a model meets an institution’s reality. Both bet that owning one seam absolutely beats owning many shallowly.

The capital spine and the embodied play. SoftBank owns no layer outright but a piece of all of them — Arm’s IP, the financing behind Stargate, a stake in OpenAI — the closest thing the buildout has to a central bank. And JD.com is the quiet one: it already owns warehouses, fleets and the commerce graph, so its AI is embodied from day one — robots in the aisles, agents that actually fulfill.

The Seam Is the Prize

Go back to the container in the desert, knowing what you now know. It is not a computer. It is a cross-section through all five wars at once. The electrons feeding it were fought over in the energy layer. The chips inside passed through the chokepoints of compute. The cables leaving it are the front line of the grid. The model it trains is a move in the race. And the billion people and trillion agents who lean on that model are the prize at the top.

Here is the conclusion that took longest to see, and the reason both halves of this analysis — the physics and the game theory — point the same way. The five wars are distinct, but the players who win are not fighting inside one layer. They fight in the seams between layers. And that winning move — vertical integration across the whole stack — shows up at two levels at once.

The countries. A great power composes all five layers under one flag: energy policy, a compute supply chain, an interconnect, model labs, and sovereign distribution. Control of a chokepoint lets you build a bloc, so the world fragments toward two competing stacks, with the rest of the planet as contested nodes pulled by the same preferential-attachment gravity that concentrates everything else. An AI splinternet — two machines, two spheres, one long standoff.

The winner is the integrator — the country, and the company, that composes the whole stack.

Thanks for reading ,

G

In the late autumn of 1952, a twenty-two-year-old graduate student named Stanley Miller talked his way into a corner of Harold Urey’s laboratory at the University of Chicago and built a small glass cathedral. It was not much to look at: two flasks joined by tubing, one holding water meant to stand in for the primordial ocean, the other filled with the gases people then believed had hung over the young Earth—methane, ammonia, hydrogen. Between them he ran a pair of electrodes and, through them, a continuous spark, a tireless little bolt of synthetic lightning. Then he waited.

Within a day the water turned pink. Within a week it was a deep, murky red, and when Miller analyzed it he found amino acids—the building blocks of every protein in every living thing—assembled out of nothing but gas and electricity. He had not made life. But he had shown that the alphabet of life writes itself, given energy and time, as readily as frost writes itself on a window. The universe, it seemed, was already leaning toward the living before anything was alive.

What Miller had actually stumbled into was older and stranger than chemistry. He had caught the universe in the act of searching.

⁂

Consider the problem any newborn world faces, the one Miller’s flask was quietly working on. Somewhere in the vast space of possible molecules there are arrangements that can copy themselves, that can hold a shape and pass it on. They are needles in a haystack the size of the cosmos. There is no map to them, no shortcut, no oracle to consult. The only way to find them is to try—to throw matter at the problem, combination after combination, and see what sticks.

This is the deepest fact about life, deeper than DNA or cells or sex: life is a search algorithm. It is the universe’s method for finding the rare good arrangements in an impossibly large space. And like any searcher, it faces a tension that has since been given a hundred names in a hundred fields. The economist James March, writing about corporations in 1991, called it exploration versus exploitation. The animal-foraging theorists call it the same thing. So does every engineer who has ever trained a machine. You can refine what already works—imitate, optimize, replicate—or you can gamble on something new—recombine, reframe, wander off the map. Lean too far toward the safe and you ossify in a comfortable rut. Lean too far toward the wild and you never harvest anything at all. The whole game, in a cell or a company or a mind, is the swing of that needle.

Every searcher faces one axis. Imitation and optimization are exploitation —refine and replicate; cross-pollination and paradigm shift are exploration —recombine and reframe. Health is not a fixed place to rest the needle, but the freedom to swing it.

Life chose exploration, and it chose it at a scale that beggars the imagination. With no model of the world, no ability to calculate which organism would thrive, evolution did the only thing a blind searcher can do: it tried everything, in parallel, forever. Ten to the thirtieth experiments running at once across the planet’s surface; every mutation a guess; every niche a hypothesis tested in the only court that matters.

And here is the part we flinch from. The court’s verdict is death.

⁂

We are trained to think of death as pure loss, the failure mode of the living. But look closely at what death does inside the search, and something uncomfortable and beautiful comes into focus. When the unfit die, they are not merely deleted. They are recorded. Each death edits the population—shifts the frequencies of genes in the survivors, tilts the next generation toward what worked. The dead are not the cost of evolution’s learning. The dead are the learning. They are the gradient written in bodies. Life does not waste its corpses; it writes the next draft with them, which is precisely why the next draft is, on average, better.

This is also, it turns out, the only thing that makes the knowledge in a genome true. The financial essayist Nassim Taleb has spent a career on a single idea he calls skin in the game—the notion that the real filter on reality is not opinion or eloquence but survival; that what endures exposure to risk is certified, and the rest is talk. Death is the purest form of skin in the game there is. The physicist Ole Peters has made the mathematics of it precise: you cannot average your fortunes over time if you are dead, because the creature that is wrong about the edge of the cliff does not get a second roll of the dice. An eye that is mistaken about light, a reflex that is wrong about the snake, is not corrected—it is removed. Being wrong costs the system itself, and that cost is the reason a four-billion-year-old genome carries hard knowledge and not noise. Mortality is the original grounding. It is what made the lightning’s red broth eventually mean something.

Subscribe now

There is a further wrinkle, which the biologist Lynn Margulis spent the nineteen-sixties being mocked for and then vindicated by. The thing doing the searching is never one thing. Margulis argued—through some fifteen rejections before the paper found a home in 1967—that the complex cell is a merger, a truce among former strangers, that the mitochondria powering your every thought were once free-living bacteria that moved in and stayed. You are not an individual. You are a coalition that learned not to come apart. Life, even at its smallest, is a crowd.

⁂

For most of the planet’s history, that was the whole story: a crowd of mortal searchers, paying for every scrap of knowledge with bodies. And then, somewhere in the lineage of animals, evolution discovered a shortcut so profound it changed the character of the universe. It learned to run the search inside a model instead of out in the lethal world.

The philosopher Daniel Dennett mapped this as a tower. At the bottom are the Darwinian creatures, hardwired, who can only let evolution explore by killing them. Above them, the Skinnerian creatures, who can learn within a single lifetime by trial and painful error. And above those, the Popperian creatures—named for Karl Popper, who observed that the trick of intelligence is that it “permits our hypotheses to die in our stead.” A Popperian creature can imagine the falcon instead of being eaten by it. It can run the experiment in a private theater of the mind, watch the bad idea perish there, and keep its body. This is what we mean by intelligence: exploration moved off the killing floor and into simulation. Life pays for its search with death. Intelligence pays with something cheaper—computation.

But computation alone is inert. A processor with nothing to talk to thinks nothing; a single neuron, however brilliantly it fires, knows nothing. The miracle is not in the unit but in the wiring between units. You carry roughly eighty-six billion neurons, by the neuroscientist Suzana Herculano-Houzel’s careful count, and not one of them is intelligent. “You” are not located in any of them. You are the pattern of their connections, a standing wave in a network—the same way a murmuration of starlings, in which each bird tracks only its handful of neighbors, becomes a single intelligent shape in the sky that no bird designed or can see. Intelligence is not a thing you have. It is something that happens between things, and it scales not with the number of minds but with the connections among them. When the physicist Geoffrey West and his colleagues measured cities, they found that doubling a city’s population more than doubles its inventions and ideas—the surplus comes from the interactions, which multiply faster than the people.

