Ever heard of “ boomerang employees” ? They are basically people who got laid off because AI replaced them, but for some reason the company needed hire them back to the same exact role they had before.

And there already estimates that say that half of the companies that cut customer service staff because of AI will be rehiring for those same roles by 2027 ( if they don’t go down before that).

In engineering the picture is not that different. At companies that bet hardest on AI replacing developers, a meaningful share of the people getting hired back are the same people who got let go.

In theory its was very simple. It always is….

If AI could write code, review it’s own code, debug it, and build simple applications in minutes, why would you need junior developers and maybe even developers at all?

The problem was that companies didn’t really test that idea and just jumped straight to it.

What could possibly go wrong?

To be fair, we know that a lot of these “AI layoffs” were just really a good opportunity to deal with the over hiring that was done during and post covid. The point still stands.

Subscribe now

Eventually, someone measured it

A 2025 study by METR, an independent research group that studies AI capabilities, decided to follow sixteen experienced developers on completing 246 real tasks on projects that they already knew well.

The developers expected AI to make them faster. Surprise surprise, it made them 19% slower.

Afterwards, when the researchers asked how it went, the developers guessed they had been about 20% more productive. Every single one of them was wrong, and every single one of them would have bet money on being right.

AI wrote plenty of code, that was not the problem. What slowed everyone down was what came next. Reviewing it. Catching the mistakes hiding in it. Making it fit into a system that already existed before the AI even showed up.

But wait! It gets worse. A separate study also found out that AI-generated code had 1.7 times the bug rate of code written by a person, and produced more of it overall. Someone will need to maintain it.

Claudeonomics

Meta employees built a leaderboard called Claudeonomics. The idea was to rank each other by how many AI tokens they burned.

Higher usage meant a Higher rank, and a title waiting at the top. “The Token Legend”.

In 30 days, employees had burned through somewhere between 60 and 74 trillion tokens. At industry pricing, that’s something like $900 million. That is 3 times bigger than Portugal’s entire annual budget for culture, youth, and sports. But most things are.

I can understand creating the pillars for the future before you fully understand the tool. And if you know you will need it anyway, why the hell not?

But to be honest, if you didn’t at least try the tech, you would be, for lack of a better word, a moron.

Actually, I lied. I have tons of better words. But they’re all just a bunch of insults and profanity.

But i cannot understand how far we went with this idea without any real results.

But i’m just a guy on the internet… what the hell do i know?

Subscribe now

If you happen to have a life insurance policy, an annuity, or a pension, there is a rule that you might never heard of.

Also, your money is being invested in AI, but more on that later.

It’s called : “ State Guaranty Association”. → It Says that if a life insurer ever fails, every other insurer in the state gets assessed to cover the shortfall.
Sounds like a cool safety net to have right?

In most of the cases, those competitors get that assessment back as credit against the premium taxes that they owe to the state.

So the bailout is not really funded by the insurance, because it is funded by the state, and state is obviously funded by the taxpayers.

And just so that you can generate cool cat pics, a ton of debt is now tied to AI.

Something like 1 trillion dollars. But it is worth it right?

1 trillion represents around 15% of the whole investment grade bond market. Greater than the total banking sector exposure. How cool right?

The rules

You might be thinking “But wait …. didn’t we build rules for exactly this after 2008?”

Yes, smarty pants. We did.

Regulators built two specific fixes for securitization, the practice of bundling loans and selling them to investors as bonds.

The first problem was that nobody really knew what was inside these bundles and to be honest they didn’t really care… as long as they were making money, everything was fine.

The fix to this problem was Regulation AB, which requires asset-level disclosure and ongoing reporting for anything classified as an asset-backed security.

Next problem was the “originate-to-distribute” model.

This model allowed the banks to write the loans, package them, sell them, and walk away without having to check if the loan was any good or not ( most of it was total shit).

Have you seen The Big Short, by the way? Great movie.

The fix was Regulation RR, requiring anyone packaging debt for investors to retain at least 5% of it themselves. So that they had some skin in the game.

These rules are still in force. But reportedly, a law firm working on AI securitization deals, recently asked the SEC if these rules apply to loans backed by data centers.

And guess what? The SEC's answer was no.

Their logic was that a data center is a physical asset, a building, not a financial asset that amortizes over time the way a mortgage or a car loan does. Fair enough, technically. A building is a building.

But if it’s not classified as an asset-backed security, then the regulation AB doesn’t apply and the regulation RR doesn’t apply either.

This means that the rules built after 2008 to stop this kind of problems from happening again don’t cover the biggest new category of structured debt currently being built. Makes sense ?

Have we learned nothing form the past?

How do life insurers make a profit?

It’s very simple honestly …

In very simple terms, they collect premiums, and invest what is collected while it waits to become a claim. That’s it…

The money they invest, while it sits there waiting, is called “The Float”.

Here’s the funny part.

If the insurer is a good investor and makes more profits than expected, policyholders don’t see a cent of it. Their return is fixed by contract, set in stone before any of this even happens.

These investments are usually on boring stuff. Bonds, real estate, corporate debt. By default boring, reliable, and predictable … But enough about myself.

Then the interest rates went lower and lower, but insurers had made promised based on a higher rate world… so they were basically fucked … or were they?

Enter private equity. whatever that means…

Private equity funds are cool. They raise money on a ten year cycle or more, invest it, and return it after collecting the profit. Wish it was me…

This means that they are always raising capital. And insurers started to look really good to them with their billions of dollars in capital.

And they weren’t really shy about it. Private equity ownership of life insurers went from close to nothing in 2009 to several hundred billion dollars in assets across well over a hundred insurers by 2024.

Total insurance assets under private equity control are estimated to be somewhere around $1.5 trillion, according to research from Daniel Oliver at Myrmikan Capital.

The conflict

Athene is a life insurance company owned by Apollo, a private equity corporation.

Annuities, or the retirement funds of regular people, are held by Athene.

Apollo also engages in loan origination. Athene is therefore a natural buyer when Apollo seeks a buyer for a loan that it originates.

Apollo provides a loan. Apollo also owns the business that is purchasing the loan.

On paper, Apollo and Athene appear to be one person with similar interests. However, in reality, they are not.

Apollo uses fees to generate revenue for its stockholders.
Regardless of the performance of the underlying loans, Athene's policyholders receive a fixed return.

One group benefits at the expense of the other when funds are transferred from Athene to Apollo.

Apollo isn't unique.

Global Atlantic is owned by KKR.

Blackstone has investments in many insurance companies.

American National is owned by Brookfield.

All the cool kids are doing it.

But where does the money even go?

If you’ve been living under a rock, i need to tell you something.
You ready?

AI infrastructure is reaaaaaallyyyyyyy expensive.

Shocker, right?

The largest tech companies are spending more than their cash flow covers, which is why they need to borrow.

Google even issued a bond that matures in 2126. A hundred-year bond. To be fair, there are trees older than that in my village. But still…

Anyway…

Building a data center directly would show up as a massive expense on a tech company’s own balance sheet and pressure its stock. So what the guys with the suits do instead is, create a separate entity to build it instead (SPV), and the tech company signs a long lease promising to rent it for years.

And that lease, is the only thing that makes the project financeable most of the time, because the SPV can then go into a lender with a Microsoft / Google / Meta signature and borrow against it.

However, the company’s own records almost never mention that lease.

According to Goldman Sachs, these businesses have lease obligations totaling around $1.5 trillion, of which about $1 trillion doesn’t show up on their balance sheets.

But the construction money still has to come up from somewhere, and banks are increasingly not the ones who lend it due to capital constraints that make agreements this large extremely challenging or even impossible.

That is when private credits steps in.

Who rates it?

A conservative insurer needs a rating before it can buy a bond. The rating determines the capital charge. A higher rating means a lower charge, which in practice means permission to buy more.

Fun fact. Some of the largest private credit sponsors, including Blackstone and Apollo, are now reportedly among the most frequent clients of rating agencies outside the traditional big three.

Former employees claim they were pressured to inflate ratings to win business, and one sponsor is reportedly under investigation over payments to a ratings agency.

Allegedly.

We’ll leave it there. The accusations are still contested and unresolved.

What happens the risk?

A big part of the risk is going offshore through reinsurance.

If you don’t know what reinsurance is, i will keep it simple and short. Insurance for the insurance companies.

Athene reportedly moved close to all of its $200 billion in retirement obligations to a Bermuda affiliate. Guess it was because of nice weather there?

According to the Bank for International Settlements, the share of US life insurer reserves held by offshore reinsurers rose from about 14% to about 40% since 2017.

Words are cheap. Here's the shape of it .

And if the loan goes bad? The insurer takes the loss first. If it can’t cover it, the state guaranty fund does. If the guaranty fund pays out, every other insurer in the state gets assessed, then quietly reimbursed through a tax credit.

Tons of words just to say “your money.”

Why you? Because the money was sitting there. What were you going to do with it anyway? Pay rent, buy food? Kinda overrated if you ask me.

Remember that rule i mentioned in the beginning? The one about who covers the shortfall when an insurer fails?

You know it now. And you also know who pays for it in the end.

But in all honesty, what else did you expect?

PS: Don’t forget to watch "The Big Short”.

Subscribe now

Something that has not happened in 28 years

The US recently spent its own money defending another country's currency. Coordinated with Japan and supported by Korea, the intervention injected approximately $87 billion into the market in two days, lifting the yen from a near four-decade low.

The last time the United States did something like this was 15 years ago, and the last time it was this direct was decades earlier.

The interesting element isn't the intervention itself. It is how the rest of the world learned about it. A Reuters photographer captured a memo belonging to the US Treasury Secretary that said, roughly, "to do: buy Japanese yen, 5 to 10 billion."

Someone that senior does not typically need to jot down the ticker symbol for a currency he trades professionally, or a dollar amount too little to impact a market worth more than a trillion dollars each day.

It makes little difference whether the note was accidental or intended. In any case, it informed every trader shorting the yen that the US government was now on the opposite side of the deal, which is often enough to shift behavior.

Why the US actually cares

The official story is about solidarity with an ally. The more informative explanation is located underneath. When the yen falls, Japanese import prices rise, pushing Japanese inflation higher.

Higher inflation causes the Bank of Japan to raise interest rates, which it would rather not do on this timetable. And if the Bank of Japan is forced to hike interest rates, it will be tempted to sell US Treasuries in order to strengthen the yen and defend its currency at home.

Japan is the largest foreign holder of US government debt, at more than $1 trillion. If that holder becomes a large-scale seller, US rates and borrowing costs climb, exacerbating an already stressed fiscal picture. The United States is not acting to assist Japan because of goodwill. It is acting because Japan's currency crisis foreshadows a bond market catastrophe that is on the verge of becoming America's.

The debt behind the decision

The US government debt surpassed $40 trillion for the first time in the week this memo was published. It stood at $34.5 trillion in March 2024. That's $5.5 trillion added in around two years, with no recession, epidemic, or conflict requiring fresh deficit spending to fight. The debt just increased on its own track, quicker than it had previously.

Moody's downgraded US national debt from Aaa to Aa1 in May 2025, becoming the third major rating agency to do so, after S&P in 2011 and Fitch in 2023. It was not the cause of the yen's current situation. However, it articulated the same underlying problem that is now pushing the United States to intervene in another country's currency market in order to defend its own bond market.

Three things, pick two

The United States is attempting to accomplish three objectives at the same time.

1. Reshore production so that it does not rely on competitors for chips and weaponry.

2. Keep prices steady, since voters will forgive practically anything except the cost of groceries and gas.

3. Maintain the whole economy and bond market running smoothly enough to keep tax revenue flowing to service the $40 trillion in debt. Only two of the three are available simultaneously.

Reshoring necessitates a weaker dollar because American manufacturers cannot compete with Chinese ones while the dollar makes US goods expensive and Chinese goods inexpensive. However, a weaker currency raises the cost of all US imports, which directly contradicts price stability. Keep the currency strong instead, and inflation remains controllable, but reshoring stalls, and the 30-year Treasury yield, which recently reached levels not seen since 2007, continues to rise as the market prices in the cost of an unresolved fiscal trajectory.

Something has to give, and both reshoring and bond market stability are viewed as non-negotiable. Which leaves price stability as the only option on the table.

Weakening the dollar without saying so

Any administration cannot formally devalue the dollar since the moment it is stated, every holder of dollar assets begins selling in expectation of the change. The alternative is to carry it out discreetly, utilizing behaviors that are individually deniable. In this intervention, the US bought yen in euros rather than dollars, suggesting that it never technically sold a single dollar on its books, despite the fact that the end outcome, a stronger yen and a lower dollar in relative terms, was the same.

Watch the same variables mentioned in earlier posts here: the 10-year and 30-year Treasury yields, the strength of the yen, and the price of gold. If long-term yields keep climbing while stocks fall, pressure is building toward some kind of policy response. If gold moves up while the dollar moves down at the same time, that is usually the clearest signal that a debasement trade is back in motion, whether or not anyone in Washington ever describes it that way.

My eToro portfolio is public if you want to see how I'm positioned

The op-ed

On June 23rd, US Treasury Secretary Scott Bessant published a Wall Street Journal op-ed titled “Hamilton Inspires Trump’s Economic Statecraft.” Two weeks earlier, the US trade representative had given a speech at Davos covering the same ground. Both were public, on the record, and named explicitly what the administration believes it is trying to do. Both largely disappeared into the news cycle within days. This is worth starting with because what follows is not speculation about private intent. It is an attempt to read what has already been stated out loud.

What even is Hamiltonian economics ?

Alexander Hamilton’s report on manufacturers, submitted to Congress in December 1791, set out a two-part framework: tariffs on foreign goods, and subsidies for domestic industry. The explicit goal was to protect young industries until they were large enough and efficient enough to compete globally. The implicit logic was that a nation which cannot make the things it needs is not fully independent regardless of what its constitution says. If the economy depends on an adversary, the political sovereignty is largely notional. Hamilton did not invent this logic. He borrowed it from the playbook that Britain had already used to build its own industrial dominance, and that Germany, Japan, and South Korea would later use in turn.

This became the US operating model for roughly 150 years. As the Davos speech noted openly, every country that has become an industrial superpower used the same model at some point, followed by preaching free trade once it had already won. Britain ran protective policies through its rise and switched to free trade advocacy in 1846 after its industrial dominance was secure. The US did the same through the 1800s. Germany, Japan, and South Korea followed similar paths in their own periods of industrialisation. The rules of the game tend to be written by whoever just won it, and the first demand of the winner is usually that nobody else be allowed to play the same way.

The Bessant op-ed named five principles, the most concrete of which are these: economic security starts with national capacity, meaning the ability to manufacture domestically rather than depend on foreign supply chains. Openness will be matched by reciprocity, meaning market access becomes conditional rather than assumed. Financial leadership is a central instrument of statecraft, meaning the dollar’s role in global trade is a lever to be used, not a neutral fact of life. The remaining two are stated in language broad enough to support almost any interpretation. What the first three describe, read from the outside, is an attempt to reverse a fifty-year process of de-industrialisation and re-establish the conditions under which domestic manufacturing can compete. That process has a well-documented historical shape, and it creates a specific problem the op-ed does not resolve.

The triangle

The problem is that three things the administration says it wants cannot all be true simultaneously. Rebuild domestic industry. Protect Main Street from rising prices. Keep the dollar strong. Pick any two. The third becomes unavailable.

A strong dollar is what makes American manufacturing uncompetitive in the first place. When the dollar is expensive relative to other currencies, American labour and American production costs are expensive relative to foreign alternatives. Factories move to where costs are lower. The arithmetic is not conspiratorial. A financialised economy operating under a reserve currency the rest of the world needs to hold makes offshore manufacturing the rational choice. To rebuild domestic industry at scale, the dollar has to come down. There is no other mechanism. If you want industry back and Main Street protected from the inflation that would accompany a weaker dollar, you cannot have both a cheaper dollar and stable prices. Something gives.

The deal most Western economies ran for fifty years looked roughly like this: consumers got cheap goods because manufacturing moved to lower-cost countries, asset owners got richer on paper because financialisation inflated valuations, and the industrial base that used to support middle-class wages gradually hollowed out. In the US, which is the clearest case because the data is most visible, the things that could be manufactured elsewhere got dramatically cheaper since 2000 — televisions down roughly 98%, toys down 74%. The things that could not be offshored moved the other way — housing up 111%, healthcare services up over 280%. The same broad pattern, in different proportions, played out across most of the industrialised world. Reversing it requires accepting that some of what got cheaper will get more expensive again, and that the people who benefited most from the financialised version are not the same people who will benefit most from what comes next.

The analyst Luke Groman, whose work the Bessant op-ed draws on in part, argues that Bessant is too sophisticated not to know this, and that he has already signalled which of the three he is prepared to sacrifice. The dollar. Not catastrophically, not overnight, but gradually and deliberately, using something else as an escape valve to manage the adjustment without triggering a financial crisis.

The escape valve

US gold reserves are valued on the Treasury’s books at a statutory price set in 1973 and never updated. At that price, roughly 8,000 tons of gold is worth approximately $11 billion. At current market prices, the same reserve is worth closer to a trillion dollars. A revaluation to market price would add that value to the balance sheet without issuing new debt. The Hamiltonian framing provides the reason why this might be necessary rather than merely a curiosity.

If the dollar is going to weaken as part of a managed re-industrialisation, something has to absorb the adjustment. A reserve asset that is not any single country’s currency, that cannot be printed, and that has historically served as the denominator through which trade imbalances get settled, is the candidate that keeps appearing in serious policy discussions. The World Bank president named it in 2010. The former IMF chief economist named it in 2016. China’s central bank named it in 2009. The US trade representative named it at Davos this year. None of these are fringe positions. They are, at different points, the stated views of the people running the institutions that manage the current system.

The $38,000 number

The price figure that appears in analyses of this framework, roughly $38,000 per ounce, is not arbitrary. The structural calculation runs as follows: China’s annual trade surplus runs in the range of $1.2 trillion. Against an annualised gold flow capacity of roughly 1,000 tons, the implied settlement value comes out in the high $30,000s per ounce. The precise figure shifts depending on which year’s data you use and how you define the relevant flow, but the order of magnitude is stable. At current gold prices, around $4,000 per ounce, gold is too cheap to function as a global settlement asset at the scale required. The number is not a price target. It is the output of asking what price would be required for the arithmetic to work.

This does not mean gold will reach $38,000, or that it is headed there on any particular timeline. It means that if the theory holds, the current price is the obstacle rather than a signal that the theory is wrong. The People’s Bank of China has been buying gold for twenty consecutive months through mid-2026, including its largest single monthly purchase since October 2023, during a period when the gold price fell nearly 30% from its peak. A central bank buying through a significant price decline is not behaving like a financial investor. It is accumulating an asset it expects to need for a purpose other than near-term return.

What is already visible

Several things predicted by this framework are already observable in the data. Central banks globally have been buying roughly 1,000 tons of gold per year for three consecutive years, roughly double the pace of the prior decade. Gold has been the largest US export category since the fourth quarter of last year, overtaking oil, pharmaceuticals, and aircraft engines. The majority goes to Switzerland for refining and then to China via Hong Kong. The dollar has been weakening against the yuan since late March, which is the opposite of what would be expected if a war were successfully pressuring China. Chinese government bond yields are steady or declining while most other major economies face rising borrowing costs.

None of these data points individually proves the framework. Taken together, they are consistent with it in a way that would be difficult to explain through coincidence. The more interesting question is whether the adjustment, if it is happening, is being managed toward a controlled outcome or whether the conditions that would allow a controlled outcome are already compromised.