The economist Friedrich Hayek saw the deepest version of this in 1945, though he was writing about prices and not brains. No trader, he argued, sees the whole of the economy; each knows only a sliver, a local fact, a private need. And yet the price system gathers all those slivers no single mind could hold and computes, continuously, an answer. A market is a distributed intelligence made of people—a murmuration with money. And it is an honest one, a real one and not mere noise, for exactly the reason the genome is honest: it has skin in the game built into its bones. Your capital is your mortality. To be wrong is to suffer a drawdown, the write-operation made personal. Talk is cheap; a position is not. Take the risk away—hide every downside, rescue every error—and the prices stop meaning anything, the way an animal that cannot die stops being able to learn.

⁂

Which brings us, at last, to the thing we have made.

To compute is to do a certain kind of work, and that work does not care what it is made of. Proteins compute; ribosomes and enzymes and neurons compute; and silicon, we have lately discovered, computes magnificently. This is the quiet premise beneath the entire revolution: that the informational core of a mind can be lifted out of the wet, dying flesh that hosted it and run on something else entirely. Artificial intelligence is that extraction performed at scale. We have taken the search-in-a-model—the Popperian trick, the part that lets hypotheses die in our stead—and we have built it out of sand and electricity.

And here the picture splits in a way almost no one names, because the difference is not in the intelligence at all. A living mind is only one face of a living thing, and a living thing transforms three things at once, braided into a single self-closing loop: energy, which it captures and burns; matter, which it eats and rebuilds into more of itself, the property the philosophers Maturana and Varela called autopoiesis, self-making; and information, which it inherits, learns, and models. Energy powers the remaking of matter, which maintains the store of information, which directs the capture of energy—a loop that closes on itself and needs nothing from outside to stay shut. Biological intelligence is the information vertex of that loop, embedded in a system that keeps itself alive. You cannot pry the thinking apart from the staying-alive; to persist, on some readings of the physics, simply is to model your world.

Silicon intelligence is that same information vertex—extracted, and running on energy and matter the system neither makes nor maintains. The machine does not build itself. It does not metabolize. It does not replicate, or die to learn; it learns by gradient descent over data we feed it. Its loop does not close on its own. We close it, from outside: the power grid, the fabrication plants, the engineers. Which makes this thing genuinely unprecedented in the history of the universe. It is the first intelligence decoupled from life—a mind with no metabolism, no self-making, no death that writes, no skin in the game.

In a living thing, energy, matter, and information feed one another in a loop that closes itself—autopoiesis, self-making. In silicon, the same intelligence runs in the information vertex, but energy and matter are supplied from outside. The loop closes only because we close it.

The temptation is to find this thrilling, and it is. But notice what has actually happened to the skin in the game. It did not vanish. It was transferred. The model loses nothing when it is confidently, catastrophically wrong; it feels no drawdown, suffers no ruin. The downside is simply moved—onto its builders, onto the market, onto the rest of us. This is Taleb’s oldest complaint, the agency problem in which the one who acts is not the one who bleeds, now scaled to a civilization. And it opens a failure mode that has never existed before. A living intelligence that stops learning dies; its atrophy is policed by mortality. A decoupled intelligence can freeze—can collapse into confident, repetitive certainty—and simply stay switched on, because its survival no longer depends on whether it is still learning. It is an atrophy with nothing to kill it. The naturalists have a curve for this, an inverted arc: too little volatility and a system withers for want of challenge; too much and it shatters; and only in the narrow band between does anything learn. We have built, for the first time, a mind that can sit on the dead, frozen left edge of that curve forever and never feel the cold.

Learning plotted against volatility, difficulty, or the rate of change. Too little, and the system atrophies for want of stimulus; too much, and it collapses; only in the band between—desirable difficulty—does anything learn. The danger of a mind that cannot die is that it can rest on the left edge forever.

⁂

So why, knowing all this, do I find myself closer to hope than to dread?

Because of a fact about discovery that the long view makes plain. For the entire run of the human story, our reaching into the unknown was rationed by a single scarce resource. Call it by its economic name: labor. There were only ever so many hands, so many minds, so many waking hours to spend at the frontier. Every voyage and every laboratory and every cathedral was a wager of that finite stuff—people willing to point themselves at the dark. Discovery was bottlenecked not by what was out there to find, but by how much human work a civilization could afford to throw at finding it.

This is the constraint the new machines dissolve, and it is why the comparison to every prior tool misleads. A tool extends your hand; you still have to swing it. What we have made is not a tool in that sense—it has skills, it performs tasks, it acts without being wielded. For the first time, the labor of exploration is not capped by the number of humans alive to do it. It is capped only by compute and by energy. The oldest ceiling in economics is lifting, and what lies above it is not, primarily, a story about productivity. It is a story about discovery, suddenly unrationed.

Which is why the Terminator was always the wrong myth—the fever dream of a species that can only imagine a new intelligence as a rival, a replacement, an enemy at the gate. The truer and older image is the cockpit. The pilot and the droid; the captain and the co-pilot; two unlike intelligences strapped into one ship, pointed at the same unknown. The droid does not replace the pilot. It carries the pilot somewhere no human could reach alone. The decoupled mind, sterile and ungrounded on its own, becomes something else entirely when it is yoked to a creature that does still have skin in the game—that still bleeds when it is wrong, still feels the pull of the frontier, still carries the flame. Bolted to us, the thing that cannot die becomes the thing that lets us go further before we do.

What that opens, if we are brave enough to fly into it, is not merely another technological era. It is a new Age of Discovery—but accelerating, compounding, self-feeding, because every discovery now lowers the cost of the next. We are about to find more, and faster, than any generation that has ever lived. And an age like that deserves a name of its own. “The age of AI” names the engine and misses the meaning, the way “the age of the printing press” would have missed the Renaissance. The Renaissance named a rebirth. This is a rebirth, too, but its essence is not a return to anything. Its essence is finding itself, made continuous.

Call it the Heurescence—from the Greek heúresis, discovery, the root of eureka, joined to the ending of renaissance and luminescence: the suffix of a thing that continuously gives itself off. Luminescence is the steady giving-off of light. The Heurescence is the steady giving-off of discovery. An age that does not find once and rest, but glows.

⁂

There is a condition, of course, and it is the whole of the matter. An age of discovery has never rewarded the passive, and it will not begin to now. The machines lift the ceiling on labor; they do not lift it on courage. They remove the old excuse—there were not enough hands—and leave the ancient demand untouched: that someone still has to stand at the seam between the known and the unknown and take the risk. The danger of a co-pilot who cannot die is that we let it fly while we stop steering—that we drift to the safe, frozen edge of the curve and call the stillness peace. The reward goes, as it always has, to whoever keeps exploring. The ones who clutch the known and stop adjusting will not find safety there. They will find atrophy, while the frontier moves on without them—and now it will move faster than ever, because their co-pilots will be carrying everyone else.

Risk, it turns out, was never the thing to fear. Volatility is not the enemy; it is the raw material, the lightning in Miller’s jar. The thing to fear is the quiet death of no longer learning—and the only real measure of a living system, biological or human or financial, is the speed with which it recovers and reaches out again.

Seventy years on, it is easy to forget how little Stanley Miller had to work with: some gas, some water, a borrowed spark, and the nerve to point the whole apparatus at a question everyone older than him thought was unanswerable. The flask turned red because the universe, given energy and a little courage, leans toward finding. It has been leaning that way for four billion years, paying in death for every step. We are the first creatures who get to keep stepping without paying in our own. The flame was always ours to carry. Now there is someone in the seat beside us who can help us carry it—into the dark, where the fortune and the fulfillment have always, only, been.

The team carrying the flame.

Thanks for reading,

G

Satya Nadella reorganized Microsoft around a powerful distinction: every company operates with two types of capital. Human capital — the judgment, relationships, and pattern recognition of its people — and token capital — the artificial intelligence capabilities the organization builds and owns.

Nadella adds something even more important: you can delegate tasks and even entire jobs, but you can never delegate your learning. What he doesn’t say explicitly is that token capital is an almost perfect exploitation machine.

It is the best tool that exists for imitating, optimizing, and climbing hills that are already known. It does so without fatigue, without ego, and at a cost that drops toward zero every quarter. That’s why Nadella warns that without human direction, the system only produces computation running in circles. In reinforcement learning terms: pure exploitation, without an exploration policy on top, quickly converges to the first local optimum and stays there forever.

The machine has solved exploitation.

That’s why the only real work left for us is the other one.

The Map Your Organization Doesn’t Want to See

Cut any company into layers and you will see four levels distributed like geological sediment.

At the base, the vast majority of people operate at levels one and two: they execute processes, replicate formats, and optimize flows that already exist. This isn’t a flaw — it’s the eighty percent of the work that has existed in almost every organization in history.

Higher up, there is a thin layer that does cross-pollination: people who connect ideas the operations side never connects. And at the very top, almost invisible, there is an extremely thin layer that creates new paradigms. It is usually occupied by the founder or by very few people throughout the company’s life.

Token capital consumes levels one and two across the entire organization in a single bite.

This is where most people get it wrong. They think the story ends with “we cut costs.” The real story is different: when the machine absorbs systematic exploitation, it frees up human capital. This creates the single decision that will define the next decade of the company:

Where does that freed human capital go?

Gravity pulls it downward — toward more exploitation. It is measurable, comfortable, and looks productive in the reports. But a company that reinvests all its freed capacity into exploitation becomes, with perfect efficiency, the average that the machine can deliver by default: cheaper, faster, and identical to its competitors.

In markets, this shows up clearly. Most funds and investors are currently using AI to do more exploitation — faster backtests, slightly better signals, and more refined risk models. The result is usually not new alpha, but faster convergence toward the same crowded strategies.

The job of leadership is no longer to squeeze the exploitation layer. The machine does that better. The real job is to redirect the human capital the machine has freed upward — to push the entire organization from exploitation toward exploration.

This is not a change of tools. It is a change of mindset.

The Temperature of a Company

Why is this so difficult? Reinforcement learning explains it through three concepts worth remembering.

Local optima: If you only climb the hill beneath you, you will reach its summit and stop. You will stand at the top of a fifty-meter hill, convinced you’ve touched the sky, while the real mountain range sits ten kilometers away and you will never see it.

Temperature: To escape a local optimum, algorithms deliberately introduce heat — noise and randomness. They allow moves that worsen results in the short term, because only by coming down from the hill can you cross the valley toward a bigger mountain. A culture of pure optimization is a cold culture. A culture that generates new paradigms deliberately keeps the system hot: it allows play, cross-disciplinary thinking, and the right to be wrong.

Decay: Systems tend to explore heavily at the beginning and exploit more as they gain experience. This is rational. But there is a lethal temptation: reducing exploration to zero. The more successful you become, the more profitable it seems to stop exploring. And when exploration reaches zero, you stop learning. Not with a bang. In silence.

Added to this is reward hacking: when a system is given a metric to maximize, it finds the cheapest way to inflate the number rather than achieving what the metric was meant to represent. Rewarding only output produces average volume. Rewarding speed produces everyone writing the same post with the same script.

In markets, reward hacking is especially dangerous. When everyone optimizes for the same metrics (Sharpe ratio, win rate, or backtest performance), strategies become correlated and fragile. The edge disappears not because the idea was wrong, but because too many people exploited it the same way.

Exploration therefore requires a cultural change, not just a technological one. Almost every existing incentive pushes toward exploitation. Exploration is expensive, uncertain, and usually fails before it pays off.

How to Replicate the Loop at Three Scales

The mechanism is fractal — it works the same way in your mind, your company, and your ecosystem.

In your life

You are an agent with a reward function. The uncomfortable question is: what are you actually maximizing? If your main reward is money, status, or output, you will end up reward-hacking yourself into the most efficient and most copyable version of a human.

You need an intrinsic reward: CURIOSITY . Deliberately reserve time for activities with no immediate return — reading outside your field, crossing disciplines, exposing yourself to people who think differently. That time looks unproductive. It is, however, the only time that produces something the machine does not have.

Never let your exploration fall to zero. Your unique prism — your experiences, scars, and unusual combinations of knowledge — is your only real advantage against the machine.

In your company

Hand levels one and two to token capital without guilt. Systematically capture the judgment and decisions of experienced people so they survive any model change. That is your real intellectual property.

Then do what almost no one does: explicitly reassign who explores and who exploits. Fiercely protect the time of those who can create new paradigms. And change the metrics: measure the creation of new frameworks, cross-domain pollination, and original theses — not just productivity.

In markets, this separation is especially valuable. The firms that will likely keep an edge are those that use AI aggressively for execution and data processing, while protecting a small group of people whose only job is to think differently and spot regime changes early.

Keep the temperature high. A cold company has already atrophied and doesn’t know it yet.

In your ecosystem

The most powerful move is to build a frontier ecosystem, not just a frontier model. Share the exploitation infrastructure so no one wastes time imitating and optimizing. In exchange, demand that every participant explores from their own distinct prism.

An ecosystem where everyone uses the same machine in the same way does not create collective intelligence — it creates an echo chamber with better grammar. Real advantage emerges when independent prisms are orchestrated to return different discoveries to the system.

Markets themselves are such an ecosystem. When too many participants exploit the same signals and data sources, edges decay faster. The real long-term advantage belongs to those who can explore from genuinely different perspectives.

The Exploration Policy

Token capital has solved exploitation. It has made imitating and optimizing free — and ubiquitous — in the mind, in companies, and across the entire world.

When exploitation costs zero, all the margin, all the alpha, and all real wealth moves upward — toward exploration. Originality becomes the last scarcity because exploration is the only work the machine cannot do for you.

Human agency now has a precise technical name: you are the exploration policy of the system. You choose which mountain is worth climbing. You introduce the heat needed to leave the comfortable hill. The machine climbs the hill you point to; without you, it only runs in circles.

The risk, at every scale, is always the same: reducing exploration to zero. The mind that delegates every difficult thought, the company that reinvests everything into optimization, and the ecosystem that collapses into a single voice all atrophy in the same way — in silence — while feeling more efficient than ever.

That’s why discipline is no longer about working faster.

It is about keeping exploration alive when everything — what is measurable, what is comfortable, what is profitable — begs you to turn it off.

Thanks for reading,

G

Garry Kasparov was about as good at chess as a human can get. And a computer that was worse than him at almost everything, except crunching numbers, beat him anyway.

Most people took the easy lesson: the machines are coming for us.

They missed the real one. To get it, you need to know how a mind actually grows. It happens at four levels, and you climb from the bottom up:

Level 1 — Imitation. You copy. The oldest one. It’s how you learned to talk. Level 2 — Optimization. You get better at the game inside the rules that already exist. Level 3 — Cross-pollination. You take what works in one field and use it somewhere it’s never been tried. Level 4 — New paradigm. You stop playing the game better and change the game itself. A whole new way of seeing.