The two roads

The Groman analysis identifies two ways this ends. The first is that the re-industrialisation project fails. Debt continues growing, confidence eventually breaks, and the financial architecture that has been in place since 1971 comes apart the way these things have come apart historically, meaning not cleanly. The period between 1922 and 1945 is the relevant precedent: depression, currency collapses, and a world war before a replacement system emerged. Gold skyrockets in that scenario because it is what people reach for when paper systems fail.

The second road is that the plan works. Re-industrialisation proceeds, trade gets rebalanced, and the world transitions gradually toward a system where a neutral reserve asset settles imbalances rather than one country’s currency absorbing all the pressure. In that scenario, gold also rises, because the revaluation is the mechanism through which the transition happens rather than an accident of crisis. Both roads lead to the same destination. The variable is how much damage occurs on the way there.

The trick

American corn farming uses roughly 20 trillion gallons of water per year. Total estimated global AI data center water use is around 260 billion gallons. Corn uses approximately 80 times more water annually than all of the world’s AI servers combined. Around 40% of that corn crop is converted to ethanol and burned as vehicle fuel. Not eaten. Burned. Every gallon of that ethanol carries an irrigation footprint of around 1,500 gallons of water, much of it drawn from aquifers that do not meaningfully recharge. You have probably not seen that comparison in any of the coverage you have read about AI and water. That absence is more informative than any of the individual figures.

The numbers in this debate get selected, framed, and released by people with a strong interest in what they appear to say. The water figures being cited on both sides of this argument are, in most cases, technically defensible. They are also, in most cases, deliberately incomplete. The question of how much water AI uses has a different answer depending on where you draw the boundary around the system you are measuring, and the boundary is almost never disclosed when the number is presented. The same mechanism runs through most resource debates. The gap between the smallest defensible number and the largest defensible number is several orders of magnitude, and both are being cited by people who believe they are being accurate.

The opening exchange in the debate shows exactly how it works.

The exchange that started it

Sam Altman, the CEO of OpenAI, stated publicly that the average ChatGPT query uses roughly a fifteenth of a teaspoon of water. Researchers pushed back almost immediately. One academic told the BBC that the Altman figure was perhaps referring to some tiny models, and that for a medium-sized large language model of the scale of GPT-3, ten to fifty queries would consume roughly 500 milliliters of water once you include cooling and electricity generation. Half a litre, not a fifteenth of a teaspoon. The BBC asked OpenAI for more details about how the per-query figure was calculated. OpenAI declined to respond.

That independent estimate used GPT-3 as its reference model. Most people querying ChatGPT today are on GPT-5.5, with more advanced models available on top of that. Both are considerably larger and more compute-intensive than GPT-3. The 500ml figure is already conservative by the time you read it, and the gap between it and Altman’s number is probably wider in practice than the research implies.

The inference step and everything that came before it

When you type a question into ChatGPT and hit send, something runs. Processors fire, a model generates tokens, cooling systems respond to the heat load, water evaporates somewhere in a data center. Inference, start to finish. The water consumed at that moment, for that query, on infrastructure that already exists and is running regardless of whether you ask anything, is small. A fifteenth of a teaspoon is a reasonable approximation. Except it is not quite one query. Complex questions trigger what are called reasoning steps, internal sub-queries the model runs without showing you. It generates an initial response, evaluates it, runs a follow-up, sometimes triggers a search, iterates until it decides it is done. One user-facing query can represent five or ten backend queries, none of which appear in the per-query figure being cited. That still does not produce a large number. Even multiplied by ten, it fits comfortably in a kitchen analogy. But it means the inference figure is softer than it looks, even before the larger boundary question enters.

None of this is independently verifiable in full, because the primary source controls much of the underlying data. OpenAI does not disclose how many sub-queries run behind a typical user interaction, what the training water footprint of any specific model was, or the full methodology behind the per-query figure. The BBC asked and was declined. The debate is therefore happening almost entirely on terms set by one of the parties to it, which matters when evaluating any figure that comes from that direction.

A proposed large data center campus in Utah was reported to emit the equivalent heat of 23 Hiroshima bombs per day, a figure about heat rather than water, but the same mechanism. That number is, apparently, approximately correct. One physicist noted it is also less than half the heat the sun delivers to the same square footage every day. Same number, different denominator, completely different emotional register. The figure was not fabricated. The baseline was not provided. The water figures run on the same mechanism: a large absolute number, stated without the context that would let a reader know whether it is large relative to anything that matters.

But the inference step is the last step in a very long chain, and the chain is where the water goes. Building a large language model requires training it, which means running enormous clusters of GPUs continuously for weeks or months. Not weeks in the sense of a project timeline. Weeks in the sense of the machines never turning off, drawing power and generating heat around the clock, cooling systems running at full capacity the entire time. According to estimates from researchers at the University of California, training can account for roughly 50% of the total resource use of an AI model over its entire operational lifetime. Half the water the model will ever be responsible for gets consumed before anyone asks it a single question.

And that is only training. The models in active use today are not the ones being trained. Somewhere else in the same company, on clusters of hardware that do not overlap with the production systems, next-generation models are being trained right now. Bigger, more capable, more resource-intensive than the current ones, because that is where the industry is explicitly pointed. The data center buildout everyone is arguing about is not for what exists. It is for what comes after. Including or excluding the training cost of models not yet deployed is a choice, and it produces numbers that look completely different while both being defensible. If you want the water number to look small, exclude training and count only inference. If you want it to look large, include training and the power generation overhead and the manufacturing footprint of the hardware. Both approaches have an argument for them. Neither is neutral.

The power plant layer

Before the power plant layer, a distinction in the data center literature that almost never makes it into the headlines matters enormously for interpreting any figure you encounter. Water withdrawal is the total amount of water removed from a source. Water consumption is the fraction that does not come back, mostly because it evaporated. A cooling system that withdraws a billion gallons but returns 80% of it tells a different water story from one that consumes 80% of what it takes. The majority of data centers in the United States currently use evaporative cooling, where GPU heat is dumped directly into water and roughly 80% of it is lost to the atmosphere as vapor. High consumption relative to withdrawal. But a power plant drawing from a river and returning most of it downstream is high withdrawal and much lower consumption. Both activities involve large numbers. They are not the same activity. The number you encounter in a headline almost never specifies which one is being counted, which means you are often comparing figures that are not comparable.

A further layer makes the larger figures even larger, included or excluded depending on the argument being made. Data centers use electricity. A significant portion of that electricity, in most parts of the world, still comes from thermoelectric power plants. Natural gas, coal, nuclear. All of them work on essentially the same principle: generate heat, expose water to that heat, expand it into steam, drive a turbine, then cool the steam back down. The US Geological Survey estimates that electricity generation accounts for roughly 40% of all freshwater withdrawals in the United States. Large, but somewhat misleading in isolation, because most of that water gets returned to its source rather than evaporating away, which makes it different in character from the water a data center loses to evaporation in a cooling tower.

Whether to include the water flowing through a power plant as water use attributable to the data center drawing its electricity is a methodological choice without a clean answer. The case for including it: the data center created the demand, the power plant fulfilled it, the water moved because of that chain. The case against: the power plant was running anyway, the water mostly goes back, attributing it to one specific consumer is arbitrary. Both arguments are made in published research. The choice between them can shift the total figure by an order of magnitude. Scope selection is where most of the interesting decisions in resource accounting happen. Nobody is inventing figures. They are choosing boundaries.

Local

Aggregate global figures, whether for AI or corn or anything else, obscure what actually determines whether a water use is a problem: the local hydrological budget. A data center drawing from a watershed in Norway, where annual precipitation is abundant and aquifer recharge is fast, is a different situation from a data center drawing from an aquifer in Arizona that has been overallocated for decades and does not meaningfully recharge. The same infrastructure, the same cooling systems, the same GPUs, two completely different water stories depending entirely on where they sit.

More than 160 new AI data centers have been built across the United States in the last three years alone, a 70% increase from the prior period. Several of them are in the southwest, in regions where the Colorado River system is already severely stressed, where aquifer levels have been declining for a generation, and where municipal water systems were already under pressure before a single server rack arrived. There are documented cases of communities losing water pressure, finding contaminants in well water, and watching municipal treatment systems become overloaded by the waste discharge from nearby data center construction. These problems are tied to particular places and particular infrastructure decisions, not a uniform planetary crisis that the global aggregate figures imply.

Data centers are built flat and wide rather than tall and narrow. Waste heat from a skyscraper rises and dissipates. Waste heat from a sprawling single-storey campus rolls along the ground and into adjacent communities. One physicist calculated that a large proposed campus in a Utah desert valley could raise local nighttime temperatures by as much as 28 degrees Fahrenheit. Permanent, local, and invisible to any water use figure. The communities closest to these facilities absorb costs the aggregate statistics were never designed to capture.

The water debate as it is currently conducted focuses almost entirely on the global aggregate, a large number in absolute terms but small relative to agriculture and to power generation. A more useful version of the same debate asks which specific facilities are being built in which specific watersheds, what the local system could absorb before construction began, and what it can absorb now. A planning question, a permitting question, a question about who has standing to object. Planetary scale makes for a better headline. Local hydrology makes for a better decision.

The bigger problem

Water is the more emotionally salient resource because wasting water feels viscerally wrong in a way that wasting electricity does not. We have all been told since childhood to turn off the tap while brushing our teeth. The moral intuition that water is precious runs deep. It is not misplaced. But the projected increase in power demand from AI infrastructure is, by most serious estimates, a larger problem than the water footprint, and it is getting less attention.

The buildout currently planned would push data center electricity consumption up by an order of magnitude relative to existing infrastructure. Power grids in most of the world are not built for it. Carbon budgets are not sized for it. Electricity prices will reflect it, in every household and business in the regions where the data centers concentrate. The water impacts are geographically bounded, hitting hardest in places that already had a water problem. The electricity impacts are more widely distributed and harder to localise. River health has very little to do with your electricity bill. They are separate problems. But if the question is which one will matter more to more people over the next decade, the answer is probably not the one generating more headlines right now.

Underneath both of them, and almost never mentioned: whether the buildout actually produces the productivity gains being assumed. The entire capital expenditure argument, hundreds of billions of dollars in data center investment, is premised on AI continuing to improve in ways that justify the infrastructure being built for it. If the improvement curve flattens, or if adoption stalls at a level below what the infrastructure was sized for, then the water and electricity conversation will look different in retrospect.

So where is the honest boundary to watch? Not the per-query figure, which is technically defensible but deliberately narrow. Not the aggregate global figure, which overcounts by conflating withdrawal and consumption and attributing power plant water to specific users.

The honest boundary is local: which specific facilities are being built in which specific watersheds, what is the actual stress on those systems before and after, and who bears the cost when the local hydrological budget runs short.

It is the only version of this debate where the numbers mean something.

$42 per ounce

On the United States government’s official books, gold is valued at $42.22 per ounce. That price was set by law in 1973 and has never been updated. At that price, the entire US gold reserve, roughly 8,000 tons of it sitting in Fort Knox and the New York Fed, among other places, is worth approximately $11 billion on the Treasury’s balance sheet.

Gold is currently trading at around $4,000 per ounce. That means that at market prices, those same 8,000 tons are worth closer to a trillion dollars. The difference between $11 billion and a trillion dollars is an accounting rule from the Nixon era that nobody has bothered to change in 53 years. It is sitting there right now, quietly, while several other things are happening around it.

What paper gold actually does

Most gold that trades in the major western markets, London and the Comex in New York, never moves. The buyers and sellers are trading contracts, which are paper claims on gold, not the metal itself. The majority of people holding these contracts have no intention of ever taking physical delivery.

The practical consequence is that the number of paper claims in circulation can, and does, significantly exceed the amount of actual gold sitting in vaults. When that happens, the market prices gold against the apparent supply, which includes all the paper, rather than the actual supply, which is only the physical metal. More paper claims in circulation means the market behaves as though there is more gold than there actually is. This keeps the price lower than it would be if only physical metal were being traded.

Nobody publishes a reliable number for how large that gap is, because nobody is required to. What is observable is the behavior of the market participants who presumably know, or suspect, that the gap exists.

The clearest recent evidence that this dynamic is real, rather than theoretical, came from the silver market. In January of this year, spot silver hit an all-time high, and at its peak, the price difference between physical silver and paper silver reached something close to 40%. People were paying a significant premium to hold the real thing rather than the contract.

The spread has since narrowed but has not returned to zero. This represents a persistent premium for physical over paper that did not exist at the same scale before. Gold’s equivalent spread is much smaller. An efficient-market reading would take this as evidence that gold’s paper market is more honest. The alternative reading is that gold’s paper market is simply larger and more deeply entrenched, meaning the suppression is more complete rather than less present.

Central banks have been buying physical gold at historically elevated rates for several years running. In the first quarter of this year alone, central banks bought a net 244 tons, the strongest first quarter on record. The World Gold Council notes that this figure includes an estimate for undisclosed purchases, meaning gold that is being accumulated but not officially reported.

China has been reducing its US Treasury holdings over several years and rotating the proceeds, at least in part, into gold. Gold is currently the number one export out of the United States, and has been for several months. The country that prints the world’s reserve currency is shipping its physical gold elsewhere. Whether that reflects arbitrage, repatriation by foreign holders, or something more deliberate about where the weight of the system is shifting is genuinely unclear, but the direction is unambiguous.

China’s two-part move

On June 24th, the Industrial and Commercial Bank of China announced it was shutting down paper gold trading for retail investors, effective July 24th. It was not the first to do so. The Postal Savings Bank of China moved first, followed by Ping An Bank, and then China Guangfa Bank. Several of China’s largest financial institutions, one after another, pulled the plug on the exact same product within the same window.

The official explanation is investor protection: gold has been volatile, margin requirements have risen to 140% of position value, and retail traders are getting hurt. The less official explanation is that China is interested in what gold actually costs when leveraged paper speculation is removed from the price-setting process.

The second part of the move is being built in parallel. China is launching a new gold clearing and settlement system running out of Hong Kong, working in conjunction with the Shanghai Gold Exchange. The design is straightforward: Shanghai sets the price based on physical delivery, meaning real metal has to move when trades clear, rather than paper claims changing hands. Hong Kong functions as the entry point for foreign participants who cannot access Shanghai directly due to capital controls.

Put them together, and what you have is a parallel infrastructure for gold price discovery sitting outside the London and New York systems, built on physical settlement and denominated in yuan. Hong Kong is reportedly expanding its physical vault capacity from roughly 200 tons to over 2,000 tons. That tenfold increase is not the kind of investment you make for a paper market.

A 2014 speech by the head of the Shanghai Gold Exchange at the London Bullion Market Association, the western paper gold pricing system, was explicit about this intent. He stated that when China has the right to speak in the international gold market, gold’s true price will be revealed. That was twelve years ago. The infrastructure being built now is the mechanism for making that statement operational.

Why this connects to the dollar problem

The post-Bretton Woods arrangement has run for over fifty years on a specific logic: countries hold their reserves in US dollars and US Treasury bonds because dollars are the currency of global trade and Treasuries pay interest. The arrangement worked as long as everyone agreed to keep playing by the same rules.

Foreign central banks stopping the growth of their Treasury holdings a decade ago, and more recently selling them, suggests the arrangement is under at least some pressure. Trading an interest-bearing asset for gold, which pays nothing, is only rational if you trust the gold more than you trust the paper promise. That calculation is being made by enough central banks, at a large enough scale, to show up clearly in the data.

If China succeeds in building credible physical gold price discovery outside the western paper markets, and if enough global trade begins to settle against that system rather than the dollar system, the dollar’s role as the default unit of account weakens at the margin. This is not a prediction that the dollar will collapse. It is an observation that the architecture underneath it is being contested in a way that is far more concrete than most geopolitical commentary tends to acknowledge.

The US counter-move, and the $42 rule again

Which brings the accounting rule back into focus. The gap between $42 and $4,000 per ounce represents roughly a trillion dollars sitting on the US government’s balance sheet, valued at a price nobody has used since Nixon.

A revaluation to market price, achievable in principle with a simple change to the statutory price, would add that trillion dollars to the Treasury’s books without issuing a single new bond. The Treasury Secretary has spoken publicly about monetizing the asset side of America’s balance sheet. There is also a proposal from economist Judy Shelton for a 50-year Treasury bond that could be redeemed in either dollars or physical gold. This would give the US bond the same kind of gold anchor that China is building into the yuan.

Whether any of this happens, and when, remains genuinely unknown. The Fed has published research on the gold revaluation idea, but it has not been implemented. It may not happen at all. But the fact that it is being discussed seriously, at the exact same moment China is building a competing settlement system, suggests that the question of what truly backs a currency is being revisited in practice, rather than just in theory.

This connects to something covered in an earlier post in this series: the gap between the statutory $42 price and the current market price is not a quirk, it is a policy option sitting dormant on the balance sheet.

Subscribe now

Source : Summary of Economic Projections

Fifty four, forty five

Warsh was confirmed by the Senate 54 to 45, the closest vote of the modern era for a Fed chair. Only one Democrat crossed over. At his confirmation hearing, Senator Elizabeth Warren called him a sock puppet for Trump, who has spent years demanding lower rates and once posted that anyone who disagrees with him will never get the job. Warsh’s answer was direct. He said he would be an independent actor. That sentence meant nothing on the day he said it. Every nominee says it. What he actually did with his first meeting is the only thing that could make it true.

A chair who arrives this way has exactly one asset that matters before anything else: whether the market believes him when he says independent. Cut rates in week one, on cue, with the president who appointed him publicly expecting exactly that, and the bond market does not hear monetary policy. It hears an instruction being followed. It prices in a risk premium for a captured central bank, and that premium shows up as higher long-term yields regardless of what the short end does. So the first move available to a chair in his position is a demonstration before it is a rate decision. Hold, while your own committee leans toward higher rates, while the president who put you there wants the opposite. That is the only way to spend the one asset you actually need before you can spend anything else.

The data gave him cover, but it was also just true

Headline CPI printed at 4.2% this spring, driven largely by the energy shock from the war. The Fed’s own median projection has the year-end rate at 3.8%, not lower, and inflation has now run above the 2% target for more than five years. Nine committee members are projecting a hike before December. None of this required Warsh to invent a hawkish posture for political cover. The data simply did not offer him a dovish one. Whatever the credibility calculation underneath it, the hold was also the only honest reading of the numbers in front of him.

The trade, if the credibility theory holds, runs further out than this single meeting. Hold now, visibly, while the data still justifies it. Let the war resolve, or claim it has. Let oil fall. Let the inflation numbers come down on their own schedule. Then cut, and say the data told him to. Nobody gets to say he was told to.

Whoever picks the ruler picks the conclusion

Two readings of inflation exist right now, and the entire plan depends on which one the Fed eventually treats as authoritative. Headline CPI and PCE put it at 4.2%, with the Fed’s own forecast bringing that to roughly 3.6% by year end, still well above target. A second story points to oil having already fallen sharply from its highs, and if the region stabilizes, that decline pulls headline inflation down on its own. A shift toward an alternative measure, the Dallas Fed’s trimmed mean for instance, strips out exactly the volatile components oil falls into, and the number moves closer to target almost by definition, not because underlying price pressure actually changed but because the ruler changed. Choosing the measurement tool is, in this case, very close to choosing the conclusion.

How the bond market answered

None of that ruler-switching has happened yet, which means for now the bond market is still grading Warsh on the numbers as they currently stand, and its verdict arrived within hours. The mechanism needs a steep yield curve to function, short rates low, long rates high, so banks have an incentive to lever up and absorb Treasury issuance. What the market actually did after the meeting was the opposite of what the plan needs. The short end of the curve, the 2-year Treasury, moved up the most. The long end barely moved. That is curve flattening, and it is the market explicitly pricing in higher rates for longer rather than the relief the mechanism depends on. Dropping forward guidance compounds this. When a central bank stops signaling its intentions, investors price in the uncertainty itself, a risk premium for not knowing, which pushes yields up independent of where actual policy ends up. Layer onto this a problem that predates any of this: the US needs to refinance roughly $8 trillion in debt over the next 12 months, into a market where foreign central banks have stepped back as reliable buyers and where, by some estimates, leveraged funds based in offshore jurisdictions have been absorbing a large share of new issuance instead. That is a buyer base that can be forced to sell into exactly the kind of volatility a flattening curve and a closed strait both produce.

The deal that lasted days

All of this assumed the war would resolve on a workable timeline and it has not. The agreement signed between the US and Iran came apart within days. Israel continued strikes against Hezbollah targets in Lebanon, and Iran closed the Strait of Hormuz again. Whatever credibility Warsh banked at his first meeting now has to last through a war that just reopened rather than one winding down.

Trump was asked about the decision to pause strikes and answered with numbers rather than victory. He talked about oil flow and reserves running out in a matter of weeks, nothing about anything happening on the ground. That lines up exactly with the Strategic Petroleum Reserve timeline flagged in the earlier post in this series. The reserve is what is actually being watched here, far more than the war.

The checklist

Tracking whether this plan is working does not require anyone’s interpretation of Fed language. Five signals are commonly cited as what successful execution of this kind of mechanism should produce: a weaker dollar, higher equities, lower long-term Treasury yields, higher gold, and higher Bitcoin. In the period immediately following the meeting, every single one moved in the opposite direction. Dollar up, stocks down, long-term yields up, gold down, Bitcoin down. None of this means the plan is dead. Mechanisms like this are not won or lost in a single week. But it means the plan, by the standard its own proponents use to judge it, is currently failing, and it is failing at the exact moment the war it depends on ending just got worse instead of better.

The base rate for new chairs

Going back through transitions to new Fed chairs, the S&P 500 has tended to fall in the first three months of a new chair’s term, averaging a meaningful decline, with the largest drops historically associated with chairs who represented the biggest break from their predecessor’s approach. The chairs associated with continuity tended to see smaller, sometimes negligible declines. Warsh is explicitly not a continuity pick. He has said as much himself, in the shorter statements, the dropped forward guidance, the abstention from his own projection. Whether that historical pattern repeats is unknowable in advance. It is at minimum a reason to expect more volatility before less.

Subscribe now

Portugal tem um dos níveis de mobilidade socioeconómica mais baixos da Europa. Nasce-se, em boa parte dos casos, no rendimento que se vai ter. Quem nasce em baixo fica perto de baixo. Quem nasce em cima tende a ficar em cima. Há sempre desvios, e acho que vão continuar a haver, mas servem mais para confirmar a regra do que para a desafiar. Os estudos sobre o tema mostram isto vezes sem conta, e acho que é a moldura que falta à maior parte das conversas sobre oportunidade neste país.

Há uma frase que se ouve muito em Portugal quando se fala de quem fez dinheiro há trinta ou quarenta anos: “Tudo parecia possível naquela altura.”

Tem alguma verdade. Nos anos 80, quando o mercado de capitais português reabriu, as regras ainda não estavam escritas. Ou estavam mal escritas. Ou ninguém sabia bem como as aplicar. Quem lá estava, com a formação certa e os contactos certos, inventava coisas que hoje seriam impensáveis. Estar no sítio certo quando o sítio ainda não tinha porteiro valia mais do que qualquer talento.

Isso já não existe, e ainda bem.

Mas acho que é por isso que a frase irrita quem a ouve hoje, mais do que inspira.

Como se fecha uma fronteira

Maquiavel já tinha percebido isto há quinhentos anos. O poder não se distribui por mérito. Distribui-se por quem está posicionado quando a ocasião aparece. É uma descrição precisa, sem precisar de cinismo nenhum, de como as coisas funcionam quando ninguém está propriamente a vigiar a porta. O Príncipe não ensina como merecer o poder. Ensina como reconhecer a fortuna, que é o nome que Maquiavel dava à ocasião, e como agir antes que ela passe. Quem está bem posicionado quando a fortuna aparece tem vantagem. Quem não está, raramente a alcança a tempo, por mais virtude que tenha.

Uma fronteira económica fecha-se por acumulação, nunca de um dia para o outro. No início ninguém sabe bem quais são as regras, porque o jogo é novo. Há espaço para inventar e improvisar. Produtos, estruturas, formas de fazer as coisas que mais tarde se tornam impraticáveis, quando já há trinta empresas a fazer a mesma coisa e nenhuma vantagem em ser a trigésima primeira.

Depois chegam os reguladores, que fazem falta. Depois os incumbentes, que consolidam posição antes que mais alguém apareça. Depois a maturidade do mercado, que é só outro nome para a ausência de vazio. Cada fase faz sentido isoladamente. O resultado acumulado nunca é o mesmo: uma porta que esteve aberta uns anos fecha-se de vez, e quem ficou de fora nunca mais a vê aberta da mesma maneira.

Mesmo enquanto a porta esteve aberta, exigia bagagem. Capital, formação, uma rede de contactos a servir de aval informal quando ainda não havia aval formal nenhum. Quem tinha essa bagagem arriscava, e às vezes ganhava em meses o que outros levavam uma vida a acumular. Quem não tinha ficava de fora, tal como fica hoje. Só que hoje já não existe nem a possibilidade teórica de arriscar, mesmo para quem tem a bagagem certa. As regras estão todas escritas. Os lugares estão todos ocupados. O que sobra para quem chega agora é competir por aquilo que já tem dono.

Ainda assim, acho que vale a pena ver de perto como era essa fronteira. A memória dela tende a ficar mais bonita do que foi. Nos primeiros anos depois de o mercado de capitais reabrir, havia regras escritas por quem nunca tinha feito as contas certas, e ninguém com autoridade para as corrigir percebia o suficiente de finanças para discutir o erro. Bastava saber um pouco mais de matemática do que a média para ganhar discussões a gente bem mais sénior. Era um vazio de competência institucional, temporário, que se fechou assim que o Estado contratou gente que sabia o que estava a fazer, e não mérito distribuído por ninguém. Havia outros vazios do mesmo tipo. O Estado tinha nacionalizado empresas em 1975, mas a indemnização aos antigos donos demorou anos a chegar, e quando chegou foi em papéis de dívida que rendiam quase nada. Pelo meio, a inflação tinha disparado para perto dos 20%, e cada ano de atraso corroía ainda mais o valor real do que esses papéis prometiam pagar. Quem entendia o mecanismo, e sabia como agir antes que sobrasse ainda menos valor para recuperar, ganhava exactamente isso: a diferença entre saber e não saber, não entre merecer e não merecer.

A mão invisível precisa de espaço para mexer

Adam Smith escreveu A Riqueza das Nações a partir de uma ideia hoje citada tantas vezes fora de contexto que perdeu a parte que lhe dava sentido: cada um a perseguir o próprio interesse, sem mais preocupação nenhuma, agrega-se em benefício comum. Mas a mão invisível é uma consequência, não uma fórmula mágica, e as consequências dependem de premissas.

As premissas são entrada livre e concorrência real. Smith descrevia um mercado onde, em teoria, qualquer pessoa com uma ideia melhor ou um preço mais baixo conseguia entrar e competir. Parece simples. Testar a ideia contra o mercado. Ganhar ou perder por mérito próprio, sem ninguém a vigiar a porta antes de a deixar entrar. Quando os lugares já estão preenchidos há gerações, e quem está dentro tem todo o interesse em manter a porta fechada, a mão invisível não desaparece, continua a trabalhar. Só que se vira para dentro. Reforça quem já estava bem posicionado, em vez de premiar quem teria a melhor ideia, se alguma vez chegasse a testá-la.

A teoria de Smith depende tanto da segunda condição quanto da primeira. Citar só a metade que interessa apaga a metade que na realidade faz a diferença.

A virtude que não tem onde aparecer

Dezasseis anos antes da Riqueza das Nações, Smith publicou um livro menos lido, mas que na minha opinião é muito mais interessante: “Teoria dos Sentimentos Morais”. Defende que o interesse próprio carrega sempre uma componente social, porque o ser humano precisa de ser visto, julgado e aprovado pelos outros para se sentir realizado. O esforço de cada um procura também simpatia, no sentido que Smith dava à palavra, mais próximo de reconhecimento do que de pena.

Esta ideia pressupõe mobilidade. Pressupõe que o esforço bem aplicado tem um sítio onde aparecer. Um palco. Uma audiência disposta a reconhecer o mérito quando ele aparece, e espaço estrutural para esse mérito se traduzir em alguma coisa concreta: um emprego, um negócio, um lugar na sociedade que corresponda ao esforço feito.

Quando os lugares já estão todos preenchidos, o talento genuíno fica sem onde aplicar essa energia. Pode trabalhar tanto quanto quiser, aperfeiçoar-se ao ponto que quiser, e continuar sem palco nenhum. A virtude de Smith pressupõe um teatro com bilhetes disponíveis. Quando a sala está cheia há décadas com os mesmos nomes, a peça pode ser excelente, mas ninguém de fora a vai ver representada.

Em Portugal o teatro nem sequer tem o bilhete certo à venda. A economia continua organizada à volta de colar solas de sapatos, servir cafés e arrumar quartos de hotel. Não há país a sério que tenha crescido só com turismo. O trabalho de baixo valor paga rendas e as vezes nem isso, e quem tem talento para mais do que isso sabe-o desde cedo. Os melhores cérebros saem, como já se falou aqui tantas outras vezes, e o teatro fica com o público que sobra para ver a mesma peça de sempre.

O que ocupa o lugar da fronteira

Sem fronteira financeira como a dos anos 80, a geração de agora tenta sempre a mesma estrutura com roupagem diferente.

Crypto teve, durante uns anos, a textura de uma fronteira sem porteiro: regras pouco claras, espaço para inventar com fartura, dinheiro a aparecer do nada para quem chegasse primeiro. Atravessa agora a mesma maturação que o mercado de capitais português atravessou nos anos 90. Reguladores a chegar. Incumbentes a consolidar. Vazio a desaparecer. Startups prometem meritocracia e continuam a premiar sobretudo quem já tem capital próprio ou acesso a quem o tenha. Conteúdo online parece democrático pela barreira de entrada baixa, mas o sucesso real concentra-se em quem já trazia audiência, rede ou capital cultural de outro lado.

E há, cada vez mais, gente a sair. A lógica é simples: se a fronteira não existe aqui, talvez exista lá. Ou exista, pelo menos, a possibilidade de competir num mercado de trabalho onde os lugares não estão ocupados pelo mesmo grupo de pessoas há trinta anos.

Estas são fendas, não fronteiras. Espaços pequenos que ainda não fecharam de todo, onde se entra com pouco e se tenta alguma coisa antes que cheguem os reguladores, os incumbentes, a maturidade, a fila toda outra vez. Em relação aos anos 80 mudou a escala. As fendas de hoje são muito mais pequenas. Fecham-se mais depressa.

O que a falta de mobilidade produz

Quando a mobilidade social é baixa durante gerações suficientes, não fica só a desigualdade. Ficam também os sistemas que se constroem por cima dela, cada um a resolver, à sua maneira estreita, o problema de viver num país onde subir é difícil e descer é fácil. Nenhum nasceu de mau-carácter coletivo. Nasceram de gente a tentar sobreviver dentro das regras que tinha. O resultado acumulado é uma engenharia de dependências que se sustenta a si própria.

A pirâmide etária está invertida. Há mais gente a sair da vida ativa do que a entrar nela, e não há substituição que chegue para inverter isso a tempo. É aritmética, simples e sem margem para opinião. Numa sociedade com mobilidade alta, uma pirâmide invertida seria só um problema de financiamento. Numa sociedade onde a mobilidade já era baixa antes da demografia se inverter, é também um problema de quem fica a competir por cada vez menos lugares, à medida que cada vez menos gente nova entra para os disputar.

Depois há a arquitetura de dependências que se construiu por cima dessa demografia, e que existe precisamente porque, durante décadas, depender do Estado foi a forma mais segura de subir, ou pelo menos de não descer. Pensionistas, que descontaram a vida toda e têm todo o direito ao que lhes foi prometido, mesmo sabendo que as contas não fecham como fechavam quando o sistema foi desenhado. Uma função pública enorme, onde uma parte significativa trabalha com ferramentas e métodos de outra época. Pertence a uma geração a quem a escola prometeu exatamente isto: estuda, entra para o Estado, tens a vida resolvida, porque o mercado privado nunca te ia oferecer essa mesma segurança. Mudar esse sistema agora não é só uma reforma técnica. É destruir, para essas pessoas, a única forma de mobilidade que a vida lhes ofereceu e que elas cumpriram à risca.

E há as empresas que dependem do Estado como cliente único ou principal, muitas vezes pequenas, municipais, ganhando concursos que cheiram a sobrinho, primo ou a cunhado. Isto também é mobilidade, só que da pior espécie. Não se sobe por produzir mais valor. Sobe-se por estar ligado a quem distribui o contrato. Uma ou duas destas empresas não seriam problema nenhum. O problema é que são centenas, talvez milhares, e a soma de pequenas ineficiências locais vira uma ineficiência nacional do tamanho de uma montanha. As pessoas envolvidas nisto não são o problema. Quase nunca são. O sistema que torna essa forma de subir mais fácil do que competir a sério é que é.

Há uma consequência ainda mais direta de tudo isto, e é talvez a mais cruel de todas, porque não depende de concurso público nem de carreira na função pública. É a forma como se vende a esperança de mobilidade a quem não tem nenhuma mobilidade real ao seu alcance.

O imposto que ninguém chama imposto. A Santa Casa da Misericórdia, instituição com a palavra misericórdia no próprio nome, vende a raspadinha como se fosse uma fronteira em miniatura. Paga dois euros, talvez hoje seja o dia. Quem compra mais raspadinhas tem, tipicamente, menos literacia financeira e menos alternativas. Os mais velhos. Os mais pobres. Gente para quem subir na vida pelo caminho normal deixou de ser uma opção realista há muito tempo, e a quem ninguém ensinou a fazer a conta de quanto se perde a longo prazo para ganhar, de vez em quando, uma fração do que se gastou. O Estado sabe isto. Tem os números todos, melhor do que ninguém. E mesmo assim continua a vender o sonho à porta de quem mais precisava de ser protegido dele, porque é precisamente a quem a mobilidade real fechou a porta que esse sonho vende melhor.

Ao lado disto, a um ritmo bem mais agressivo, estão as apostas desportivas e os casinos online. Betclic, Solverde, Betano, e a fila toda atrás delas. Caras conhecidas da televisão e da internet a vender a marca como se fosse só mais um patrocínio inocente, entre o futebol e o intervalo. Outdoors em todo o lado. Rádio o dia inteiro. Anúncios encavalitados uns nos outros, e no metro para quem ainda tem a sorte de ter metro perto de casa. Código de desconto AtrasadoMental10. Free spins. Cinquenta rodadas grátis com qualquer disparate que se inventou para a campanha estúpida desse mês. As mesmas técnicas de recompensa intermitente que tornam o scroll difícil de parar, só que aqui não se perde tempo. Perde-se dinheiro. E o produto está desenhado por gente que sabe exatamente o que está a fazer.

Começa quase sempre a tentar a única fronteira que ainda parece estar aberta. Depois o desenho do produto faz o resto. A dopamina não distingue entre ganhar a sério e quase ganhar, e o cérebro aprende a perseguir o quase, tanto quanto perseguiria o ganho verdadeiro. Junta-se a isso o instinto mais básico de recuperar o que já se perdeu, a recusa em aceitar que o dinheiro de ontem já não volta, e o que era uma aposta pontual na ocasião torna-se um hábito que continua muito depois de a esperança racional ter morrido. Chamar a isto vício, em muitos casos, não é exagero nem desonestidade. É descrição. A estrutura cria o vício de propósito, e o vício depois garante que a pessoa continua a alimentar uma fronteira que sabe, lá no fundo, que está fechada há muito. Que a casa ganha sempre. Exatamente como nos anos 80 a casa também era sempre quem já estava lá dentro.

A parte cómica, se é que há alguma graça nisto, é o Estado regular o jogo com uma mão e cobrar a fatura com a outra. Há licenças, há impostos especiais sobre o sector, há campanhas oficiais de jogo responsável encaixadas entre dois anúncios da mesma operadora. Ninguém está propriamente a fingir que isto não existe. Está só a gerir-se a si próprio em paralelo, regulador de manhã, sócio à noite, e a receita que entra pela porta das apostas paga uma fatia do mesmo orçamento que devia, em teoria, andar a corrigir o problema que torna essa receita possível em primeiro lugar.

Há ainda uma última consequência, mais difícil de medir mas talvez mais perigosa do que todas as outras: o voto. Quem não vê mobilidade nenhuma no sistema tal como está organizado tende, mais tarde ou mais cedo, a votar contra o sistema, não a favor de uma alternativa concreta dentro dele. Reconhece-se a lógica sem ter de gostar do resultado. Se subir pelas regras normais já não parece possível, votar em quem promete deitar a mesa ao ar passa a ser, para muita gente, a única fronteira que ainda resta. Vota-se na ocasião, no sentido mais antigo da palavra, o mesmo que já se discutiu aqui a propósito de Maquiavel, mais do que em qualquer ideia: quando o sistema normal de subir parece fechado a sete chaves, qualquer fenda que prometa rompê-lo de vez vale a pena tentar, mesmo sem garantia nenhuma do que vem depois.

Há um número que resume isto melhor do que qualquer adjetivo. O Banco de Portugal revelou este ano que o salário mínimo já vale 91% do salário mediano no setor privado, o valor mais alto de toda a zona euro. Quer dizer que quem ganha o mínimo está praticamente ao lado de quem ganha o salário típico do país. Não é que os de baixo tenham subido. É que quase todos estão a ganhar perto do mínimo. Estudar mais, qualificar-se mais, arriscar mais, deixou de se traduzir numa diferença que se sinta no fim do mês. Quando subir e ficar onde já se está dão praticamente o mesmo resultado, a palavra que falta para descrever isso não é igualdade. É estagnação distribuída por igual.

E mesmo para quem ganha esse mínimo, ou pouco mais do que ele, ter um trabalho deixou de ser garantia de ter casa. Não se trata de não querer trabalhar, ou de não conseguir emprego. Trata-se de ter os dois e continuar sem conseguir arrendar uma casa ou até mesmo um quarto nas cidades onde o trabalho existe. É a versão mais literal possível de não haver janela: a porta de entrada na vida adulta costumava ser ter emprego e arranjar casa, e agora é possível ter o primeiro sem nunca conseguir o segundo.

Vale a pena dizer isto sem açúcar a mais nem a menos, porque é fácil de exagerar nos dois sentidos. Há uns anos, com um salário normal, comprava-se uma casa, comprava-se um carro, e ainda sobrava o suficiente para uma vida que se podia chamar normal sem ironia. Isso não significa que tudo fosse melhor. Os salários eram mais baixos em termos absolutos, o crédito era mais difícil de obter, havia menos direitos no trabalho, menos opções de tudo, e quem teve de viver essa época sabe que está longe de ter sido um paraíso. Mas a relação entre o esforço normal e o resultado normal existia. Trabalhava-se, e ao fim de algum tempo havia casa e havia carro. Hoje essa relação partiu-se num dos lados: trabalha-se, às vezes mais e com mais qualificação do que nunca, e o resultado normal já não chega nem para pagar a renda, longe de chegar para comprar qualquer porcaria. O passado não tem de ter sido bom para esta comparação se sustentar. Basta que um passado imperfeito ainda entregasse o básico a quem cumprisse a sua parte, e que o presente, com tudo o que tem de mais avançado, tenha deixado de entregar isso a muita gente que está a cumprir exactamente a mesma parte.