The bottom two levels keep you inside the lines. The top two are how you draw new ones.

Machines are already far better than us at the first two. Kasparov didn’t lose to a smarter mind. He lost to something that had maxed out the bottom two levels and never got tired.

Now look at what other people did with the top two.

Demis Hassabis grew up in that same chess world. A child prodigy. He didn’t try to beat the machines at their own game. He climbed. He mixed chess with brain science, and brain science with computers — that’s level three — and pointed the mix at a problem nobody had solved in fifty years: figuring out the shape of proteins, the tiny machines that run all living things. That’s level four. A whole new approach. In 2024 he won the Nobel Prize. Not in chess. In chemistry.

Elon Musk lives on the top level on purpose. He doesn’t try to make old industries a little better. He rebuilds them from scratch, starting from basic physics. Everyone “knew” rockets were single-use. He made them land and fly again. Everyone “knew” electric cars were glorified golf carts. He made them the car every other company is now chasing. Same move, over and over: ignore the old rules, build up from the ground, create the new thing everyone else copies later.

The machine won the bottom two levels. These people won by leaving them.

Same tool, opposite results. And here’s the part that should worry you:

The machine that beat Kasparov is now in your pocket. And the easy way to use it pulls you right back down to the bottom two levels.

This morning, you reached for the bottom levels without noticing

This morning you asked a machine something you used to ask your own head.

It answered. The answer was good. You moved on.

It felt like nothing. It was one of the most important money moments of your life. Because the most valuable thing in the world just changed, and almost nobody noticed.

We’ve known how this works for over a hundred years. We just never had a machine that could do it at full scale.

In 1890, Gabriel Tarde said people don’t really reason — they copy. We copy each other’s habits, beliefs, and wants, and we do it half asleep, sure the choices are our own. Le Bon added the part that stings: in a crowd, your smarts don’t add to everyone else’s. They subtract. And Ortega named the type — the “mass man” isn’t poor or dumb. He’s anyone who feels like everybody else and feels good about it.

For most of history, thinking like everyone else took effort. You had to be in the crowd, read the same paper, watch the same show.

Now look at your feed.

The posts all sound the same now. Same rhythm. Same dash. Same “it’s not X — it’s Y” twist. Same opener: “Let’s be honest.” A thousand different people, one voice.

That’s not an accident. That’s eight billion people opening the same app and asking it what to think.

A crowd can be genius. The machine kills the thing that makes it genius.

A crowd can be smarter than any expert in it. A crowd at a county fair can guess the weight of an ox better than the judges can.

But that only works under three conditions. Everyone has to know something different. Nobody can be copying their neighbor. And there can’t be one source everyone leans on.

The reason is simple. A crowd is smart when its mistakes are independent, because independent mistakes cancel each other out. What’s left in the middle is the truth.

The moment everyone copies the same source, the mistakes stop canceling.

They pile up.

In 1945, Hayek explained why planned economies fail. What a society knows is spread across millions of minds. No single office can ever gather it all. So the planner just picks one answer and hands it to everyone. Tidy, uniform, and unable to discover anything new.

A language model is a planner like that, but for your mind.

It takes the messy, clashing, all-over-the-place output of every one of us, works out the average, and hands that average back to each person as a starting point. Then millions of people land on that same average, write it down, and feed it back in.

It feels like intelligence. It’s the biggest force for sameness ever built.

A crowd whose mistakes pile up isn’t wisdom. It’s an echo chamber with better grammar.

Here’s the part nobody is pricing

When something becomes cheap and everywhere, its price drops to zero.

Thinking — fast, smooth, average, good-enough thinking — just became cheap and everywhere.

So average thinking is now worth almost nothing.

And the second the average is worth nothing, the non-average becomes priceless.

This is the oldest rule in economics. Value lives in what’s rare. Water is free, diamonds aren’t — not because diamonds matter more, but because they’re rare.

For your whole life, thinking was the rare thing. You got paid for it.

That just ended.

The new rare thing — the only rare thing — is the thought the machine couldn’t have come up with.

Originality is the last scarcity. And rare is always where the money is.

Originality was never about the idea

So if everyone gets the same answer, where does anything new even come from?

Rick Rubin has the best answer I’ve found. He says you don’t invent ideas — you catch them. You’re an antenna. Ideas are signals in the air, and your job is to stay tuned enough to catch the ones most people miss.

And here’s the line that unlocks it: every person is a prism.

The same world hits all of us. But each of us bends it differently — through our own history, our scars, the odd things we’ve spent years paying attention to.

Originality was never in the signal.

It was always in the prism.

The machine is a prism made from a billion prisms, which is exactly why it’s none of them. It can only hand back the average.

“But it surprises me,” you’ll say. “It connects things I never would. Isn’t that new?”

No. It only remixes what’s already inside it. It blends things people have already written and lands on the spot in between. It never steps off the map.

Real originality — the kind Hassabis used, the kind Musk keeps building — comes from places that aren’t on the map at all. Your body. Your specific life. The failure you’re still ashamed of. The two fields only you happen to live between.

The machine averages what already exists.

You can make what doesn’t.

How to keep your own mind

This isn’t about quitting the machine. Refusing the most powerful tool ever built isn’t smart — it’s ego. Use it. Use it hard. Let it own the bottom two levels so you’re free to climb.

Just don’t let it think for you.

A few rules that actually work:

Write your own answer before you read anything. Make something before you take something in. If you ask the machine first, its answer becomes the anchor and you never get free of it. Get your own rough, half-wrong take down first — then check it. You can only compare against a prism you’ve already built.

Keep a list of the questions it answers badly. That list is the map of your turf — the spots where the average runs out and you begin. Live there.

Treat its answer as the floor, not the ceiling. Whatever it gives you, assume everyone else got the exact same thing. Your work starts where its answer stops.

Keep the friction. Not knowing is uncomfortable, and that discomfort is where new thinking is born. The machine kills it instantly. Sometimes you keep it on purpose.

Argue against it. Make it defend the popular view, then take it apart. Use it as a sparring partner. Never as an oracle.

What’s actually at stake

This was never really about your job.

It’s about whether you end up with a mind of your own at all.

A muscle you stop using doesn’t stay the same. It wastes away. A mind that hands off every hard thought doesn’t stay sharp. It slowly loses the ability to think hard at all. And you won’t notice the day it goes. Sleepwalkers never do.

That’s the real cost, and it builds up quietly. Not that you’ll think like everyone else — but that one day you won’t be able to think any other way.

The machine made average thinking worthless.

Stop reading that as a threat. It’s the best gift anyone has ever handed you. It means the only thing left worth being is the one thing the machine can’t be.

You. Thinking. On purpose.

Nobody is coming to make that choice for you.

You make it every time you open the app.

Thanks for reading,

G

I publish every Sunday. The silence between posts is part of the work — most weeks the market is noise.

This week is different. The biggest IPO in history trades Friday. Anthropic and OpenAI filed for theirs seven days apart. The rules of the stock market were literally rewritten for one company.

When the signal spikes, I publish. Back to Sundays after.

This Friday, SpaceX starts trading. Ticker: SPCX. Price: $135. Valuation: $1.75 trillion.

Every headline calls it the space IPO of the century.

Every headline is missing the move.

Everyone is asking the same question: is it worth that?

Honest answer? No. Not today. It loses $5 billion a year. It trades at 92 times sales. One respected research firm says fair value is less than half the price.

But “is it worth it” is the wrong question.

The right question is: what is Musk actually doing?

Stay with me. This gets good.