Por baixo de tudo isto está uma classe trabalhadora com salários baixos, que desconta pouco porque ganha pouco, e uma classe média a sustentar o resto através de impostos que sobem sempre que o resto não chega. O IVA, as taxas e as taxinhas, os juros e os jurinhos, as tarifas e as tarifinhas, é tudo a mesma lógica: o Estado vai buscar receita onde consegue, porque qualquer reforma estrutural a sério demora uma geração a dar resultado e nenhum político vive politicamente o tempo suficiente para a ver terminada.

Reformas assim não ganham eleições.

Os filtros que deviam corrigir isto e não corrigem

Se a mobilidade fosse só um problema de sorte de nascimento, haveria menos a fazer. Mas uma sociedade tem filtros precisamente para corrigir essa sorte. Mecanismos cujo trabalho é peneirar quem nasceu em baixo e tem talento para subir, e separá-lo de quem nasceu em cima e não tem nada de especial para oferecer. Quando esses filtros funcionam, a mobilidade sobe, mesmo que lentamente. Em Portugal, os três principais filtros têm vindo a falhar ao mesmo tempo, e isso ajuda a explicar porque é que a mobilidade aqui é tão baixa mesmo quando comparada com economias de tamanho parecido.

O primeiro filtro é a escola, que devia premiar quem se esforça e aprende. Cada vez mais acaba a premiar quem já chega com vantagem de casa. O segundo é o partido político: dificilmente se chega a primeiro-ministro sem passar primeiro por ser escolhido líder de partido e antes disso andar a abanar bandeiras da juventude do partido, e o filtro que devia trazer ao topo quem tem mais para oferecer ao país está, na prática, a escolher por outros critérios. O terceiro é a comunicação social, que devia distinguir competência de barulho e cada vez mais confunde as duas coisas de propósito, porque o barulho vende mais anúncios. Juntos, estes três filtros formam o funil que decide quem sobe e quem fica. Quando os três falham ao mesmo tempo, o que sobe já não é necessariamente quem merece subir mas sim quem sabe navegar três sistemas avariados em simultâneo. É uma competência genuína, sem dúvida, só que não a competência que devíamos querer premiar, e cada vez que isso acontece, a mobilidade que devia existir fica, mais uma vez, por entregar.

Há, a fechar este retrato, uma comparação que eu acho que vale a pena fazer e que ninguém gosta de fazer em voz alta: comparar as oportunidades que existiam para entrar no mercado de trabalho há quarenta anos com as que existem hoje para quem está a começar agora. Quem teve a sorte de viver a fronteira aberta raramente olha para trás com essa régua. É mais confortável lembrar o mérito próprio do que admitir a diferença de ponto de partida. Mas a régua existe, e aplicada com honestidade não deixa grande margem para dúvidas: uma geração inteira apanhou a fronteira aberta, beneficiou de uma mobilidade que os filtros de então, mesmo imperfeitos, ainda conseguiam produzir, e deixou-a fechar-se sem deixar atrás de si nada parecido para quem veio depois.

Há uma frase que se dizia muito, e que ainda se diz, normalmente a quem está a passar por um momento difícil: “Quando se fecha uma porta, abre-se uma janela.”

É uma frase de conforto, e como a maior parte das frases de conforto, funciona melhor quanto menos se olhar para ela de perto. Pressupõe que o mundo tem uma espécie de equilíbrio automático, que nenhuma porta se fecha sem deixar, em troca, outra abertura à mão. Era talvez verdade numa altura em que havia mais janelas do que portas, numa economia a abrir-se, com fronteiras novas a aparecer a um ritmo mais rápido do que as antigas se fechavam. Cada vez que se repete a frase hoje, sem se perceber, está-se a assumir que essa proporção continua a existir. Não existe. As portas continuam a fechar-se ao mesmo ritmo de sempre, ou até mesmo mais depressa. As janelas, essas, estão cada vez mais raras, mais pequenas, e mais bem guardadas por quem já está lá dentro. Continuar a dizer a frase já não é otimismo. É só ainda não ter feito as contas.

Algures num universo onde o número de portas e janelas é constante, alguém ainda acredita nessa frase com toda a razão. Não é aqui. Aqui contam-se as portas, contam-se as janelas, e a diferença não fecha. O universo não devia nada a ninguém em matéria de simetria, e por acaso não deve mesmo nada a este país em particular.

Sem solução, só descrição

Não tenho resposta para isto. E já desisti de fingir que ter percebido o problema ajuda a resolver alguma coisa. Sei exactamente como a fronteira se fechou, sei quem a manteve fechada, sei os mecanismos todos que sustentam o fecho. De nada disto sair da cabeça surge sequer um caminho. Conhecer a anatomia de uma coisa não a torna menos definitiva…

A nostalgia de quem viveu o “tudo parecia possível” não ajuda quem está a tentar entrar agora num país, ou num mercado, ou numa carreira onde já só há fila. Mas também não ajuda escrever isto. Não muda a fila. Não abre porta nenhuma. Fica só a sensação estranha de estar a fazer um diagnóstico rigoroso de uma doença para a qual já não há tratamento, e de continuar a escrevê-lo de qualquer forma, porque a alternativa é o silêncio, e o silêncio também não muda nada.

Fico aqui a fazer as contas. Como toda a gente da minha geração, entre continuar e sair, entre tentar uma fenda pequena ou aceitar que algumas portas só se abrem para quem já as encontrou abertas. Não tenho a ilusão de que isto vai mudar a tempo de me servir a mim. Talvez nem a tempo de servir a quem vem a seguir. Continuo a fazer as contas porque é o que resta fazer, mesmo que não acredite que elas um dia fecham a meu favor.

Subscribe now

A rate decision that was never the story

The market is pricing in roughly a 97% probability that the Fed leaves rates unchanged today. That was never going to be the interesting part of this meeting. Kevin Warsh chairing his first FOMC matters because of what he says, not what he does, and because the plan he is reportedly trying to execute is more elegant and more fragile than a simple rate decision. Understanding the mechanism requires going through three steps, and then understanding why a war on the other side of the world might quietly determine whether any of it works.

The mechanism

Step one is straightforward. Warsh lowers short-term interest rates by cutting the overnight Fed Funds Rate, the one the Fed directly controls, which pulls down front-end yields like the 2-year Treasury along with it. This makes borrowing cheaper at the short end and gives the administration the growth story it wants heading into the midterms.

Step two is where it gets more interesting. At the same time, the Fed shrinks its own balance sheet, meaning it holds fewer bonds itself. Normally this pushes long-term rates up, which sounds bad, but here it is intentional. The combination of low short-term rates and higher long-term rates creates a steeper yield curve. A steeper curve means banks can borrow cheaply at the short end and lend or invest at the long end, pocketing the spread. That spread is how banks make money, and the steeper it is, the more profitable it becomes for them to buy Treasury bonds.

Step three is the part that makes the whole thing work: deregulating the banks specifically around something called the supplemental leverage ratio, or SLR. This is a post-2008 rule that limits how many Treasury bonds banks are allowed to hold relative to their total assets, even though Treasuries are considered close to risk-free. The rule effectively puts a ceiling on how much government debt the banking system can absorb.

During the COVID shock in 2020, the Fed temporarily exempted Treasuries from the SLR calculation for about a year. Banks immediately piled into Treasuries and the bond market functioned smoothly. When the exemption expired in March 2021, banks had to reduce their holdings again and the bond market did not like it. The reported plan now is to make that exemption permanent, or loosen the SLR significantly.

Once the limit comes off, banks can borrow cheaply, buy Treasuries with leverage, and pocket the spread at scale. The Fed shrinks its balance sheet. The commercial banks expand theirs. The net effect on the bond market looks almost identical to the Fed buying the bonds itself, except it never shows up as quantitative easing on the Fed’s books. It is QE laundered through the commercial banking system, with a deregulation press release instead of a money-printing headline.

Worth being precise about something here. Warsh is not a dictator who simply implements this on arrival. He is one vote in a room where Jerome Powell, his predecessor, still sits on the Board of Governors with an active FOMC vote through 2028. The committee is genuinely divided right now. Core inflation ticked back up in May, and the hawks on the committee are pointing directly at that print as a reason to slow down on any of this. What looks like a master plan from the outside is, from inside the room, a negotiation Warsh has to win incrementally rather than a switch he gets to flip.

That was the plan a few months ago, before something happened that the plan did not account for.

What the Iran war did to the spread

The entire mechanism depends on a steep yield curve, short rates low, long rates high. War is inflationary, almost without exception throughout history, because it disrupts supply chains and pushes up energy prices. The Iran conflict shut down meaningful flow through the Strait of Hormuz, which under normal conditions carries roughly a fifth of the world’s oil. Oil went up. Inflation expectations went up. Short-term rates, which move with inflation expectations, went up too, in exactly the direction the plan needed them to go down.

The 2s/10s yield spread is compressed to a razor-thin 40 basis points, against a healthy historical range of roughly 100 to 150 basis points. This is not a modest narrowing from a normal positive slope, it is a curve that spent much of the last two years inverted and has only recently fought its way back to barely positive territory, now getting pinned there by the war.

When that spread compresses, banks have far less incentive to lever up and buy Treasuries, because the profit on the trade shrinks along with it. The entire SLR mechanism depends on that spread being wide enough to make the leverage worthwhile. Without it, the plan does not break loudly. It just quietly stops working.

The two paths available today

Warsh effectively has two options today, and the choice is not entirely his to make freely. The dovish path involves hinting that looser policy is coming, or that the Fed has tools available to support the bond market if conditions worsen. Markets would likely read that as liquidity on the way, and historically gold, equities, and risk assets across the board tend to rise on that signal. The problem is that path only works cleanly if the Iran situation is genuinely resolving, because persistent war-driven inflation makes it very difficult to credibly promise lower rates.

The hawkish path involves acknowledging that inflation remains a serious concern and that the economy, evidenced by a recent jobs report that beat expectations by a wide margin, is strong enough that cutting now would be premature. There is something almost absurd in that dynamic: a strong jobs report is normally good news, and right now the market would treat it as the opposite, because a strong economy with persistent inflation removes the room to cut. In fact, a truly hawkish presentation today would completely strip out the Committee’s long-standing easing bias, signaling to the market that the next move might not be a delayed cut, but an active rate hike if energy shocks keep bleeding into broader price-setting. Markets would likely read this path as tighter conditions for longer, and risk assets typically respond by falling.

There is a third layer worth watching that has nothing to do with dovish or hawkish at all. Warsh has been openly critical of how much the Fed talks. He has argued in the past that forward guidance and the dot plot of rate projections box the committee into corners it later has to walk back from. If that view shows up today, the signal will not be in what he says so much as how little he says. A noticeably shorter statement, fewer specific forward commitments, more deliberate vagueness, that itself would be the story, regardless of which way the rate decision or the tone leans.

The genuinely interesting part of today’s meeting is not the rate decision, which is already priced in. It is which of these stories Warsh tells, how much he chooses to say at all, and how the market reacts in the hours afterward, particularly in the dollar index. A weakening dollar after the press conference would signal markets pricing in future liquidity. A strengthening dollar would signal the opposite: tighter conditions, higher rates for longer, more pain ahead.

There is also a structural irony sitting underneath all of this. Warsh has written that he believes AI represents a genuine disinflationary force, a productivity story closer to the 1990s than to anything happening in the Middle East right now. He is, in effect, looking through the windshield at a structural productivity boom that should bring prices down over time, while the rest of the committee is looking out the side window at a literal gas station, watching the price of oil decide the next several months of policy. Both views are defensible. They are also pulling in opposite directions on the same decision.

The two clocks running underneath

There are two separate countdowns that make this more urgent than a single Fed meeting suggests. The first is the US Strategic Petroleum Reserve, the emergency oil stockpile that has been absorbing the shock of the Hormuz disruption and keeping domestic gas prices from rising further than they already have. According to reporting at the time, the reserve was estimated to run out in under 80 days. Once it does, there is no remaining cushion, and oil prices move directly with whatever is happening at the strait.

The second clock is the US Treasury’s refinancing need: roughly $8 trillion in debt that needs to be sold or refinanced over the next 12 months, into a market where several of the usual buyers have become less reliable. Foreign central banks have been replacing a portion of their Treasury holdings with gold rather than adding more dollar-denominated debt. The buyers stepping in to fill that gap are increasingly leveraged hedge funds, which is a structurally fragile substitute for what used to be sovereign, relatively price-insensitive demand. None of this resolves on its own. It resolves if Hormuz reopens in time, or it does not.

What Iran is actually doing

It is worth being precise about the strategic logic here, because it is easy to misread the conflict as primarily military. Iran does not need to defeat the United States militarily, and by most assessments cannot. The leverage is different. If keeping the strait constrained for long enough pushes oil high enough, for long enough, the inflation feeds directly into the spread compression described above, and the US bond market bears the cost regardless of what happens on any single battlefield.

This is the framing I have used before in this series: Hormuz functions as a trigger that reveals pre-existing vulnerability rather than a cause of one. The vulnerability, an $8 trillion refinancing need landing in a market with a shrinking buyer base, predates the war entirely. The war just determines the timing.

Subscribe now

The rules changed

In the next few weeks, if you hold any broad market index fund tracking the Nasdaq 100, the Russell 1000, or any other major technology index, you are going to become a buyer of SpaceX. The rules of index inclusion make it automatic. And right behind SpaceX are OpenAI and Anthropic, both preparing their own public listings. Three companies with a combined valuation of roughly $4 trillion, all going public within months of each other, all designed to plug directly into the passive investment flows that your index fund feeds into every time you save. This is not an American story. Passive investing is now the dominant form of retail investment globally. Anyone holding a Nasdaq-tracking ETF in Lisbon, London, or Tokyo is part of this.

The rules that made this possible were changed on May 1st, weeks before the biggest IPO in history. Nasdaq introduced a fast entry rule cutting the inclusion window from up to a year down to 15 trading days. The float requirement that would have disqualified SpaceX was waived. A hidden multiplier was added: for companies listing with a float below 20%, the index treats the float as three times larger than it actually is. A 4% float gets weighted as if it were 12%. Index funds are now legally required to buy more than the available supply would normally justify. The FTSE Russell moved to 5-day inclusion. The S&P held its line on profitability, which is why SpaceX cannot enter the S&P 500 until June 2027 at the earliest. But the Nasdaq and Russell passive flows still represent billions in forced demand. And SpaceX is allocating 30% of the IPO to retail investors, three times the standard 10%. The venture capital investors who got in years ago at much lower valuations need to exit. Someone has to buy their shares.

Not a valuation bubble, but an earnings bubble

Most people looking at the AI trade right now conclude it does not look like a bubble. The price-to-earnings ratios of the major technology companies are not insane. The valuations look kind of reasonable. The companies are profitable. This is not 1999. So the argument falls apart before it even gets started. The problem is that everyone is making the wrong argument. The observation worth making, from BCA Research among others, is that this is not a valuation bubble. It is an earnings bubble. Those are different things and the difference matters enormously because earnings bubbles are much harder to see coming.

In a normal bubble, the stock price goes up while the earnings stay flat, so the price-to-earnings ratio inflates visibly and obviously. In an earnings bubble, the earnings themselves are being inflated. The price-to-earnings ratio stays low and looks healthy because the E is being manufactured. The ratio looks fine right up until the moment the earnings are revealed to be built on something other than what they appeared to be. This is what happened with homebuilders before 2008. Perfectly healthy earnings numbers right up until the underlying assets were shown to be worth a fraction of their book value. The same thing happened with banks. The theory is that it is happening now with AI, through a mechanism that is genuinely elegant in how it works and genuinely alarming once you see it.

The circle

Big technology companies, Microsoft, Google, Amazon, Meta, invest billions into AI startups. Anthropic, OpenAI, and others. The AI startups then use that capital to rent computing power back from the same big technology companies. Big tech books the investment gains from those AI startups as profit on their balance sheets. Those paper profits justify lending and spending even more. The money is mostly going in a circle. OpenAI and Anthropic spending commitments together account for between 43% and 54% of the entire revenue backlogs of Microsoft, Oracle, and Amazon, according to multiple analyst estimates. The companies that are supposedly benefiting from AI demand are also the ones funding the AI companies that are generating that demand. The circularity is not hidden. It is visible in the financial statements if you read them carefully enough. It just requires reading them together rather than individually.

The Financial Times looked at what Microsoft, Google, Amazon, Meta, and Oracle are planning to spend on AI infrastructure between now and 2030, then asked how much revenue those same companies are expected to generate from those investments over the same period. Under the most generous assumptions possible, assuming zero costs, no salaries, no electricity, no overhead, Microsoft’s implied return was negative 9%. Google was negative 15.7%. Meta negative 28.8%. Oracle negative 35.6%. The only company clearing positive territory is Amazon at 7.2%. And remember, these are not worst case numbers. These are best case. Assuming it costs nothing to build and run the infrastructure, almost all of them still lose money on the AI investment. Those same companies issued over $120 billion in corporate bonds to fund the AI spending in 2025, more than four times their annual average from 2020 to 2024. More than double what they were doing two years ago. Their free cash flow margins, the actual money left over after spending, are collapsing toward zero when you account for the AI capital expenditure.

The cost of the demo

The circular spending shows up at the corporate level too, and in a way that is easier to see. The CTO of Uber recently ran a two-hour internal demo to show his executive team how AI would save the company money. The demo cost $1,200 just in tokens. The act of demonstrating the cost savings cost more than the thing it was supposed to replace. Uber had set an annual AI budget for 2026. They blew it in four months. The output was some summarized emails and a handful of internal specs that nobody read. They were running an internal leaderboard ranking teams by how many tokens they burned, giving standing ovations to the biggest spenders. They eventually flipped it. Now, whoever spends the least wins. Microsoft is cancelling its own cloud licenses for cost reasons. GitHub Copilot moved customers to token-based billing and some saw their costs jump a HUNDRED TIMES over, not a hundred percent ( that is 10,000% for those of you keeping score). OpenAI loses approximately $1.22 for every dollar it makes. The technology is being adopted at scale before anyone has solved the economics of running it. The companies burning the most capital on AI infrastructure are being rewarded with the highest valuations. That is the same incentive structure as the leaderboard, just at a larger scale.

Why the IPOs need to happen now

The circle only keeps spinning as long as everyone agrees about the valuations. The private funding rounds set the valuation. The investment gains get booked as profit. The profit justifies the stock prices. The stock prices justify more spending. But the second these AI companies go public, the valuations stop being whatever a private funding round between friendly parties decided they were. They become whatever the open market says they are worth on the day of trading. If the open market prices them lower than the private round valuations, every profit that was booked by Google and Amazon and Microsoft against those private valuations has to be revised downward. The earnings that justified the stock prices get revised. The loop runs in reverse. This is the most dangerous moment for the AI bubble so far. For the last few years, the trade has mostly been about belief. Existing AI-linked stocks went up. Private companies raised money at enormous valuations. Nobody had to actually convert confidence into cash at scale. The IPOs are the moment that changes. Confidence has to become real money. And not everyone who said they would buy the car actually buys the car when they are standing at the dealership.

Notice that Google did not wait for SpaceX to price. They went first, raising $85 billion before the IPO queue opened. That is not coincidence. If Google absorbs $85 billion of AI investor demand, that money cannot go to SpaceX or Anthropic or OpenAI. It is a race to find the liquidity before it runs out. The person who bought Anthropic stock three years ago and is now sitting on a position worth many multiples of their original investment wants to sell. The private market investor who got in early needs an exit. Someone has to buy all of it at these prices. That someone is the open market. And the open market has finite cash.

The railroad argument

But SpaceX is no longer just a rocket company. Three businesses under one ticker: the launch services business, Starlink, and xAI with X, the loss-making AI and social media operation that turned a $791 million net income in 2024 into a $4.94 billion net loss in 2025. The valuation is priced on that third business. The S1 claims a $26.5 trillion AI addressable market. 93% of the valuation justification comes from a segment with no current revenue.

In the late 1800s, railroads changed America forever. They connected the whole country, expanded commerce at a scale that was impossible before, transformed agriculture, manufacturing, supply chains. The technology was real and the railroads are still here today. Almost every single investor who funded the railroad boom got wiped out. The companies went bankrupt and the stocks collapsed. A new group of investors came in, bought the same railroads out of bankruptcy at pennies on the dollar, and made fortunes. The same pattern played out with the fiber optic cable boom of the 1990s. Hundreds of billions spent laying cable across the country and under the ocean. The companies went bankrupt in the dot-com crash. A decade later, that same fiber optic infrastructure became the backbone of the internet. Technology won but the initial investors lost. The people who bought those assets afterward built the world we live in today.

The technology works. Starlink has 10 million subscribers and 63% margins. The reusable rocket changed the economics of getting to orbit in a way that will not be reversed. None of that is in dispute. What history has a clear record on is the gap between a technology succeeding and the people who funded it at peak valuations doing the same. Those are two completely separate outcomes and they rarely travel together.

What this means right now

The S&P 500 has just hit four consecutive record highs on negative market breadth. The index keeps going up but more stocks are going down than going up. The gains are being driven by a handful of the very biggest companies while everything else deteriorates. AI-related stocks now make up approximately 45% of the entire S&P 500’s market cap. 41 stocks out of 500. If anything goes wrong with AI, it will not just hurt tech stocks. It will hurt anyone holding a broad index fund or pension invested in US equities, because almost half of the major index is now a bet on AI. And those same companies are about to be joined by the three biggest IPOs in history, all competing for the same pool of capital, into the same liquidity squeeze the bond market has been documenting for months.

The US personal savings rate just hit 2.6%, the lowest in four years. Real wages are flat. The top 10% of earners are the only reason the economy is not in a recession (or is it? it really depends on how you measure it). The passive investment money that SpaceX and OpenAI and Anthropic are counting on to buy their stock only exists as long as workers globally have paychecks to invest every month. The moment AI displaces those paychecks at scale, the very thing these companies are exploiting starts to get smaller. There is less money for them because they got rid of their own workers. Which is another reason why the window for these IPOs may be closing and why the people on the inside want to move now.

If the AI bubble bursts, the index funds collapse, the pension savings evaporate, and the spending that has been holding up corporate earnings goes with it. If AI keeps dominating and the concentration continues, nearly half of every broad index fund on earth is a single-thesis bet on a technology that has not yet demonstrated it can pay for itself. Either way, the passive investor did not ask for any of this. The rules were changed, the allocations were set, and the car was bought on their behalf.

Once the whole world is in the bubble, that’s just called atmosphere.

Subscribe now

I have been covering the SpaceX IPO across several posts in this series, starting before the S1 was public.

What you need to know about the SpaceX IPO

SpaceX IPO update

The xAI complication that i identified early turned out to be more damaging than it appeared. This post is about what the S1 says when you read past the executive summary, and about a detail regarding the index inclusion rules.

The filing is public at SEC EDGAR

Three businesses under one ticker

SpaceX prices its IPO around June 11th. The S1 is around 300 pages and i am sure most people will not read it and some will ask their favourite chat bot to sum it up and decide if they go long or short on the IPO. The number that matters most requires reading past the executive summary into the historical comparatives, and i think it changes the picture quite a bit. Before getting to that number, it helps to understand what is actually being sold. The S1 describes three distinct businesses packaged together under one ticker:

• Launch Services: Falcon 9 and Falcon Heavy launch services. The reusable rockets changed the economics of getting to orbit in a way that will not be reversed. The business is growing, it is defensible, and remarkable in a way that almost no technology business is.

• Starlink: 10.3 million subscribers, $11.4 billion in revenue in 2025, functioning infrastructure that is genuinely difficult to replicate.

• xAI and X: The social media and AI infrastructure business that merged into SpaceX in 2025.

The first two businesses are palpable and the competitive advantages are there. The third is where the valuation friction lives.

The number buried in the comparatives

In 2024, before the xAI merger closed, SpaceX posted $791 million in net income. Profitable. Growing. A genuine aerospace business.

In 2025, after the merger, SpaceX posted a $4.94 billion net loss. In Q1 2026 alone, the loss was $4.28 billion in a single quarter. The accumulated deficit now stands at $41.3 billion.

The xAI merger did not add a profitable AI business to a profitable rocket company. It added a loss-making AI and social media liability to a profitable aerospace balance sheet, using the word AI to reframe a legacy industrial moat into a tech-multiplier valuation. AI losses account for approximately $2.5 billion of that quarterly burn. X, the social media business that came with the merger, accounts for most of the remainder. Neither is tapering.

That is not R&D spending that will taper. AI infrastructure at the scale xAI is building requires sustained capital expenditure that compounds in the near term. The S1 calls this a strategic investment phase. It does not model when that phase ends or at what revenue level the losses taper.

Will it pay off in the end? I simply can’t tell…

Starlink

Starlink’s subscriber growth is genuinely impressive. From 5 million in Q1 2025 to 10.3 million in Q1 2026, more than doubling in a year. However the average revenue per subscriber fell 23% year over year.

Revenue is growing because there are more subscribers. The revenue per subscriber is declining as Starlink pushes into lower-priced international and consumer markets to hit volume targets. The Starlink bull case is built on the assumption that a large subscriber base at healthy margins produces durable cash flow. The actual trend in the document is a large subscriber base at shrinking margins. Those are different businesses with different long-term implications, and the executive summary features only one of them.

The valuation

The S1 breaks the total addressable market into three segments:

• Space: $370 billion

• Connectivity: $1.6 trillion

• AI: $26.5 trillion (Enterprise applications within AI alone is listed at $22.7 trillion)

The businesses generating actual revenue today, launch services and Starlink combined, represent under 7% of the total TAM being used to justify the valuation. The other 93% is a placeholder for the AI segment.

The S1 states a $22.7 trillion enterprise TAM but entirely omits the mechanism by which SpaceX captures the market share from entrenched incumbents. You are being asked to pay 2026 prices for a market position that does not yet exist, against competitors who have been building in that market for years.

The governance problem ( is it a problem? )

Elon Musk holds approximately 42% of SpaceX equity and approximately 85% of voting power. Every shareholder who is not Elon Musk controls 15% of the votes collectively in a company priced at up to $2 trillion.

The S1 also includes an arbitration clause and class action waiver. Shareholders who have a grievance cannot bring a coordinated legal action. They must pursue individual arbitration. The combination means that no mechanism exists for shareholders to hold management accountable if capital allocation decisions destroy value.

This is disclosed in the risk factors section. It is presented as a feature of how the company operates, not as a risk to investor returns.

In fairness, this is not a straightforward negative. Companies where the founder retains control have a documented tendency to outperform over long time horizons. Bezos, Zuckerberg, Page and Brin all ran dual-class structures. The argument is that insulation from quarterly shareholder pressure allows better long-term capital allocation. And this is not a random founder. Musk built Tesla from near bankruptcy to the most valuable car company in the world. Is it even a car company? He built SpaceX from scratch into the dominant launch provider on the planet. The track record is genuinely remarkable.

In earlier posts in this series I flagged the possibility of a Tesla and SpaceX consolidation into a single Musk empire. If that happens, the voting structure you are buying into today does not just apply to SpaceX. It applies to the combined entity. One person with 85% of the votes across a $2 trillion aerospace and AI company that may eventually absorb the world's largest electric vehicle manufacturer. That is a level of consolidated control over a corporate empire that has no real precedent in modern markets. Whether that is reassuring or alarming probably depends on how you feel about Musk specifically. Which is itself a risk worth naming.

Who is buying this

Standard practice for a mega-cap IPO is to allocate approximately 10% of the float to retail investors. The SpaceX IPO is allocating 30% to retail, three times the standard. When venture capital and private investors who have held SpaceX equity for years at much lower valuations are exiting, someone has to act as the liquidity provider. At three times the normal allocation, that someone will be retail.

There is a further detail that most coverage missed until Thursday evening. On June 4th, S&P Dow Jones Indices held the line and refused to grant SpaceX a fast-track exemption into the S&P 500.

Their statement was explicit:

“Exceptions to the financial viability, seasoning, and investable weight factor (IWF) requirements should not be granted solely based on market capitalization.”

SpaceX fails on all three counts:

• The Seasoning Rule: Requires 12 months of public trading, meaning the earliest possible S&P 500 entry is June 2027.

• The Profitability Rule: Requires four consecutive quarters of cumulative GAAP profit and a profit in the most recent quarter. The $4.94 billion net loss in 2025 and the $4.28 billion Q1 2026 loss lock SpaceX out indefinitely until the xAI and X cash burn stops.

• The Float Rule: Requires at least a 10% public free float. Because of Musk’s heavy inside ownership and the tight allocation structure, initial calculations put the public float tracking at just 3% to 4%.

Nasdaq and FTSE Russell did change their rules. Nasdaq cut its inclusion window to 15 trading days for qualifying megacaps. FTSE Russell will allow inclusion roughly 5 trading days after IPO. Those two indexes will absorb a combined 24% of the float through passive funds.

But the S&P 500 passive cushion, which Bloomberg Intelligence had estimated at 19% of the float, has completely evaporated. That 19% does not disappear. It falls entirely on retail and active buyers who choose to participate. The argument that passive index flows would underwrite the valuation at launch no longer holds. The rules built since 2002 to protect passive investors from buying loss-making companies at IPO prices held for the most important index in the world.

The honest position

SpaceX is a great company. The launch services moat is very defensible, the Starlink infrastructure is genuinely difficult to replicate and the long-term vision of a vertically integrated space and AI company is coherent, even if the execution is loss-making right now.

The xAI merger is probably not a terrible idea in theory. xAI brought the AI models and the software ambition. SpaceX brought the launch infrastructure, the Starlink backhaul, and the balance sheet. The logic of combining them is coherent. The problem is that xAI is burning approximately $2.5 billion per quarter building the infrastructure to make that vision real, and X, the social media business that came with the merger, is also loss-making. The Q1 2026 total loss of $4.28 billion reflects both. Whether $1.75 to $2 trillion is the right price for what is being sold today, under this governance structure, at this stage of the xAI integration, in this liquidity environment, is the part the S1 cannot answer for you.

The Google Cloud Contract Reality

One development worth acknowledging. On June 5th, SpaceX filed an amendment to the S1 disclosing a Cloud Service Agreement with Google. Google has agreed to pay $920 million per month from October 2026 through June 2029 for access to approximately 110,000 NVIDIA GPUs and related compute capacity. That is roughly $11 billion per year in contracted revenue from a single customer. It partially changes the timeline to profitability for the xAI infrastructure business and the bull case is stronger for it.

The termination clause is the caveat worth reading. After December 31, 2026, either party can exit the agreement with a 90-day notice. That is a tight six-month committed period at full rate before the contract becomes effectively cancellable on a rolling basis. Google retains ownership of its AI models and data. This is real contracted revenue, but it is not the durable, structural long-term anchor that the headline numbers imply.

One more thing worth noting for anyone holding Tesla. In an earlier post in this series I flagged that Tesla’s valuation has been partly sustained by its association with Musk’s broader empire, and that a SpaceX IPO would test that relationship. If SpaceX trades at a significant premium to its private valuation, the pressure to eventually consolidate the Musk empire into a single entity grows. If it disappoints at IPO, the dynamic runs the other way. Either outcome has implications for Tesla holders.

SpaceX is going public the same week that Google completes the largest follow-on offering in the stock market history and Anthropic files its S1 confidentially. All of them are competing for the same pool of capital at the same time. That is not the ideal moment to be buying at the top of the range.

Subscribe now

Há uma estátua no centro de Lisboa que me irrita sempre que a vejo. Felizmente não sou de Lisboa e visito muito poucas vezes a cidade.

O Marquês de Pombal, de braço estendido, olhar sobranceiro, como se estivesse a apontar o caminho para o futuro. É uma das estátuas mais icónicas da capital. Ensinado nas escolas como um dos maiores estadistas da história portuguesa. O homem que reconstruiu Lisboa após o terramoto de 1755. O iluminista que modernizou o país. O herói.

Disseram-nos isto nas escolas. Repetem-no nos livros. Está gravado em pedra no centro da capital.

O problema é que a história que nos contaram está errada. Não está apenas incompleta. Errada mesmo.

As consequências dessa história mal contada ainda hoje se sentem na forma como o país é gerido, na relação que os portugueses têm com o Estado, na dependência crónica de um aparelho que nunca funcionou bem, e na fraqueza persistente das nossas instituições.

Pombal não foi o herói que nos ensinaram. Foi o ponto de partida de um efeito borboleta que ainda não terminou. E enquanto continuarmos a erguer estátuas aos responsáveis pela nossa destruição, não teremos razão nenhuma para mudar.

Ainda mais irónico a estátua ficar ao lado da Avenida da Liberdade.

Antes de Pombal: o país que poucos conhecem

Para perceber o que Pombal destruiu, é preciso primeiro perceber o que existia antes dele. E aqui começa o primeiro problema, porque o que existia antes dele não tem nada a ver com o que nos ensinaram.

A narrativa oficial diz que Portugal antes do século XVIII era uma monarquia absoluta, em que o rei mandava em tudo e todos obedeciam sem questionar. Essa narrativa foi inventada no século XIX pelos chamados liberais, que precisavam de pintar o passado como absolutamente negro para justificar a sua própria existência, mais ou menos o que os partidos da oposição fazem sempre, não interessa qual o partido. Propaganda política com duzentos anos de vida.

Portugal tinha parlamentos. Chamavam-se Cortes, e existiam desde a Idade Média. Os municípios portugueses, todos os duzentos e cinquenta e tal, elegiam representantes chamados Procuradores que iam a Lisboa negociar com o rei. O rei não podia subir impostos sem autorização das Cortes. Não podia tomar decisões políticas relevantes sem as consultar. Havia freios e contrapesos (checks and balances) ao poder real.

Votavam nestas eleições locais pessoas comuns. Sapateiros. Artesãos. Agricultores. Sufrágio universal não era, mas era uma proto-democracia com legitimidade representativa genuína. Há registos de eleições ganhas por um voto, de Procuradores que perderam por margem mínima. Isto era o século XVII em Portugal.

Depois da Restauração de 1640, as Cortes voltaram a ter muito poder. O rei não fazia o que lhe apetecia. Havia negociação constante. Portugal estava numa trajectória política que, em termos comparados europeus, não era de todo má.

As mulheres portuguesas neste mesmo período trabalhavam no mercado de trabalho em proporções comparáveis às do norte da Europa. Podiam até ser cabeças de casal. Os seus salários relativos aos dos homens eram semelhantes aos da Holanda da altura. A ideia de que as mulheres no sul da Europa eram muito mais discriminadas do que no norte é um mito construído por historiadores holandeses do século XX que extrapolaram para trás uma realidade do século XIX.

Os níveis de literacia e numeracia de Portugal no início do século XVIII eram comparáveis aos dos países mais avançados da Europa. Medem-se estas coisas de formas criativas: um indicador clássico é a percentagem de pessoas que, quando perguntadas pela idade, respondem com um número preciso em vez de arredondar para a dezena. Sociedades mais letradas respondem com precisão, 33, 42, 67, 21. As outras arredondam para as dezenas, 30, 40, 60, e é possível ver uma distribuição anormal das idades da população em termos estatísticos. Portugal no início do século XVIII estava ao nível europeu normal.

O sistema educativo existia e funcionava. Os jesuítas tinham praticamente o monopólio do ensino pré-universitário em Portugal. O Colégio de Santo Antão em Lisboa tinha entre dois mil e quinhentos e três mil alunos e era gratuito. Não havia três mil nobres em Lisboa, o que significa que pessoas de todos os estratos sociais iam ali aprender. A Universidade de Évora ensinava física avançada para os padrões da época. Havia uma rede real, funcional, e acessível.

Este é o Portugal que Pombal herdou.

Mas havia um problema crescente que vai ser o contexto que permite a Pombal fazer o que faz. O ouro do Brasil.

O ouro que nos envenenou

A partir de finais do século XVII e especialmente durante o século XVIII, o ouro do Brasil chegou a Portugal em quantidades enormes. O efeito foi destrutivo.

Quando um rei tem dinheiro próprio, os parlamentos tornam-se irrelevantes, porque deixa de ser necessário negociar. Em todo o século XVIII, as Cortes não reuniram uma única vez. O absolutismo português não tem raízes no século XVI como nos ensinaram. Tem raízes no ouro do Brasil, que deu ao rei independência financeira suficiente para ignorar o único mecanismo de controlo que existia.

O efeito económico foi igualmente grave. Quando entra muito dinheiro externo numa economia, as indústrias exportadoras perdem competitividade. Torna-se mais barato importar e mais caro exportar. Portugal em finais do século XVII tinha um processo de industrialização em curso, estava até a exportar têxteis para Castela. Tudo isso foi destruído. A economia desindustrializou. O país ruralizou. Em meados do século XIX, dois carregamentos de Vinho do Porto a caminho de Inglaterra valiam mais do que todas as fábricas portuguesas juntas.

O Tratado de Methuen de 1703, que facilitava a entrada de têxteis ingleses em Portugal em troca do acesso preferencial do Vinho do Porto a Inglaterra, já tinha comprometido a indústria têxtil nacional. O ouro do Brasil financiou esse défice comercial durante décadas, tornando-o suportável mas nunca resolvido. Pombal herdou uma economia já distorcida. O que destruiu foi o único activo que poderia ter ajudado a sair dessa distorção.

É a maldição dos recursos. Aconteceu com a prata na Espanha no século XVI. Aconteceu com o petróleo em países do Médio Oriente no século XX. A abundância de dinheiro fácil destrói os incentivos para construir, produzir, e competir.

A dimensão política foi ainda mais grave. O ouro do Brasil tornou as reformas institucionais desnecessárias no curto prazo. Para quê mudar quando o dinheiro chega? Para quê negociar com parlamentos quando não precisas deles?

É neste contexto que o Pombal aparece.

O terramoto e o mínimo aceitável

Em Novembro de 1755, um dos maiores terramotos da história europeia destruiu Lisboa. Seguiram-se incêndios e um tsunami. Dezenas de milhares de mortos. A cidade em ruínas.

Pombal reconstruiu-a. É por isso que tem uma estátua, dizem eles.

Reconstruir a capital após uma catástrofe natural é o mínimo aceitável para qualquer pessoa que detenha o poder e a responsabilidade de o fazer. Se amanhã Lisboa fosse destruída por um terramoto, esperávamos que o governo a reconstruísse. Não lhes erigiríamos estátuas por isso, ou talvez sim, porque somos um bocado provincianos.

Apesar de ser sempre uma tragédia humana para quem a sente na pele, o impacto no PIB do país foi temporário e relativamente limitado. Foi antes de mais uma crise de Lisboa, não uma crise de Portugal. O país já estava mal muito antes. Já estava a divergir da Europa. O terramoto não causou o atraso, apareceu quando o atraso já existia mas faziam de conta que não.

O que o terramoto fez foi dar a Pombal exactamente o que ele precisava: caos suficiente para centralizar poder. Antes do terramoto, era um dos vários secretários de estado do rei D. José. Depois, aproveitou a crise para se tornar o homem indispensável, afastar os outros do centro de decisão, e concentrar nas suas mãos um poder que não poderia ter obtido em circunstâncias normais.

A narrativa do herói do terramoto foi construída em grande parte pelo próprio Pombal. Encomendada, financiada, e espalhada por ele. Camadas de propaganda que as gerações seguintes foram reciclando, cada uma por razões próprias. A Maçonaria no final do século XIX financiou a estátua que ainda hoje está no centro de Lisboa, precisamente porque a visão antirreligiosa e centralizadora de Pombal servia os seus interesses. A Primeira República adorava Pombal pelo mesmo motivo. O Estado Novo encontrou nele a figura do reformador autoritário centralizador com quem se identificava.