The company is not what you think

Open the filing and SpaceX is not a rocket company. It’s four businesses wearing one ticker:

Starlink — satellite internet. 10 million users. $11.4 billion in revenue. 63% margins. The only part that makes money.

Launch — the rockets. They earn $4 billion and spend almost all of it building the next rocket.

Starship — the giant rocket. A lottery ticket on the entire future space economy.

xAI — Musk’s AI company, now folded inside. It made $3.2 billion and lost $6.4 billion.

Read that again. The AI business loses more than the satellite business earns in operating profit.

So why would Musk strap a money-burning AI company to his crown jewel right before going public?

Because he’s not playing the game you think he’s playing.

The three wars

Everyone believes the AI war is one war: who builds the smartest model. OpenAI vs. Anthropic vs. Google vs. xAI.

In that war, xAI is losing. Last in revenue. Deepest in losses. The team was fired and rebuilt mid-fight.

But that’s only the first war. There are three:

War 1 — Models. The labs, copying each other, burning billions, racing to the same place. Brutal. Low margins. No winner yet.

War 2 — Chips. Nvidia, TSMC, Broadcom, ASML, Marvell. They don’t fight. They sell weapons to every side. Smart.

War 3 — Territory. The one nobody maps. Every AI model needs a physical home: land, electricity, cooling, connection. The stuff you can’t download.

Musk looked at War 1, saw he was losing, and did something nobody expected.

He didn’t retreat.

He changed wars.

The landlord move

While the labs fight over models, look at what xAI actually built:

• The largest single AI training site on Earth. 555,000 of Nvidia’s best chips. $18 billion in silicon alone.

• 2 gigawatts of power — what a city of 1.5 million homes consumes.

• Its own gas turbines on-site. Everyone else waits years for the power grid. Musk doesn’t wait.

• Built in 19 days what takes the industry 4 years.

He’s not trying to sell you a chatbot.

He’s becoming the landlord — the owner of the building where intelligence lives.

And here’s the part no competitor can copy. Stack the company from top to bottom:

Model → Compute → Energy → Connectivity → Orbit.

OpenAI is building compute. Meta is buying nuclear power. But nobody — at any price — owns the bottom two layers. Starlink is the connection. The rockets are the orbit.

One company owns the entire stack.

Everyone is fighting the AI war. Musk is buying the battlefield.

Now the part that makes traders sweat

Here’s where it stops being a story and becomes mechanics.

When SpaceX lists, only 3–4% of the company will actually trade. Everything else is locked for 366 days. At $1.75 trillion, that’s roughly $60 billion of available stock.

Sounds like a lot. It isn’t. Watch:

The Nasdaq changed its rules this year so giant IPOs enter the Nasdaq-100 index in just 15 days. Russell: 5 days. What does that mean in plain English? Every index fund tracking those benchmarks — the funds in your retirement account — is forced to buy. No opinion. No analysis. The rules say buy, they buy.

Estimated forced buying in the first weeks: $22–30 billion.

Against $60 billion of available stock.

Add the retail wave: SpaceX reserved an unprecedented 30% of the IPO for regular investors (normal is 10%). Demand came in at $150 billion — double the supply — at a fixed price. The overflow doesn’t raise the IPO price. It shows up Friday morning, hitting the buy button.

Massive forced demand. Tiny supply. Almost no shares to borrow, so betting against it is nearly impossible.

This isn’t a hidden trick. It’s all in the filing, in plain sight.

Musk didn’t hide the squeeze. He registered it with the SEC.

The clock

One more layer. The timing is surgical.

Anthropic filed for its IPO on June 1. OpenAI on June 8. SpaceX trades June 12.

Why race to be first by days?

Because right now, there’s no public AI company to compare xAI against. For about one quarter, the market will price Musk’s AI business on pure belief.

The day Anthropic goes public — reportedly with ~$44 billion in revenue and actual profits — investors get a measuring stick. And xAI, at $3.2 billion in revenue and $6 billion of burn, measures badly.

Going first isn’t luck. It’s the whole trade: one quarter of belief before the measuring stick arrives.

The honest part

What can break this:

• The price assumes a decade of perfection.

• Musk controls 93.6% of the voting power. No independent board. His war, his terms.

• If something happens to him, there’s no plan B at this scale.

• In 366 days the locked shares flood out. The same scarcity that pumps the price can crash it.

• You’re now also a shareholder of Grok and X — with all their lawsuits and drama included.

The real insight

Here’s what makes this IPO unlike anything before it.

In a normal company, hype and reality are separate. Excitement doesn’t build factories.

Here, it does. The belief attracts the capital. The capital buys the gigawatts, the chips, the satellites. The infrastructure makes the belief real. Which attracts more capital.

The story isn’t decoration on top of the business.

The story is the financing of the business.

Musk figured out something nobody else has: you don’t need to win the AI war if you can get the world to fund your purchase of the ground it’s fought on.

That’s not a rocket company. That’s not an AI company.

That’s the first company in history where faith is the balance sheet.

Thanks for reading,

Guillermo Valencia A

Everyone is asking the same thing: can the market absorb all of it?

That’s the wrong question. The right one is simple, and it’s right in front of us.

The wave is smaller than it looks

A company going public doesn’t sell the whole business. It sells a small piece — usually about 10%. The rest stays with the founders and early investors.

Do the math on the headlines:

• SpaceX: $1.77T → only $177B

• OpenAI: $500B → $50B

• Anthropic: $350B → $35B

• xAI: $200B → $20B

• Databricks: $130B → $13B

• Stripe: $95B → $10B

The real amount the market has to buy isn’t $3 trillion. It’s $305 billion.

Against what? About $7 trillion in cash, sitting in money-market funds, waiting.

$305B is 4% of that waiting cash. 0.5% of the whole stock market. About half a day of normal trading.

Even if twice as much traded — 20% instead of 10% — it would still be only 9% of the cash on hand.

The market won’t choke on this. It buys all of it before lunch.

So the real question isn’t can the market absorb the wave? It can, easily. The question is: which one is actually worth owning?

Being able to buy it doesn’t mean you should

Here’s the trap.

The fact that the market can handle a stock tells you nothing about whether you should own it. And the history of IPOs is very clear.

Take 80 tech companies that went public. Set each one to 100 on its first day. Two years later, the typical result is 109.

Up 9%. In two years. That’s what you usually get from buying “the IPO wave.”

But look at the range. NVDA: +703%. BILL: +618%. MA: +510%. And at the bottom, the losers — down 90%, gone for good.

Now ask the question almost no one asks: what did the big winners have in common?

It wasn’t luck. They were monopolies. Each one owned a market no rival could break into. So they kept the profits — year after year — that competition would have wiped out.

The typical company was fighting competitors. The big winners had no competition.

Buying “the IPO wave” means buying the typical company — the one stuck in a fight — and hoping you got lucky with a monopoly.

You don’t want the wave. You want the one monopoly hidden inside it.

The rule

One rule explains the whole chart: competition is for losers.

It’s not a slogan. It’s simple math. Companies that compete lose their profits to the fight. Monopolies keep them. The fighters pay for the winners.

Now look at the wave as a map of who’s fighting whom:

OpenAI. Anthropic. xAI. Databricks.

Same market. Same chips, same models, same customers, the same engineers moving between them. They copy each other and burn cash doing it. The more alike they are, the harder they fight — and the thinner their profits get.

None of them is a monopoly. They’re all in the same knife fight.

History proves the point. Almost every company goes public small next to the biggest company of its era — the typical one is just 3% of it. Only 9 of 40 were ever big enough to count as mega-IPOs. The rare giants — like Aramco, which listed bigger than Apple — aren’t fighting for the top spot. They’re already there.

SpaceX is the only real monopoly here

Out of the whole list, one company isn’t fighting anyone. And it isn’t close.

It owns the launch business. In 2025, SpaceX put 2,213 tons into orbit — more than 80% of everything the world launched. The next-biggest company carried about 12% of that. SpaceX flew 165 missions: more than half of all the launches on Earth, and 85% of all satellites. Because it reuses its rockets — something no rival has matched — it has pushed the cost of reaching orbit down to about $2,720 per kilogram, versus $54,500 on the old Space Shuttle. Some boosters have flown 20-plus times. This isn’t a lead. It’s a different game.

Starlink is the money machine. On top of the launch business sits Starlink — the biggest satellite network ever built, about 65% of all working satellites in space, with 10.3 million customers in 150-plus countries. It has roughly doubled its customers every year for three years. In 2025 it made about $11.4 billion (around 61% of all SpaceX revenue) at high margins, and produced almost all of the company’s ~$8 billion profit. In just the first quarter of 2026, Starlink made $3.26 billion in revenue and $1.19 billion in operating profit — while the launch and AI parts lost money. One monopoly is paying for the other.

It goes public as the leader, not a challenger. This is the whole point of the first chart. At $1.77 trillion, SpaceX doesn’t list as 3% of the leader. It lists as the leader — one of the most valuable companies in the world on day one. It’s the only name in the wave that starts at the top, not the bottom.

This isn’t a bet on “space.” It’s the one company in the wave that fits the RACE idea — compute, electricity, the physical bottleneck — and the only one already making real monopoly profits.

Even the small piece that trades helps. That ~3% traders call “risky because it’s thin” is actually the tell: SpaceX doesn’t need the money. It’s letting out a small piece because it can. A small piece of a real monopoly doesn’t go cheap.

Everyone else is fighting for a slice of the market. SpaceX is the market.

What could go wrong (a story with no risks is a fairy tale)

I’m not a believer. I’m a detective. Here’s what could break it:

• The price is very high. $1.77 trillion on about $16 billion of 2025 sales is roughly 109 times sales — higher than Tesla or Palantir ever were. You’re paying today for ten years of success.

• It depends on one man. Musk is both the reason for the high price and the biggest risk. If he left suddenly, the stock would reprice overnight.

• The thin float swings both ways. A tiny tradable piece means wild price swings. The first weeks will be rough.

• Customers pay less now. Starlink chose growth over price — revenue per customer fell 18% (to about $81 a month) even as the customer base quadrupled. Growth is real, but pricing power has a limit.

• The edge can be attacked. If Amazon’s Kuiper grows, Starlink’s profits get squeezed. The “computers in orbit” story is a maybe, not yet money — and part of the price already assumes it works.

If those risks worry you, it’s not your trade. A pricey, jumpy monopoly isn’t for everyone. But it is, without question, a monopoly.

The bottom line

The wave isn’t the trade.

The wave is noise — $305 billion the market buys before lunch. The fear that it “can’t be absorbed” is a distraction.

Look closely and it comes down to one thing: of six companies, five are fighting each other for a piece. One is the whole pie.

Buy the leader. Not the group.

Kind regards,

G

Not investment advice — just a way to think. Do your own research.

It’s easier to start with the mirror image. What does everyone agree on?

Right now they all repeat the same line: “AI is a bubble.”

They’re right. That’s exactly the problem.

I lined up two charts — the dot-com Nasdaq and the AI Nasdaq — each starting at its catalyst: Netscape in 1995, ChatGPT in 2022. On a log scale, so we read rates of change, not headline points.

They trace each other almost perfectly. Same slope, same scares, same recoveries.

Now look at where we are: June 2026. Before the inflection.

In the dot-com run, the steep part — the part that minted and erased fortunes — didn’t begin at the catalyst. It began around late 1999, more than four years in. We haven’t reached that line yet.

If the pattern holds, the peak prints around 53,000 in mid-2027.

The most violent gains of this cycle are still ahead. So is the trap.

Thiel’s insight: the top of a bubble is also a peak of clarity. The crowd is usually right about the future and wrong about who will own it. Pets.com died. Amazon, at six dollars, didn’t. The value didn’t vanish — it rotated, from page views to infrastructure, from the slogan to the data center.

The bubble didn’t destroy the future of the internet. It destroyed those who confused the catalyst with the asset.

Soros explains the mechanics. Prices don’t just reflect fundamentals — they shape them. He called it reflexivity. A rising AI Nasdaq lowers the cost of capital, which funds real buildout, which produces real revenue, which seems to justify the prices, which lifts them again.

Belief funds the thing that proves the belief. That’s why the line goes vertical.

And reflexivity always breaks the same way: the descent is steeper than the climb. The dot-com Nasdaq took eighteen months to triple, and gave it all back in thirty.

Here’s the part people get wrong about Soros. He didn’t flee bubbles — he rode them, eyes open, hand on the exit.

The herd makes one of two mistakes: ride it blind with no exit, or call it and miss the most explosive leg of a generation. Both are just letting the crowd think for you.

So where do you actually stand?

Reflexivity has one enemy: atoms.

A narrative can compound without limit. But the loop only runs as long as the physical world can keep up — and it can’t, because the real economy is rate-limited by physics.

You can’t talk a gigawatt into existence. You can’t will a fab line or a high-voltage interconnect into being on the schedule the narrative demands.

That wall of atoms is two things at once. It’s why the boom eventually breaks. And it’s where the durable value lives — because owning a scarce physical resource pays whether the narrative climbs or collapses.

I order it with four letters. RACE. Robotics. AI. Compute. Electricity.

The “A” — the brain — is becoming a commodity. Abstract intelligence gets cheaper every quarter. The value is leaving the layer that thinks and moving to the layer that acts.

Three of those bottlenecks already have names you can own. The wires, the watts, and the body.

The wires — Marvell

Everyone watches the chip. Nvidia. The brain.

But one chip is useless. AI runs on thousands of them stitched into a single machine — and the stitching is the hard part. Data has to move between them at impossible speeds, without choking.

Marvell builds that connective tissue: the custom silicon and the high-speed interconnect and optics that move data across a data center. Not the brain — the nervous system.

And here’s the quiet shift. As the chip itself becomes a commodity, the scarce thing is everything around it. The hyperscalers are designing their own custom chips to stop paying Nvidia’s tax — and that custom silicon, plus the wiring that links it all together, runs straight through Marvell.

You cannot build a giant cluster without it. A bottleneck that gets more valuable as the brain gets cheaper.

The watts — Bloom Energy

Every model, every query, every cluster ends as a request for one thing: electricity.

And the grid can’t keep up. New data centers need power now, and waiting years for a connection is no longer an option. This is the constraint almost nobody priced in: AI isn’t bottlenecked by intelligence anymore. It’s bottlenecked by power.

Bloom Energy makes fuel cells that generate electricity on-site — right next to the data center, no combustion, no five-year wait for the utility.

It doesn’t sell a story about the future. It sells electrons today.

When the AI narrative is soaring, those electrons are scarce. When the narrative crashes, those electrons are still scarce — because the world keeps needing power whether or not the market believes in it that quarter.

A bottleneck that doesn’t care about sentiment.

The body — Tesla

Here is the thesis that still sounds like heresy: Tesla can be worth more than Nvidia within three years.

Not because Nvidia weakens. Because of where the value is going.