O que ele realmente destruiu

Em 1759, Pombal expulsou os jesuítas de Portugal. Foi o primeiro país da Europa a fazê-lo.

Os motivos não foram religiosos nem filosóficos. Os jesuítas opunham-se por princípio ao poder absoluto, o que os tornava inimigos naturais de alguém que queria exactamente poder absoluto sem contestação. O irmão de Pombal era governador no Brasil e estava envolvido em negociatas das quais o próprio Pombal beneficiava financeiramente. Os jesuítas no Brasil protegiam as populações nativas e entravam em conflito constante com esses interesses.

Havia também a questão política mais imediata. Em 1759, houve uma suposta tentativa de assassinato do rei D. José. As circunstâncias nunca foram completamente esclarecidas e a probabilidade de ter sido encomendada pelo próprio Pombal é, no mínimo, elevada. A partir desse pretexto, Pombal eliminou fisicamente os seus inimigos: execuções públicas com tortura, derrube do palácio dos Távora, terra salgada onde o palácio esteve. Onde hoje é o Rato.

Meteu um irmão à frente da Inquisição para a usar como instrumento de perseguição política. Nepotismo puro. Poder familiar centralizado disfarçado de modernização iluminista.

Mas nada disto foi tão destrutivo como o que fez à educação.

Pombal fechou todas as escolas dos jesuítas. Fechou a Universidade de Évora, ficando apenas a Universidade de Coimbra que ficou quase sem alunos. Expropriou todas as propriedades e recursos financeiros da Companhia de Jesus em Portugal. Prometeu usar esse dinheiro para construir um sistema educativo alternativo, mas nunca o fez.

O dinheiro desapareceu.

Pode argumentar-se que o ensino jesuíta era demasiado focado na teologia e no latim, que perpetuava dogmas em vez de promover o pensamento crítico. É um argumento legítimo. Mas a questão não é o conteúdo ideológico do que os jesuítas ensinavam. É a infraestrutura que garantiam. Uma população que sabe ler, escrever e contar, mesmo com manuais religiosos, está infinitamente melhor equipada do que uma população 80% analfabeta. Pombal destruiu a rede. O que ia dentro da rede era um problema secundário.

O Colégio de Santo Antão, com os seus dois mil e quinhentos a três mil alunos gratuitos, foi substituído pelo Colégio dos Nobres, que como o nome diz, só aceitava nobres e nunca chegou a ter cem alunos. Substituíram três mil por menos de cem.

A reforma da Universidade de Coimbra é frequentemente citada como contraprova. Pombal modernizou-a, introduziu as ciências exactas, criou o gabinete de física e o jardim botânico, limpou o escolasticismo. É verdade. Mas modernizar o topo enquanto destroís a base é o equivalente a construir o telhado e deitar as fundações abaixo. Coimbra ficou moderna e quase sem alunos, porque não havia escolas preparatórias para criar candidatos à universidade.

Portugal ficou com um sistema educativo gravemente amputado.

Existiam outras escolas, municipais, de outras ordens religiosas, privadas. Mas o que os jesuítas tinham construído era qualitativamente diferente: a maior rede de ensino gratuito e acessível a todas as classes sociais que o país alguma vez teve. O Colégio de Santo Antão não foi substituído por nada remotamente equivalente. A Universidade de Évora ficou fechada mais de duzentos anos. O dinheiro expropriado desapareceu.

Nos 140 anos que se seguiram, houve tentativas de reconstruir. O liberalismo do século XIX aprovou leis de escolaridade obrigatória. A Primeira República lançou campanhas de alfabetização. Todas falharam por razões que já discutimos, falta de recursos, oposição cultural, motivações políticas contraditórias. Mas há uma pergunta que fica no ar: teriam falhado da mesma forma se tivessem herdado um país com uma rede educativa funcional em vez de um país que levava já um século a tentar reconstruir o que tinha sido destruído?

Não é possível saber. O que é possível dizer é que o défice criado em 1759 provou ser demasiado profundo para ser recuperado por qualquer dos regimes seguintes durante mais de um século. Pombal não causou sozinho o analfabetismo do século XX. Criou as condições que tornaram esse analfabetismo muito mais difícil de evitar.

Este tipo de decisão faz-me lembrar o que acontece hoje com medidas políticas tomadas de forma impulsiva para benefício próprio rápido sem pensar nas consequências a médio e longo prazo. Quem vier que se amanhe.

De jure e de facto

Antes de rastrear o efeito borboleta que se seguiu, há um conceito que é preciso fixar.

Existe uma diferença fundamental entre o que é a lei escrita, o de jure, e o que é a aplicação efectiva da lei, o de facto. Em Portugal a divergência entre as duas tem sido, ao longo de séculos, sistematicamente enorme.

A União Soviética tinha uma das constituições mais bonitas do mundo. No papel, garantia direitos a toda a gente. Na prática, era um Estado policial que assassinava os seus próprios cidadãos.

Em Portugal, esta divergência é um padrão. Um padrão que explica porque tantas reformas, tantos liberalismos, tantas constituições, tantos pacotes de modernização não mudaram nada de fundamental.

A escolaridade era obrigatória em Portugal desde os anos 40 do século XIX. Estava escrito. Mas na prática, absolutamente ninguém ia à escola, não havia escolas suficientes, e não havia nenhum esforço efectivo do Estado para fazer cumprir a lei. Durante décadas.

O efeito borboleta

Uma borboleta bate as asas em Lisboa em 1759, e 150 anos depois três quartos dos portugueses são analfabetos.

No início do século XX, Portugal era o país mais atrasado em educação de toda a Europa Ocidental, de longe. Um abismo que não existia no início do século XVIII. O que aconteceu nesse intervalo tem um nome.

A cadeia causal é directa e documentada. Sem escolas, não há literacia. Sem literacia, não há operários qualificados. Sem operários qualificados, não há industrialização. Há registos de finais do século XIX, de donos de fábricas portuguesas, a dizer que não conseguem pôr as suas fábricas a funcionar porque o país é de analfabetos.

O século XIX foi um século perdido para Portugal. A Inglaterra, os Estados Unidos, e a Alemanha cresciam a ritmos nunca antes vistos mas ao mesmo tempo Portugal divergia. Cada ano que passava, o fosso aumentava.

Houve investimento. O fontismo da segunda metade do século XIX construiu caminhos de ferro, estradas, pontes. Capital físico sem capital humano para o acompanhar. Dois carregamentos de Vinho do Porto a caminho de Inglaterra valiam mais do que todas as fábricas portuguesas juntas.

Portugal teve revoluções, liberalismos, constituições, parlamentos reabertos, retórica de modernização. De jure, havia progresso. De facto, o país real não mudava, porque o país real não tinha o capital humano para absorver e exigir essas mudanças. Uma sociedade com 80% de analfabetos não tem sociedade civil. Sem sociedade civil, as leis ficam no papel e os políticos fazem o que querem porque ninguém consegue fiscalizar nada.

A Primeira República, em 1910, chegou ao poder com vontade declarada de acabar com o analfabetismo. Tinha vergonha do atraso e queria genuinamente mudar as coisas. Mas o que fez foi repetir os erros de Pombal: destruir sem construir, impor sem consultar, mudar a lei sem mudar a realidade.

Expulsou os jesuítas novamente. Proibiu símbolos religiosos nas escolas. Misturou meninos e meninas nas mesmas salas. Mudou os programas para servir a sua agenda jacobina. Os pais, na esmagadora maioria rurais e católicos, simplesmente não mandavam os filhos para a escola porque não se identificavam.

Havia também um motivo político mais sórdido. A República tinha medo de alfabetizar demasiado rápido a população rural porque sabia que essa população votaria contra ela, como hoje. Construíam escolas nas cidades onde tinham apoio, não na província onde precisavam de as construir.

O analfabetismo infantil só foi resolvido durante o Estado Novo, por uma razão que desconforta muita gente.

Não estou a defender o Estado Novo, longe disso. Foi uma miséria moral. Um regime asfixiante com censura, polícia política, perseguições, tortura, tribunais plenários, e uma guerra colonial sangrenta que hipotecou gerações inteiras. Isso não se discute.

O que se discute é outra coisa. Em termos estritos de capital humano básico, o regime percebeu o que a República e o Liberalismo falharam por completo: para resolver o problema de facto, tinham de se alinhar com o país real. Mesmo que esse país real fosse analfabeto, rural, e profundamente clerical. Em vez de tentarem impor o país idealizado que só existia nas cabeças das elites de Lisboa, puseram crucifixos nas salas de aula, separaram rapazes e raparigas, e usaram manuais que não chocavam com as famílias. Pararam de tentar convencer o padre da aldeia de que eram aliados. O padre da aldeia provavelmente nunca foi aliado de ninguém que não fosse ele próprio, mas havia excepções como em tudo na vida. Os pais mandaram os filhos para a escola. A partir dos anos cinquenta, o analfabetismo infantil estava essencialmente resolvido.

Isto não apaga nada. Serve para demonstrar a tese central deste ensaio: durante um século, os regimes liberais e democráticos portugueses falharam a alfabetização porque governaram de jure, no papel, divorciados da realidade. O autoritarismo resolveu a base não por bondade nem por visão iluminada. Resolveu porque não tinha eleições para perder.

Esta é a cadeia. 1759 para 1900 para 1926. Uma borboleta que ainda voa.

Seria desonesto dizer que Pombal explica tudo.

Portugal partilha com quase todo o sul da Europa uma industrialização tardia que tem causas que vêm de muito antes do século XVIII. A geografia importa. O acesso ao carvão importa. A posição periférica em relação aos grandes centros comerciais europeus importa. Espanha não teve um Pombal equivalente e também ficou para trás. Itália também. O clima mediterrânico não produz as mesmas dinâmicas de poupança e acumulação de capital que o norte frio e cinzento. Há factores estruturais que nenhum secretário de estado do século XVIII poderia ter criado nem resolvido.

As invasões francesas entre 1807 e 1814 destruíram capital físico e institucional, forçaram a corte a fugir para o Brasil, e deixaram um vácuo de poder que levou décadas a estabilizar. A independência do Brasil em 1822 eliminou de um dia para o outro a principal fonte de receita da coroa. A guerra civil entre absolutistas e liberais entre 1828 e 1834 não foi apenas destabilizadora politicamente, foi economicamente devastadora, com destruição de infraestrutura, fuga de capital, e paralisia institucional. O condicionamento industrial do Estado Novo criou distorções que duraram muito além do regime. A gestão desastrosa da transição democrática nos anos 70 teve consequências culturais que ainda sentimos.

Tudo isto contribuiu.

Mas há uma diferença fundamental entre um país pobre com um sistema educativo funcional e um país pobre sem escolas. O primeiro tem as ferramentas para diagnosticar os seus problemas, debater soluções, e exigir mudança. O segundo passa gerações inteiras a tentar recuperar o ponto de partida antes de sequer poder começar a resolver o resto. As invasões francesas causaram dano, mas um país com capital humano absorve e recupera de invasões. A independência do Brasil foi um choque económico severo, mas economias com trabalhadores qualificados adaptam-se a choques. A guerra civil foi destabilizadora, mas sociedades instruídas constroem instituições mais resilientes depois das guerras.

O colapso educativo que Pombal criou não foi a única causa do atraso português. Foi o multiplicador que tornou todas as outras causas muito mais difíceis de superar.

O arquétipo que Portugal continua a reproduzir

Pombal não é apenas uma personagem histórica com uma estátua no centro de Lisboa. É um arquétipo que Portugal continua a reproduzir geração após geração.

O reformador centralizador que acumula poder pessoal sob a capa da modernização. Que anuncia grandes reformas com retórica iluminista, progressista, ou modernizadora. Que destrói o que existe sem ter um plano real para o substituir. Que persegue os que têm opiniões diferentes porque representam um obstáculo ao poder que quer concentrar. Que produz reformas de jure, no papel, em decreto, sem a implementação de facto que daria vida a essas reformas.

Cada governo anuncia um grande pacote de transformação que existe nos comunicados de imprensa. “Desta vez é que vai ser.” Cada lei de anticorrupção é aprovada sem recursos, sem dentes, sem aplicação. Cada reorganização institucional serve para controlar em vez de libertar. Cada pacote de fundos é apresentado como solução transformadora e é absorvido pelo mesmo sistema que impede a mudança.

A Primeira República queria criar o homem novo e destruiu o que existia sem construir o que prometeu. O fontismo investiu em capital físico e ignorou o capital humano, exactamente como hoje se constroem autoestradas em paralelo enquanto as escolas e os hospitais decaem. Cada regime pinta o anterior como o espantalho de todos os males e depois reproduz os mesmos erros com nomes diferentes.

A consequência mais profunda não está nos governantes. Está na população.

Os portugueses habituaram-se, ao longo de séculos, a que o Estado resolva tudo. Por ausência de alternativa que durou gerações. Quando o Estado destruiu o sistema educativo e nunca o reconstruiu verdadeiramente, quando controlou quem podia fazer o quê na economia através do condicionamento industrial, quando foi simultaneamente o maior empregador, o maior investidor, o árbitro de todos os conflitos, e o distribuidor de todas as benesses, a população aprendeu a relacionar-se com ele de dependência.

A forma de sobreviver em Portugal, durante gerações, foi navegar o Estado. Quem tivesse acesso ao Estado prosperava. Quem não tivesse, não.

O novo ouro do Brasil

No século XVIII, o ouro do Brasil chegou em quantidades suficientes para o rei não precisar de negociar com as Cortes. O ouro não tornou Portugal mais rico a prazo. Distorceu a economia, destruiu a indústria, e eliminou os mecanismos de controlo político que existiam. A abundância de recursos externos tornou as reformas desnecessárias no curto prazo.

Os fundos europeus são o novo ouro do Brasil.

Há quarenta anos que Portugal recebe transferências da Europa que representam uma parte enorme do investimento público nacional. Mais de oitenta por cento do investimento público em Portugal é pago com fundos europeus. Os políticos podem fazer inaugurações, construir estradas, anunciar projectos, sem que as suas decisões económicas tenham consequências imediatas visíveis para a população.

A população não sente as verdadeiras consequências da má gestão porque há sempre uma transferência europeia que cobre o buraco. Imaginem que cada vez que o Estado injectava dinheiro público numa empresa como a TAP, cada cidadão recebia uma carta a dizer: vá ao multibanco e deposite trezentos euros. As pessoas tinham-se revoltado muito mais cedo.

A pressão para reformar não existe, nem existe incentivo político. A dor suficiente para exigir mudança também.

As empresas portuguesas que recebem fundos europeus são na sua maioria pequenas empresas sem capacidade transformativa, que usam esses fundos como forma de sobreviver. Nunca vão criar nada para competir internacionalmente. Mas enquanto os fundos chegarem, não há pressão para tentar.

Portugal não converge com a Europa há mais de vinte anos. Recebe fundos há quarenta.

Mas pelo menos faz sol, temos o pastel de nata, a comida é boa e o Ronaldo é português.

Os empresários, os políticos e o espelho

Há um ódio latente aos empresários em Portugal. E é um ódio mal dirigido.

O tecido empresarial foi moldado por séculos de incentivos perversos. Num ambiente em que as regras mudam conforme quem manda, em que os contratos não são garantidos pelos tribunais com consistência, em que o acesso ao mercado depende de relações políticas, os empresários que prosperam são os que aprenderam a jogar esse jogo. Os que ficaram perto do poder. Os que transformaram a proximidade ao Estado numa estratégia de negócio.

O sistema produziu exactamente os empresários que era suposto produzir. Um ambiente com estes incentivos não produz empreendedores que destroem criativamente o mercado. Produz quem protege as suas posições e caça rendas.

A mobilidade social em Portugal é vergonhosamente baixa. Os mesmos nomes de família estão em todo lado. Os mesmos círculos, as mesmas escolas, as mesmas redes. Numa sociedade estagnada, mesmo que os mais pobres vivam marginalmente melhor em termos absolutos, a distância relativa em relação aos que estão no topo não diminui.

Os políticos portugueses são produto da mesma sociedade, das mesmas escolas, da mesma cultura, do mesmo sistema de incentivos. Se uma pessoa é genuinamente competente, porque razão escolheria a política portuguesa? O escrutínio público constante é só triste nesta era de infotainment. A remuneração não compete com o sector privado em funções com o mesmo nível de responsabilidade. O sistema recompensa a lealdade partidária mais do que a competência. O mandato dura quatro anos, as reformas estruturais levam décadas a dar frutos.

Os que se sujeitam a isto ou não conseguem fazer melhor noutro sítio, ou vão com a intenção de aproveitar o que o sistema oferece a quem estiver dentro. Uma sociedade com baixo capital humano, com fraca sociedade civil, com baixa literacia económica e financeira, com dependência histórica do Estado, não cria as condições para que os melhores queiram ou consigam governar.

Existem excepções. Claro que sim. Mas infelizmente são mesmo isso, excepções.

A literacia que nunca foi construída

Portugal tem os níveis de literacia económica e financeira mais baixos da Europa Ocidental.

É uma consequência directa de séculos de subinvestimento em capital humano de qualidade. Saber ler e escrever não chega. Uma democracia funcional precisa de cidadãos que percebam como funciona uma economia, que consigam avaliar as promessas dos políticos, que entendam o que é produtividade e porque é que importa.

Sem essa literacia, a população é governada por slogans e subsídios. É convencida de que mais Estado resolve os problemas que o Estado criou. É distraída com inaugurações de infraestrutura física enquanto o capital humano se deteriora. É mantida numa ilusão de bem-estar financiada por fundos que vêm de fora.

Quem controla o acesso ao Estado, quem captura as instituições, quem define quem entra nas universidades e quem aparece nos órgãos de comunicação social, tem todo o interesse em que a população continue com baixa literacia e baixa capacidade de fiscalização. A mobilidade social baixa é um sistema, não um acidente.

O clube de que não fazemos parte

Há um bit do George Carlin, comediante americano, com mais de vinte anos, em que ele diz que os donos do país não querem uma população capaz de pensamento crítico. Querem trabalhadores obedientes, suficientemente inteligentes para operar as máquinas e fazer a papelada, e suficientemente dóceis para aceitar passivamente empregos cada vez piores com salários cada vez mais baixos. O contexto é americano mas a descrição encaixa em Portugal como uma luva, com as devidas adaptações provinciais.

Temos um povo com uma dependência estrutural do Estado que não é acidente nem preguiça colectiva, é o resultado lógico de décadas de sistema. Temos uma massa enorme de pensionistas, muitos deles a não conseguir escolher entre comida e medicamentos com o dinheiro que recebem depois de uma vida a trabalhar. Temos uma máquina de funcionários públicos que nunca para de crescer, assessores, tesoureiros, chefes de gabinete, presidentes de mesa, presidentes de junta, assessores do assessor, secretários do chefe do gabinete do secretário de estado, observatórios de isto e daquilo que se multiplicam constantemente com a sua lógica própria de existência. Temos empresas cujo único ou principal cliente é o Estado e os seus devidos funcionários. Temos os subsídios de deslocação dos deputados que casualmente vivem todos a quinhentos quilómetros do local de trabalho. Temos o tachismo crónico, as burlas, os favores, o branqueamento tranquilo de geração em geração.

Os subsidiodependentes no sentido estrito existem, mas são residuais. O problema não é esse. O problema é a proporção do país que vive directa ou indirectamente do dinheiro público sem que alguém consiga explicar com clareza o valor que produz.

E depois sobram os tristes que trabalham e tentam fazer alguma coisa da vida.

A estimativa razoável é que menos de vinte por cento da população activa produz efectivamente valor transacionável, o tipo de valor que se pode vender fora do país e que traz riqueza real para dentro. O resto trabalha para gerir, redistribuir, e consumir o que esse vinte por cento produz, com uma fatia crescente a trabalhar directamente para o Estado ou para quem vive do Estado e se serve do Estado.

E o sistema perpetua-se com uma elegância quase admirável. De vez em quando acenam os pensionistas com aumentos de cinco por cento e a eleição está ganha. Os pensionistas não têm culpa, a maioria não sabe se compra comida ou medicamentos porque o dinheiro não chega, e os erros que os trouxeram até aqui foram em parte deles mas em parte de um sistema que nunca lhes explicou como funcionava nem lhes deu alternativas. É uma ineficiência gigantesca que não resolve o problema de ninguém mas mantém toda a gente suficientemente anestesiada para não o resolver.

A corrupção é dispersa e por isso quase invisível. Não é o grande escândalo ocasional que aparece nos jornais. É a soma de mil pequenas cumplicidades, o contracto da câmara, a obra da junta, a nomeação do amigo, a empresa pública que nunca fecha porque fecha empregos. Cada caso individualmente é demasiado pequeno para mobilizar alguém. Em conjunto representam uma sangria constante de recursos que este país não tem para desperdiçar.

O sistema não precisa de conspiração para funcionar. Só precisa de incentivos alinhados. E os incentivos em Portugal estão todos alinhados para manter as coisas exactamente como estão.

A geração que chegou tarde

A geração mais nova é genuinamente boa. Bem formada, competente, cosmopolita, com acesso a oportunidades que as gerações anteriores infelizmente não tinham. A programação, a engenharia, a medicina, a enfermagem, as ciências exactas produzem portugueses que competem sem complexos a nível internacional. Nunca foi falta de talento ou vontade.

O problema é a pirâmide etária e o peso do passado.

Portugal tem uma massa crítica enorme de gerações formadas num sistema educativo fraco, numa cultura de dependência do Estado, num mercado de trabalho que não recompensa o mérito. Essa massa não desaparece porque apareceu uma geração melhor preparada. Vota, consome, influencia a cultura, ocupa posições de poder, e é a maioria. Faz parte.

Os melhores estão a sair. Os jornais descrevem isso como tragédia nacional. É uma resposta racional a um sistema de incentivos que não mudou. Quando o teu país não recompensa o mérito, quando os salários são uma fracção do que consegues noutro lado da Europa, quando as perspectivas de progressão dependem mais de quem conheces do que do que fazes, sair é matemática.

Um em cada cinco portugueses vive fora do país. Somos o país da Europa Ocidental com maior percentagem de cidadãos emigrados. Um historiador económico que passou a vida a estudar este país diz publicamente aos jovens que saiam, porque Portugal não lhes vai dar um bom futuro, e eu infelizmente concordo. Com a frustração de quem vê um país que não quer mudar, que não sente dor suficiente para mudar, e que vai continuar assim enquanto o dinheiro de fora continuar a chegar.

Quando a torneira fechar

As grandes mudanças em Portugal foram quase sempre impostas de fora. A globalização do século XV. A influência inglesa no século XIX. A entrada na CEE nos anos 80. A troika em 2011. O país raramente mudou porque quis mudar. Mudou porque foi forçado. Tivemos de esperar que o outro senhor caísse da cadeira e, mesmo assim, precisámos de mais uns anos de vácuo até que a liberdade acontecesse. E o mais extraordinário é que o símbolo máximo dessa liberdade não veio de decretos de políticos iluminados. Veio do país real: da coroa de dignidade e coragem do capitão Salgueiro Maia na rua, e da generosidade espontânea da Celeste Caeiro que, ao partilhar os seus cravos com os soldados, batizou a democracia. Eles fizeram mais pelo país real do que um século de burocratas de jure.

O próximo choque virá quando os fundos europeus diminuírem ou acabarem, porque todos sabemos que nada dura para sempre. A Alemanha está em recessão, mesmo que isso não seja muito falado. A França tem défices insustentáveis. Os extremos crescem por toda a Europa. A União Europeia pode reformar-se, contrair-se, ou simplesmente não ter dinheiro para continuar a financiar a inércia portuguesa ao ritmo actual.

Quando a torneira fechar, a ilusão acaba. Noventa por cento do investimento público financiado de fora é uma dependência, não uma política. E dependências têm sempre um momento de abstinência.

O que acontecerá depois é genuinamente difícil de saber. Pode ser uma oportunidade de reforma real. Pode ser o caos. Pode ser a ascensão de populismos que aproveitam a frustração acumulada. A verdade costuma estar a meio, por isso provavelmente uma mistura dos três.

O país não vai mudar porque alguém explicou bem a história. Vai mudar quando a realidade forçar a mudança, como sempre forçou. Não sei é se quando isso acontecer haverá capital humano, literacia, e consciência histórica suficientes para fazer as escolhas certas.

O que a estátua representa

A estátua do Marquês de Pombal representa a capacidade que Portugal tem de transformar os seus maiores fracassos em narrativas de sucesso. De olhar para a destruição e ver heroísmo. De celebrar o mínimo aceitável como se fosse extraordinário. De erguer estátuas aos responsáveis pelo atraso enquanto continua a não perceber porque é que está atrasado.

Enquanto a estátua estiver lá, a história oficial permanece intacta. E enquanto a história oficial permanecer intacta, as perguntas certas não são feitas.

Porque é que somos o país mais pobre da Europa Ocidental?

Porque é que as nossas instituições são tão fracas?

Porque é que as reformas ficam sempre no papel?

Porque é que os melhores saem?

Porque é que depois de quarenta anos de fundos europeus ainda não convergimos?

A resposta está numa cadeia de decisões que começa no século XVIII, quando um homem que hoje tem uma estátua no centro de Lisboa destruiu o sistema educativo do país, gastou o dinheiro noutras coisas, e deixou um Portugal sem escolas que levaria gerações a reconstruir.

O de jure dizia que ia construir escolas. O de facto foi destruir as que existiam.

Pombal fez o mínimo aceitável pelo terramoto. Pela educação, fez o máximo dano possível.

Ainda estou cá. E cada vez que leio mais um relatório sobre divergência económica, cada vez que vejo outro jovem ir embora, cada vez que faço as contas entre o que ganho e o que ganharia noutro lado da Europa, a matemática fica mais difícil de ignorar. Por agora fico. Mas se algum dia sair, não vai ser por raiva nem por desistência. Vai ser exactamente pela mesma razão que explica tudo o resto neste ensaio.

Entretanto, a estátua continua a irritar-me.

Subscribe now

The market most people ignore

Most people watch the stock market. It gets the headlines, the CNBC segments and the Reddit threads. But did you know that the bond market is $140 trillion? Crazy, right? It is roughly twice the size of global equities and it is the foundation that every other interest rate in the economy is built on. Your mortgage rate ( if you can afford a house in this market). Your car loan. Your credit card rate. Every business loan. Every government’s ability to pay its bills. All of it is priced off the bond market. When the stock market crashes, it is painful but eventually recovers. When the bond market breaks, the cost of borrowing money goes up for everyone simultaneously, governments cannot fund themselves without printing, and the economic damage spreads in ways that take years to unwind. The bond market breaking is a different category of problem from a stock market correction. Right now, the bond market is telling you something very important.

What is happening

The US 30-year Treasury yield just hit over 5%, the highest level since July 2007, right before the financial crisis. The 10-year yield is up 75 basis points since the Iran war started. Bond yields are at multi-decade highs across the UK, Germany, France, Canada, Australia, and Japan simultaneously. This is not a US-specific problem. When you see the yields rising everywhere at the same time, it means investors globally are losing confidence in governments globally at the same time. And let’s be honest… it is very difficult to trust governments … Investors are demanding higher interest rates before they will lend. The worse the trust, the higher the rate. Same logic as a bank charging you more when your credit score drops.

Why it is happening

Three reasons are converging simultaneously:

The first is inflation, already present and more still coming through the supply chain. Since the Iran war started, crude oil is up 60%, jet fuel up 58%, gasoline up 52%, fertilizer up 30%. PPI hit 6%, the highest since 2023. CPI is back at 3.8% against a 2% target. The fertilizer price increases from February have not yet fully shown up in grocery prices ( but they will). That lag is usually 3 to 6 months. The full inflationary effect of what happened to commodity prices since February is still working its way through the system.

The second reason is that the biggest foreign buyers of US debt are pulling back. China has reduced its direct Treasury holdings from a peak of $1.3 trillion to around $652 billion as of March 2026, the lowest since 2008. Some analysts argue that a portion of those apparent sales moved into custodial accounts in Belgium and Luxembourg rather than leaving the system entirely. It is the same custodial bias caveat that complicates the headline number. The directional trend is real even if the precise scale is contested. Japan, the largest foreign holder at around $1.1 trillion, has been selling to defend the yen and fund oil purchases. In the first quarter of 2026 alone, Japan sold more US Treasuries than in the previous four years combined. Japan’s position deserves its own explanation, which I covered in an earlier piece [How Japan Could Break the Global Economy], but the short version is that Japan raising rates to save the yen forces it to sell US Treasuries, which pushes US yields higher, which strengthens the dollar, which forces Japan to sell more Treasuries. There is no clean exit from that loop.

The third reason is math. The US carries $39 trillion in official debt, adds roughly $2.5 trillion per year, and is now spending over $1 trillion annually just on interest payments. Over 20% of all federal tax revenue is now consumed by net interest payments alone, before a single dollar is spent on anything else. That figure rivals defense spending and sits at a multi-decade high. Bond investors who lend money for 10, 20, or 30 years look at that math and reach the same conclusion: there is only one exit from this that does not require politically impossible tax increases or spending cuts. Print your way out. And if you expect the government to print its way out of the debt, you demand a higher interest rate before you will lend your money today. That is why yields are going up.

The trap

The Federal Reserve is sitting with an impossible choice and both options are bad. If it cuts rates with PPI at 6%, CPI at 3.8%, and oil above $100, the bond market revolts. Investors sell Treasuries, yields go up anyway, and borrowing costs rise despite the cut. The thing the Fed is trying to prevent happens regardless of the cut. If it raises rates, the government pays even more on $39 trillion in debt, credit card delinquencies already above 12% get worse, auto loan defaults increase, private credit markets crack further, and the housing market slows more. Both options make the underlying problem worse. The trap is that there is no move that does not accelerate something bad.

Historically, every time a government has faced this math, it has chosen the invisible option. Inflate the way out. Print money. Lower the real value of the debt by making each dollar worth less over time. In the 1970s, the US was looking at the same impossible math. The dollar lost 50% of its purchasing power between 1970 and 1980. Gold went from $35 to $850 an ounce in the same decade. Bond investors looking 10 to 30 years ahead are pricing in a version of that outcome. That is why yields are high and going higher.

The ruler change

Here is the detail that deserves more attention than it is getting. Kevin Warsh, the newly installed Fed chair, took office promising structural regime change. During his Senate confirmation hearings he called core PCE a "rough swag" and stated a preference for the Dallas Fed's trimmed mean PCE, which strips the top and bottom outliers across all spending categories. The trimmed mean is a legitimate methodological preference with genuine academic support. It currently reads around 2.3%, well below core PCE's reading above 3%. Brookings noted in April that the trimmed mean is currently excluding much of the oil and tariff shock from its calculation. The preference is real. The timing is convenient. The bond market is drawing its own conclusions. It is now pricing over 63% probability of a rate increase by March2027. Chris Waller, one of the first Fed members to vote for cuts, said he can no longer rule out rate hikes if inflation does not abate. The Fed chair was hired to cut rates. The bond market is telling him to raise them. The stock market is at all-time highs pricing in a bailout. And my spidey sense is telling me that they cannot both be right at the same time

This is the same bond market tripwire that has constrained the Iran war throughout the conflict. I have been covering this across several posts for a while now, but i’ll say it again: Every time the 10-year yield approached 4.4%, the US military posture softened and we got a lot of TACO (Trump Always Chickens Out). Iran noticed that. Yields are now well above that level and the pressure on the Fed is structural rather than situational. The Japan doom loop, the sovereign creditors selling, the inflation still working through the supply chain, and a Fed chair proposing to change the inflation ruler are all pointing in the same direction. The pressure is not going away regardless of how the conflict resolves. The bond market was the constraint on the war. It is now the constraint on everything else.

But no one seems to notice… no one seems to care…

Subscribe now

AGI stands for Artificial Gaslighting Intelligence. For now.

AI may truly become something genuinely transformative. The people building it believe that sincerely and some of them are probably right about the direction even if wrong about the timeline. This post is about the right now, not that future.

Right now, the largest amount of money ever deployed in human history is being justified by a technology that cannot define its own terms. AGI is the stated destination for most of the investment flowing into this sector. The problem is that nobody agrees on what the hell general intelligence means, nobody agrees on what consciousness means, and David Deutsch, the father of quantum computing, does not believe dogs are intelligent or conscious. I am not making this shit up. We are out here debating whether a Python script can think and this man is like: “dogs are not even alive bro”. At that point you have to step back and ask what exactly we even are all arguing about. You cannot arrive at a destination that has not been defined.

But wait…. It gets worse. Every time LLMs fail a benchmark, that benchmark gets rewritten. Every time they pass one, the definition of intelligence shifts to exclude what they just did. The goalposts are not fixed. They move every time the technology falls short of them. You cannot calculate a return on investment against a target that will not stand still. But the capital is being deployed anyway…

We honestly could be looking at the largest sunk cost fallacy in human history. Trillions already committed to data centers, chips, and energy capacity whose returns have not been proven at the scale required to justify spending that money in the first place. The sunk cost fallacy is when you keep spending because of what you have already spent, not because the future return is clear. At some point the question changes from whether AI will change everything to whether it will change enough, fast enough, to justify what was already spent before the answer was known.

There is a specific financial problem sitting underneath this that almost nobody is discussing. A traditional factory depreciates over 20 to 30 years. An AI data center built on today’s chip clusters faces technological obsolescence in 3 to 5 years as chip designs keep evolving (assuming current rate). The companies building this infrastructure are booking it like long-term real estate, when in fact it has the shelf life of a phone. The return on what they are building degrades faster than the accounting assumes, which means the actual return on invested capital is significantly worse than the balance sheet implies. The equity prices sitting on top of those balance sheets are pricing in the long-term real estate version, not the phone version.

The token problem

Someone recently posted a screenshot of $1.3 million spent on OpenAI tokens in 30 days, 603 billion tokens, as a demonstration of what AI-powered development looks like at scale. The reaction in most places was AWE... A developer pointed out that this is the same energy as the startup era of 2016 to 2020, when companies had more internal tools than actual customers and nobody questioned whether any of it made sense because it was fashionable and looked good on LinkedIn. The same thing is happening now with tokens. People are getting fired for not using enough AI. Token spend is being used as a measure of productivity. At some point the companies paying those bills will ask who is actually getting things done, not who is spending the most.

Google at its IO conference this year bragged about token usage going from 9.7 trillion in 2024 to 480 trillion in 2025 to 3.2 quadrillion in 2026. Those numbers look exponential on a chart. They feel like a growth story that never ends and is perfect to put in Excel to show everyone. But if you look at the rate of change rather than the absolute numbers, the year-over-year growth multiplier collapsed from roughly 50x down to roughly 6.7x. That is an 86% drop in the growth rate. The absolute number keeps going up but the acceleration is gone. That is the second derivative going negative, and it is the number the infrastructure investment thesis depends on staying positive. Worth noting that these are Google’s numbers specifically. Other providers may look different. But Google is not what you would call a small sample.

An LLM is a system trained to predict the next word given everything it has read. It does this well enough that the output feels like reasoning, feels like understanding, feels like thinking. The feeling is the product. The bot does not have to be intelligent. It just has to be more convincing than the median CEO is skeptical. That bar has been cleared comfortably. The money followed the feeling.

I can smell it… The consulting class is 100% coming. Slop masters. Token coaches. I can already see them on the horizon. Same guys who were blockchain specialists, then NFT specialists, then B2B sales gurus, and now AI specialists.

Btw, leave the Doggos alone. They are more intelligent than most people I know.

You can polish a lie forever

More parameters make the outputs more convincing. They do not make the system understand anything in any sense that has been agreed upon. You can polish a lie forever and it never becomes the truth. The outputs get better of course, but what is underneath is the same. That gap between what these tools are good at and what the investment story requires them to be good at is where the sunk cost fallacy lives. That gap is being covered by narrative rather than by results.

Artificial Gaslighting Intelligence. For now.

Subscribe now

Quick update, Sunday May 24th. Keeping this one informal because a lot happened last night and some of it does not really add up.
Might do a longer update when/if something else materializes.

So… Yesterday at 4:32pm Trump told Axios he was solid 50/50 on deal or bombing, but by 9:30pm the same evening he announced on Truth Social that an agreement had been largely negotiated and the Strait of Hormuz would be opened.

Just 5 hours between those two statements. The full Gulf players were on that call. Saudi Arabia, UAE, Qatar, Pakistan, Turkey, Egypt, Jordan, Bahrain. Netanyahu separately according to Trump. Either those calls were really good, or this is one more theatre play. Iran immediately suggested the latter of course.

Iran’s Fars news agency said the Strait would remain under Iranian management, which directly contradicts Trump’s announcement. Senior Iranian sources told Reuters that Tehran has not agreed to hand over its highly enriched uranium stockpile and that the nuclear issue is NOT part of the preliminary agreement. Tasnim confirmed that Iran has not accepted any actions on its nuclear file. Rubio, hours after Trump declared the Strait would open, said there was a possibility of good news on the Strait over the next few hours. Which suggests the opening was not actually confirmed when Trump announced it.

The actual structure per Axios and the NYT: a 60-day ceasefire extension during which the Strait reopens over 30 days, Iran freely sells oil, sanctions on Iranian oil are waived, and nuclear negotiations are held. The uranium stockpile question gets kicked to a 30 to 60 day negotiation window. Iran is also demanding Israel end the war in Lebanon as a condition of the MOU. That condition is not in Trump’s announcement.

The conflict’s stated purpose was to prevent Iran from ever acquiring a nuclear weapon ( some of my previous posts say otherwise). That is not in this deal. Both sides declared victory and deferred the hardest problems.

Since April the conflict has had two documented patterns.

The TACO pattern : Trump escalates, sets deadlines, backs down. The April 8th ultimatum expired with Trump walking it back an hour and twenty minutes before the deadline.

A Bloomberg trader coined NACHO on April 29th: Not A Chance Hormuz Opens. The second pattern is more specific and something I have been tracking in this series: progress gets announced before market closures and walked back after markets open.

This announcement follows both simultaneously. Saturday night before a three-day Memorial Day weekend. Earlier this week a $129 million options position was placed betting on a sharp drop in oil by Tuesday. Deal holds means oil craters. Deal collapses means oil spikes. That position is a bet on downward capitulation. Whether it reflects information or coincidence is a question worth sitting with.

This is the most advanced iteration of a deal in the conflict and i think that the Beijing summit context last week is the key here. The 60-day nuclear window is where the conflict’s stated purpose either gets resolved or gets deferred again. Watch what happens in that window, not what was announced on Saturday night.

Subscribe now

What follows is a theory. I am going to present it as clearly as I can. I do not think all of it is correct, and I will explain why in the second half.

Subscribe now

What they did to Japan

In 1985 the US had the same problem it has now. The trade deficit was too large. The dollar was too strong. Manufacturers who could not compete. The Reagan administration called a secret meeting at the Plaza Hotel in New York with France, West Germany, Japan, and the UK, and the deal they made was simple in the way that expensive things often are: everyone would intervene in currency markets to weaken the dollar against the yen. The yen nearly doubled in value almost overnight. Japanese exports became expensive, American goods became cheap, the deficit narrowed, Japanese companies built factories in the US, money flowed, and for a while everyone called it a success. Then Japan flooded its own economy with cheap money to absorb the shock, created the largest asset bubble in modern history, watched it collapse, and spent the next thirty-five years trying to recover. Its stock market did not regain the 1989 peak until 2024. One weekend at a hotel in New York destroyed the second largest economy on earth for an entire generation. The Plaza Accord is why. China was watching all of it and probably has been thinking about it ever since.