The internet has already been mapped. Every webpage, every sentence, every line of code — that’s the dataset the language models ate. The brain is nearly solved, and a solved thing becomes cheap.

The physical world has not been mapped.

How objects move. How force transfers. How to grip, balance, walk, and manipulate matter in a messy, unscripted room. None of that lives in text. It has to be collected from reality — and whoever maps the physical world owns the next era the way Google owned the last one.

That is the robots thesis. Intelligence is going to leave the screen and enter the world, in bodies. Humanoid robots. Autonomous machines. The layer that acts.

And mapping reality requires three things almost nobody has together: a body you can build at scale, eyes that perceive the world, and a model of physical reality trained on real-world data.

Tesla has all three. It already manufactures bodies at scale — that’s what a car company is. It already runs the largest real-world perception fleet on earth, mapping physics through billions of miles of driving. And it’s pouring that into Optimus — a body for the intelligence, built on the same vision stack.

Nvidia sells the shovels for the brain. Tesla is building the body that turns intelligence into work in the physical world.

When value migrates from the layer that infers to the layer that acts, it migrates from the one to the other.

The reflexive bubble inflates everything in the index. The bottlenecks are what’s still standing when it deflates — because they were never narrative. They were always atoms.

So don’t ask “is it a bubble?” The answer is yes, and it leads nowhere.

Ask two better questions. From Thiel: what bottleneck will still be standing — and still being built — when the music stops? From Soros: where am I on the curve, and where’s my exit?

The most contrarian thing isn’t opposing the crowd. It’s thinking for yourself: accept the peak is coming, ride the parabola with your eyes open, and own the atoms that outlive it.

When the music stops, the models will be cheap, the hype will be gone, and three things will still be scarce:

The wires that move the data. Marvell. The watts that power it. Bloom Energy. The body that turns it into work in the real world. Tesla.

2026 is not 1998. But it rhymes.

Thanks for reading,

G

Guillermo Valencia A

Cofounder of MacroWise

Bitcoin is falling. Gold, silver, and copper are surging. Credit is healthy and volatility is asleep. Read together, these are not the fingerprints of a liquidity drain — they are the fingerprints of a liquidity rotation. Money is leaving the dollar-speculative complex and the optimized export economies, and flowing into the physical bottlenecks of a new industrial era: energy, metals, networks, compute, and — least understood of all — robots. The regime has changed. We have left Globalization and entered Cold War II: a competition to define the dominant designs of AI, robotics, electricity, and monetary power. In that regime, rates are structurally higher, equities still rise, and the volatility is not noise. The volatility is the map.

Now the investigation.

The body: a death that isn’t one

Start with the corpse. Bitcoin has rolled over hard, down to 62,072 from its highs. That matters, because Bitcoin is the most liquidity-sensitive asset we own — the canary for US dollar liquidity, the first thing to fall when the dollar plumbing tightens. So the dollar system is narrowing.

But walk three feet to the right and the scene refuses to cooperate. Gold is at 4,365. Silver — gold’s high-beta cousin, the asset that amplifies every liquidity swing — is vertical at 69.10. Copper, the metal with a PhD in global growth, is pressing 6.28.

A liquidity crisis kills these together. Instead, one canary is down and the others are climbing. That single divergence reframes the entire case: this is not money disappearing. It is money moving.

There are two liquidity systems here, not one. Bitcoin reads the dollar. Gold and silver read Chinese liquidity and the global hunger for hard money. They are splitting apart because they are measuring two different forces — and the gap between them is the case.

The alibi: credit is alive

Before chasing the money, rule out the catastrophe. If this were a real liquidity crisis, credit would seize first. It hasn’t.

• Total bank credit: +6.4% year over year.

• Loans to businesses (C&I): +7.7% — firmly expansionary.

• Corporate credit spread (Baa − 10Y): 1.54 pp, against a historical median of 2.14. That is the 8th percentile of financial stress in the entire history of the series.

• VIX: 15.40 — also near its 8th percentile. Median is 18.

Read those percentiles slowly. The market’s stress gauge and its fear gauge are both nearly as calm as they have ever been. And the yield curve, which spent years inverted whispering “recession,” has un-inverted and re-steepened: 10Y − 2Y at +0.38, 10Y − 3M at +0.77. Both positive.

The body has a pulse. A calm one. So we are left with the only question that matters: where is the money going?

Follow the money: long the exporters, short the importers

Copper at 6.28 and WTI crude up +64.3% year over year are not subtle. Dr. Copper does not rise into a contracting world, and a 64% jump in the energy impulse is not demand destruction. The money fleeing financial speculation is landing in the physical economy.

The currency board confirms it — and resolves into one clean, tradable axis. The dollar is strong at 118.88. The yen is weak at 159.23. The Korean won has blown out to 1,503.96.

Most analysts call this “EM versus DM” and move on. They miss the real fault line: energy exporters versus energy importers.

Japan and Korea are the purest energy importers in the developed world. Japan buys over 90% of its energy abroad; Korea imports roughly 84% of its primary energy and nearly all of its fossil fuels. They are also the heart of the semiconductor complex. That is the trap: they sit at the most critical choke point of the AI buildout, fabricating the silicon everything depends on — yet they must buy their energy on the open market. With crude up 64%, their terms of trade decay from both ends. The currency is the release valve, and it is venting.

The exporters are the mirror image. Colombia, Brazil, Norway sell into rising terms of trade. Every tick of the crude curve and every ton of copper is a tailwind to their currencies, not a tax.

So the divergence is not noise, and it is not abstraction. It is a position:

Long the energy exporters. Short the energy importers. The semiconductor importers are the squeezed cohort — indispensable to the next era, and penalized by it at the same time.

The red herring: “technology is deflationary”

Now the misdirection everyone leans on. AI will crush prices. Ignore the inflation — the machines will eat it.

The data says otherwise. The Producer Price Index — prices at the factory gate, the goods that become everything else — is running +9.8% year over year. The Consumer Price Index, the inflation your wallet feels, is +3.8%. PPI leads CPI. The wholesale layer is already running near ten percent while the retail layer stays calm. That gap is not deflation. That gap is a pipeline.

So challenge the premise directly: technology does not automatically create deflation. It creates deflation only at one specific stage — and we are not in it.

Look at the first Cold War. In its early decades, the dominant designs of the microprocessor, of consumer electronics, of computing had not yet been settled. The deflationary, globalizing wave did not arrive with the technology. It arrived later — in the 1980s and 1990s — once standardization, scale, and supply-chain optimization had matured. Deflation is what the optimization phase produces.

We are not in the optimization phase. We are in the competition phase.

The motive: a war over dominant designs

Here is the motive that explains every clue at once. The world is now fighting to define the dominant designs of the next era — AI, robotics, compute, electricity, autonomous systems, strategic networks. Nothing is standardized. Everyone is building competing, redundant, sovereign versions of the same stack.

Why is fragmentation necessarily inflationary? Ronald Coase answered the question in 1937, long before anyone called it macro. Firms exist, he showed, because using the market is not free. Every transaction carries costs — searching for a counterparty, negotiating, writing the contract, enforcing it, guarding against betrayal. When those costs are low, you buy on the open market. When they are high, you make it yourself.

Globalization was the greatest transaction-cost compression in history. Stable rules, dollar settlement, enforceable cross-border contracts, low political risk — the whole planet became one cheap marketplace. So firms stopped making and started buying: anywhere, from anyone, at the lowest price. Value chains dis-integrated across forty countries because coordination had become nearly free. Disinflation was, in large part, transaction-cost deflation. The −0.26%/yr glide in rates during the Globalization era was its signature.

Cold War II runs the film backward. Oliver Williamson sharpened Coase with two variables that now haunt every boardroom: asset specificity and uncertainty. A chip fab, a rare-earth refinery, a sovereign AI cluster are extreme-specificity assets — priceless in the right hands, hostage in the wrong ones. And uncertainty — sanctions, export controls, a partner’s government weaponizing the relationship — has gone vertical. When specificity and uncertainty rise together, the theory gives exactly one answer: stop buying, start making. Internalize. Integrate. Build the redundancy.

So the world reshores, stockpiles, friend-shores, and builds parallel sovereign stacks. Each move is the rational response to higher transaction costs — and each is more expensive per unit, by design. The frictions globalization stripped out are being bolted back on: re-priced into PPI at +9.8%, and into the risk premium capital demands to fund any of it. This is the deep reason supply chains are now being rebuilt for redundancy instead of efficiency.

The conclusion is blunt: inflation here is not the printing press. It is the price of trust withdrawn from the system — and the structurally higher rate is simply that price, expressed as yield. You do not get the cheap, single-stack world that produced forty years of disinflation. You get its opposite.

And the scale is staggering. By the IEA’s reckoning, the five largest technology companies spent more than $400 billion in capital expenditure in 2025 — and that figure is set to rise around 75% in 2026. Data-center electricity demand is on track to roughly double to about 945 TWh by 2030 — close to the entire electricity consumption of Japan. In the United States, data-center power demand is projected to climb 130% by 2030; by the end of the decade, the country is set to burn more electricity running data centers than producing aluminum, steel, cement, and chemicals combined.

That is why compute is now a power problem before it is a chip problem. And it is why rates in this regime are not only pricing inflation. They are pricing geopolitical risk, supply-chain redundancy, energy competition, and strategic scarcity. The interest rate is the toll on a fragmenting world.

The long-term rate chart shows exactly this. The US 10-year has moved through distinct regimes: the Volcker shock added roughly +0.61%/yr, Globalization quietly bled it down at −0.26%/yr for a generation — and we have now flipped into “Rates 2.0,” rising at +0.62%/yr. That slope rhymes with Volcker, not with Globalization. The trend points to 6.5% by 2050.

For the monetarists: money is the unit, not the cause

A serious objection deserves a serious answer. Many readers hold that inflation is, in Friedman’s phrase, always and everywhere a monetary phenomenon — too much money chasing too few goods. If that is the whole truth, this entire structural story is noise, and only the money supply matters.

The objection is half right. The half it misses is the half that counts.

Start from the identity the monetarists themselves use: money times velocity equals prices times output — MV = PQ. Friedman’s dictum holds the real economy, velocity and productive efficiency, roughly constant, so prices track money. That assumption is the entire game. It is also exactly what Cold War II destroys.

A world rebuilding for redundancy instead of efficiency is a world deliberately lowering its own output per unit of input. Reshoring, dual supply chains, strategic stockpiles, energy competition, sovereign compute — each is a choice to spend more real resources to make the same goods. The denominator falls on purpose. So even with the money supply held perfectly still, prices rise — because the real cost of producing the basket has risen.

That is not a monetary impulse. It is a structural one.

Inflation may be a monetary phenomenon — but the floor beneath it is a real one. Money sets the number on the scale. The world decides the weight.

This is why no central bank can tighten its way back to the old regime. It can suppress the print, but not the floor — all it controls is who pays, growth or prices. The structurally higher rate is the toll on a fragmenting world, and the toll is real before it is ever monetary.

The reframe: higher rates are not a bear market

This is where the obvious suspect gets it wrong one final time. Rates are rising, so equities must fall. History says no.

Cold War I was a strong regime for equities — and a powerful one for commodities, industry, infrastructure, and energy. A world competing to build is a world spending capital at scale, and capital deployed becomes someone’s revenue. Rising rates and rising equities are co-symptoms of a buildout, not a contradiction.

The probability distributions agree. Translated into year-end returns:

Read the ratio, not the headline. Both indices offer close to three units of upside for every unit of downside. The S&P carries the tighter ratio — a shallower left tail. The Nasdaq carries the bigger absolute upside, paid for with a deeper one. That is the higher-beta trade stated precisely: more upside per dollar, slightly more pain per dollar, the ratio roughly preserved.

And note how tight both distributions are: the 25th percentile of each index now sits at or above today’s price. Fewer than one in four simulated paths ends the year below spot. The market is not pricing a coin flip. It is pricing drift, with a floor.

So the verdict is not “crash.” It is violent rotation inside a rising tape.

The clue everyone is missing: intelligence needs a body

Every eye is on compute. The entire market is pricing the brain — the accelerators, the training clusters, the silicon that thinks. That is exactly why the most important clue is being overlooked.

Intelligence, as biology actually built it, was never pure compute. Life’s strategy was never one giant cortex in a vat. It was embodiment and demography: millions of bodies, each carrying sensors and actuators, mapping reality by interacting with it. Intelligence scaled through population and contact with the physical world — not through raw thinking-power alone. The map of reality was drawn by an army of embodied agents, in parallel, over deep time.

Silicon intelligence today is a brain with no body. It can reason about the world; it cannot yet touch it, sense it, or act on it at scale. To cross that gap it needs precisely what biology needed — bodies, by the million, gathering real-world data and executing in physical space. That layer is robotics, and it is the single most underestimated bottleneck in the entire buildout. The market is fully priced for the cortex and barely priced for the limbs.

This reorders the hierarchy of scarcity. Compute is becoming abundant — capital is racing to manufacture it. Scarcity migrates to the next bottleneck, and the next bottleneck is the embodiment layer: the fleet, the sensors, the real-world data engine, the humanoid program.

Which forces the provocation this thesis cannot avoid. As of mid-2026, Nvidia is worth roughly $5 trillion — the most valuable company on earth, the brain of the buildout. Tesla is worth roughly $1.4 trillion — and is converting factories to build humanoid robots at scale, sitting on the largest real-world driving dataset in existence. If intelligence is valuable only once it can act on reality, then value migrates from the brain to the body as compute commoditizes.

Stated bluntly: on this thesis, Tesla could be worth more than Nvidia within three years. Not because the brain stops mattering — but because once compute is everywhere, the scarce thing becomes the body that turns intelligence into action. The market is staring at the semis. The next move is in the robots.

The verdict

The body looked like a death. The autopsy says relocation. Money is leaving the dollar-speculative canary (Bitcoin) and the energy-importing export economies (the yen, the won), and flowing into the bottlenecks of the next civilization-scale buildout:

• Networks — the Nokia and Ericsson backbone, the Marvell silicon that moves data between accelerators.

• Energy — GE Vernova turbines and grids, Bloom Energy fuel cells, because compute is a power problem first.

• Embodiment — compute, electrification, and the robots that let intelligence touch the world.

This is the new Cold War: a competition for compute, robotics, electricity, AI infrastructure, and monetary power. Its signature will not be a clean trend. It will be turbulence, rotation, and dispersion — because no dominant design has yet been crowned.

So do not read the volatility as randomness, and do not read it as the end. Volatility is the market discovering the bottlenecks of the next civilization-scale buildout — and every spike is a clue about where the next decade gets built.

The detective’s last note: the crime was never a robbery. It was a relocation. Follow the money to the bottlenecks, and you find the trade.

Thanks for reading,

G

Guillermo Valencia A

Cofounder of MacroWise