Beijing

Trump flew to Beijing with eighteen CEOs. Elon Musk, Tim Cook, Jensen Huang, Larry Fink from BlackRock. It was the most concentrated deployment of American private capital ever assembled for a foreign trip. Xi opened by invoking the Thucydides trap, the Greek observation that when a rising power threatens an established one, the result is almost always war in twelve of sixteen historical cases, which is a strange way to start a conversation about tariffs unless the conversation is actually about something else. On the margins of the summit Xi showed Trump trees that are over 300 years old. Older than the United States, which turns 250 this year. I think we don’t need a translator for that one.

The official story from the summit covered tariffs, manufacturing investment, and the Iran nuclear question. Both leaders agreed that Iran could never have a nuclear weapon. Worth sitting with that for a moment because Russia and China both have nuclear technology. Both also have every geopolitical motive to give Iran a weapon if they wanted to end the war on their own terms. But they have not done it. Whyyyyyy, you may ask?

A closed Strait of Hormuz is a slow, controllable drain on the Western financial systems. A nuclear Iran is an uncontrollable event that serves nobody’s interests cleanly (probably not even Iran). The Strait being closed is the weapon of choice precisely because it is manageable.

Plaza Accord 2.0

There is a macro thesis gaining traction in institutional circles that most financial media is not covering. It proposes that what we are watching is not a series of chaotic geopolitical accidents but the deliberate, painful unwinding of the post-war dollar standard and its replacement with something new. The thesis says the Beijing summit was not really about tariffs but about a new monetary order structured around gold rather than currency pairs, one that lets both sides claim victory while the dollar loses purchasing power gradually rather than catastrophically.

Beijing has been explicit for years that it will not allow a direct revaluation of the yuan against the dollar. That is probably the institutional memory of Japan speaking. China watched a currency revaluation destroy a generation of Japanese wealth and has spent four decades trying to make sure that the same mechanism is never applied to the yuan. So if there is a deal, the revaluation cannot happen through the currency pair. It has to happen through something else, something that lets both sides claim they did not blink. The thesis says that something else is a shiny piece of rock. GOOLDDDD.

The US government holds approximately 8,133 metric tons of gold, the largest national reserve in the world. On the Treasury’s books that gold is valued at $42.22 per ounce, a statutory price set by Congress in 1973 and never updated. The current market price is over $3,200 per ounce. Nearly a trillion dollars of value that does not appear in any debt-to-GDP calculation, any deficit discussion, or any serious conversation about American fiscal capacity. If those reserves were marked to market, the debt burden looks dramatically more manageable without printing a dollar or issuing a bond. For China the mechanism runs the other direction: Beijing has been accumulating gold aggressively for years, and if gold reprices significantly upward, China’s reserves appreciate without touching the yuan exchange rate. Both sides get richer. Both sides claim victory. Neither has to describe publicly what just happened.

The structural observation underneath all of this stands regardless of whether the deliberate coordination interpretation is correct. The same countries that are the largest holders of US Treasury bonds, Japan, China, the United Kingdom, France, are also the most exposed to a Hormuz closure. Japan sources approximately 75% of its oil from the Middle East. China roughly 50%. These are not countries that can afford to abandon the dollar system at the same moment their primary energy supply is disrupted. Canada is the anomaly worth noting. It is the one major creditor that is energy-independent, and it is also the one country on that list currently being pressured by the United States in ways that look less like a trade dispute and more like something structural. If the thesis holds, pressuring Canada is about ensuring the one energy-independent creditor ends up inside the American energy orbit rather than becoming an alternative supplier to the countries you are trying to bring into dependency.

What is leaving

Physical gold has been the largest US export category since October 2025, overtaking oil, pharmaceuticals, and aircraft engines. The majority goes to Switzerland first for refining and then to China via Hong Kong. In February 2026, precious metals reached 15% of total US export value against a historical average of around 4%. When the British Empire wound down, gold left London and came to New York, and by the time World War II ended the US held more than half of all the world’s gold. That is why the dollar became the reserve currency. Bretton Woods was not an ideological project. It was a reflection of who had the metal. The country with the gold writes the rules. Right now the US is the single largest exporter of gold in the world and it is going to China.

The market signals point in the same direction. Since late March the dollar has been falling against the yuan, which is the opposite of what happens when a war is successfully pressuring China. Chinese government bond yields are steady or declining while everyone else’s borrowing costs are rising. Gold has been on an extended run that predates any announcement. The yuan is strengthening while Hormuz is closed and the official story is that China is under pressure. None of those things are consistent with each other unless the deal is already being priced in by people who expect it to happen. There is a second dynamic running underneath this: for decades, global investors borrowed yen at near-zero rates, converted to dollars, and invested in US assets earning 5%. Estimates put total carry trade positions at over $20 trillion. In August 2024 the Bank of Japan raised rates a fraction and the Nikkei fell 12% in a single day. Analysts estimate only half the carry trade unwound in that event. The other $10 trillion still exists, still dependent on low Japanese rates that are no longer low. As that unwinds, capital flows interact with the gold move, the dollar weakness, and the yuan strength in ways that are difficult to read in isolation but more significant taken together.

That is the theory. I find it very compelling.

I also find it suspicious that I find it very compelling.

The first and most fundamental problem is that the theory is unfalsifiable by design. Every piece of evidence consistent with deliberate strategy is equally consistent with a simpler explanation: a sequence of decisions made for unrelated reasons that happen to produce a structurally favorable outcome. How convenient, right? A leader who stumbles into an advantageous position looks identical from the outside to one who engineered it. Unfalsifiable theories are not necessarily wrong…. They are just impossible to trust…

The Venezuelan infrastructure problem is the structural hole in the middle of the theory. Venezuelan crude is extra-heavy, low quality, and requires upgrading infrastructure that has been degraded by two decades of mismanagement and sanctions. The reserves are real but the delivery capacity is probably not. A global supply shock triggered by a Hormuz closure would need a response measured in weeks or months. Venezuelan supply cannot fill that gap on any timeline that matters. This is not a footnote to the theory.

The petrodollar transmission mechanism is also weaker than the theory requires. The assumption that energy dependency in 2025 automatically converts into continued Treasury demand treats a historical political arrangement as a mechanical law. Countries under acute financial and energy pressure do not automatically buy bonds. They sometimes even sell them to raise liquidity. The last two years of Treasury market volatility suggest the transmission is already weaker than it once was. And the gold flow, while genuinely anomalous in scale, it is mostly private bullion rather than sovereign reserves. Western investors swapping gold for 5% yielding cash while Asian retail buyers absorb the physical supply is not the same as governments deliberately repositioning. There is also the renewable energy escape route. The theory requires affected countries to remain trapped in fossil fuel dependency long enough for the mechanism to work. Europe has been moving away from that dependency, inconsistently but directionally, for over a decade. The pressure itself may accelerate the very escape it depends on preventing ( maybe nuclear will be back ?).

What stands regardless

The structural overlap between the largest Treasury holders and the countries most exposed to a Hormuz disruption is real and genuinely underappreciated. The geographic logic of Western hemisphere resource consolidation as a long-term strategic play is real. The fragility of the petrodollar system is real. The Japan carry trade stress is real. The gold flow is real even if the interpretation is contested. These observations stand on their own even if the intentionality claim does not. Empires do not necessarily require masterminds. They require math that works in their favor, and structural incentives that push in the same direction regardless of whether anyone is consciously coordinating them.

There is a framework that makes all of this less surprising. A century ago geopolitical thinkers identified two types of power: Heartland powers controlling the interior of Eurasia, resource-rich and largely landlocked, and Rimland powers controlling the coastal edges, the trade routes, and the sea lanes where economies actually connect. The competition between them has never really stopped. The countries most exposed to a Hormuz closure are Rimland nations whose prosperity depends on maritime access. The countries accumulating gold and building alternative settlement infrastructure are Heartland powers trying to reduce that exposure. If you read through the Heartland/Rimland frame, Plaza Accord 2.0 looks less like a financial conspiracy and more like the latest expression of a geographic competition that has been running for a century. Whether or not Trump and Xi signed anything specific in Beijing, the underlying forces pushing in this direction do not require a meeting to keep operating.

You do not buy home insurance because you are certain your house will burn down tonight. You buy it because if it does, you would be ruined. Whether Plaza Accord 2.0 is an accurate description of what was negotiated in Beijing or a compelling narrative built on real but misread signals, the direction it points is the same: In an environment where the dollar's long-term purchasing power is structurally uncertain, holding assets that governments cannot print is basic civilizational insurance. The end of the world is not required.

The theories that explain the most tend to be the ones quietly assuming the most. Sit with that before you decide how much of this one to believe.

Subscribe now

Most people should be Bogleheads. I understood compound interest before I was legally old enough to open a brokerage account, which made me the strange kid in a country where most people around me treated investing as something other people did. I am writing this because the conversation most people are having about money is wrong, and the right version is so simple it feels like it must be a trick.

But i can hear your question: “WTF is even a Boglehead?”

A Boglehead is someone who follows the investment philosophy of Jack Bogle, the founder of Vanguard. The term started as a nickname on the internet for people who took his ideas seriously. It stuck because the ideas work. The philosophy has no secrets, no proprietary signals, no edge you need to pay for. It is built on three observations that have held true across every market cycle in modern history: costs compound against you the same way returns compound for you, time in the market beats timing the market, and the average active fund manager underperforms a simple index over long periods after fees. That is the whole thing. People built a community around it because it turns out most of financial media exists to make you think the answer is more complicated than it is

Step 1: Buy a broad market index fund.
Step 2: Keep costs as low as possible.
Step 3: Invest every month regardless of what is happening in the world.
Step 4: Do not touch it for decades.

That’s it. That is the entire strategy… Sounds simple, right ?

Jack Bogle built Vanguard around this idea in 1976. A dollar invested at 7% annual return over a long period produces roughly three times more than the same dollar at 5% after fees. The 2% is not the market. It is what you pay someone that wears a tie to manage it. Morningstar found their sophisticated rating system performs almost as well as simply ranking funds by cost. They published that. The confessional booth, Bogle called it.

You do not need a thousand dollar course from someone on the internet with a rented Lamborghini in the thumbnail.

Wanna know what are Warren Buffett’s instructions for his wife’s trust after he dies? 90% in a Vanguard S&P 500 index fund and 10% in short-term government bonds. That is from his 2013 letter to Berkshire Hathaway shareholders. The person who has arguably generated more alpha than anyone in the history of markets thinks his wife should be mostly a Boglehead. Sit with that.

The enemy is you

A JP Morgan study looked at the S&P 500 from 2001 to 2020. Stay fully invested: 7.5% average annual return. Miss the 10 best days: 3.4%. Miss the 20 best days: 0.4%. Nearly flat. Two decades of participation for roughly zero return.

The best days almost always happen inside the worst periods. When the market crashes, the sharpest single-day recoveries happen during the crash. You cannot exit to avoid the bad days without missing the good ones. They are the same days. The person who sells to wait for things to calm down misses the days that generate most of the long-term return. This is a mechanical property of how markets move, and it will keep being true regardless of how reasonable your timing thesis sounds.

Did you know that women tend to outperform men in investing? They trade less. Ego is expensive. DCA removes the problem at the source. You set it up, it runs, and when everything is falling you do not have to be smart or brave or emotionally stable because the action is already decided. You buy more at lower prices automatically without meaning to.

Ronald Read was a janitor and gas station attendant in Vermont. He drove an old car. He wore flannel shirts held together with safety pins. When he died in 2014 at 92, he left $8 million to his family, his local library and hospital. Nobody knew. He had spent decades quietly buying shares in companies he understood and never selling them. The fortune was discovered after he died. No research edge. No Bloomberg terminal. No macro thesis. Just time and the discipline to do nothing with money that most people would have spent.

The 1929 argument

Yes it can happen again. If you are five years from retirement when it does, it is genuinely painful, but it would still be painful if you were not invested. The people who kept investing through 1929 were back to even by the mid-1940s. The people who sold locked in the loss and missed the recovery. The recovery only came to the people who stayed.

The only scenario where none of this works is genuine civilisational collapse. Like nuclear war. The kind of thing where the problem is not your portfolio. If there is nuclear war, it will not matter a hell of a lot whether you owned stocks or bonds or cash. You can spend your life worrying about that or you can invest. Both options end the same way if the bombs actually go off. Only one of them builds anything if they do not. Your call.

The objections

The market is corrupted and manipulated. Yes, in the short term. In the long term it is a claim on the earnings of real companies selling real things to real people. The price moves. The earnings compound regardless. Every major index has recovered from every major crisis in modern financial history. Companies kept selling things. That is why.

A genuine caveat before the arithmetic. There are people for whom investing right now is simply not the conversation to be having. If you are choosing between food and rent, the Boglehead strategy is not your immediate problem. There is no shame in that and no clever reframe that changes it. This post is not for that moment. It is for everyone who is not in that moment but tells themselves they are.

“It is only for the rich.” Twenty euros a month. Twenty dollars. Twenty of whatever currency you have. Invested consistently over thirty or forty years, twenty euros a month becomes a number most people would not believe without running the calculator. Growing up in Portugal, most people around me never invested anything. The amounts were not the barrier. The conversation simply never happened because most people had no idea how it works.

Here is what $100 a month looks like over 40 years, assuming you never increase the amount. No salary adjustments. No inflation adjustments. Just $100 a month, every month, left alone.

If your financial situation is what scientists would describe as "not a complete disaster" and your situation allows you to invest $300 a month, run the same calculator. The number on the other end is close to a million.

Let's say I failed mathematics at every level of education ( i didn't), my calculator is broken ( i don't think it is), and I have never been right about anything in my life ( no objections here). Cut it in half. You still have $500,000. From $300 a month. Sorry for ruining your day.

Most people have more investable money than they have honestly accounted for. In Portugal a pack of cigarettes costs around five euros. Smoke one every two or three days and you have fifty to one hundred euros a month sitting there doing nothing. You do not need to quit. You do not even need to smoke less if you genuinely do not want to. But if you smoke a pack a day and have absolutely no intention of stopping, I have genuinely good news: you probably do not need to worry about this at all. The timeline will take care of itself.

For everyone else: one fewer dinner out per month. One cheaper holiday. Buy the game two months after launch when it is a bit cheaper. Skip the subscription you forgot you had. Drive a slightly cheaper car for one more year. None of these are suffering. They are just arithmetic that most people have never done. The gap between zero and twenty euros a month is almost never as large as it feels from inside the month.

The inflation argument

“In thirty years a million will be worth nothing due to inflation.”

Okay…. Give it to me! I will take good care of it. I promise.

But also, if inflation makes future money worth less, your salary rises with inflation, which means your monthly DCA rises too, which means you automatically invest more in nominal terms over time. The inflation argument for not investing is an argument for investing more. It defeats itself.

The economy is inflationary by design, that’s just how it works. Holding cash is not a neutral position. It is a slow and guaranteed loss of purchasing power dressed up as caution. The index at least tracks the economy doing the eroding. You are not trying to beat inflation. You are trying to stay on the train instead of watching it leave.

The lucky streak problem

Most people who think they have edge are confusing a streak with skill. This works in both directions and it gets dark very quickly.

You can be long something for two years, feel like an absolute genius, write about it online, start a newsletter, gain followers, and then the thing corrects 60% over six weeks and all of that was just the tide coming in.

You can be short something that is obviously overvalued, be completely correct about everything, and still blow up because the market stayed irrational longer than your broker stayed patient. Michael Burry was right about 2008 and almost lost everything being right too early. Being right is not the skill. Surviving being right is the skill. Most people do not have the capital structure, the risk management, or the psychological constitution to do that. Most people, if they are honest, are just on a streak.

Yes, genuine edge exists. But is very very very rare, and you statistically are not in that group ( Don’t worry, i am also not part of the club). For the small minority who have it, the Boglehead approach would be destroying value. But the honest question is: how do you know you are in that minority and not just in a bull market or a lucky run? The market is very good at making you feel like you have edge right up until it removes your capital. A long track record across multiple market cycles with documented risk management is the only real answer. Almost nobody has that.

What I actually do

I have spent years following geopolitics and markets. Pandemics, oil shocks, bond market dynamics, petrodollar mechanics, the strategic logic of conflicts most people read as random. AND YET, My core portfolio through all of it is broad market indices and technology ETFs held on a long horizon with consistent DCA. I also trade occasionally when I think I have an edge on something specific. I am probably just being an idiot when I do that, which is statistically what the evidence suggests about most people who think they have edge. But I keep the trading allocation small enough that being wrong does not matter much.

The analytical work tells me the world is complicated, the macro picture is genuinely uncertain, and anyone claiming to know exactly how it resolves is either lying or mistaken.

We, common twats, do not have access to enough information fast enough to compete with algorithms or insiders who get phone calls before things even happen. The portfolio tells me I do not trust myself or anyone else to navigate that consistently enough to abandon the boring answer. The boring answer has a very good long-term track record. The exciting answer mostly benefits the person selling it.

If you want to see what this looks like in practice, you can check out my public portfolio

This is a follow-up to my first SpaceX IPO piece, which covered the valuation structure, the NASDAQ rule change, the Morgan Stanley conflict of interest, and the Tesla acquisition thesis. If you have not read that one, you can find it here.

This update covers three developments since then that change parts of the picture. The S-1 still has not dropped yet so everything before it is still a pitch.

If you want to follow the portfolio behind the analysis, you can find me here.

Three things that have changed

The SpaceX IPO narrative has moved considerably since my first piece. The valuation has held at $2 trillion. The filing has not dropped. But three specific developments have added new analytical layers that are worth mapping before the S-1 arrives.

The Google deal and what it signals

Google is in talks with SpaceX to secure preferential rocket launch access. Google owns approximately 6.1% of SpaceX, making it one of the earliest and largest outside investors. The deal centres on Project Suncatcher, Google’s internal initiative to put its TPU chips into orbital data centers. Google has already done the foundational work: its TPUs have been tested in space and survived 15 kilrads of radiation in testing, roughly three times the expected five-year LEO exposure at small scale. Most commercial space-rated systems targeting decade-long reliability aim for 50 to 100 kilrads, so this is a proof of survival rather than a proof of readiness. It has also partnered with Planet Labs, providing two prototype satellites to evaluate whether TPUs can operate at commercial scale in orbit.

Google has been working toward orbital data centers for years. Securing preferential launch access before the IPO locks in its position as anchor tenant if the Starship cost curve plays out. The deal also strengthens the IPO narrative in the short term, which benefits Google's existing stake regardless of whether the long-term space thesis works out.

The space data center reality check

The orbital data center thesis deserves more scrutiny than it typically receives. There are three unsolved engineering problems between the current state and the commercial reality the $2 trillion valuation assumes.

The first is launch cost. Current SpaceX Falcon 9 costs run between $250 and $600 per kilogram to orbit. SpaceX is targeting $50 per kilogram at Starship scale, which would require full routine reusability at a level not yet demonstrated in regular operations. The commercial viability threshold for space data centers is estimated at below $200 per kilogram. The gap between current reality and the threshold is significant and requires Starship to perform at a level it has not yet sustained.

The second is radiation hardening. Google’s TPU testing is genuinely encouraging. Surviving 15 kilrads at small scale is a meaningful proof of concept, roughly three times the expected five-year LEO exposure at that scale. However most commercial space-rated systems targeting decade-long infrastructure reliability aim for 50 to 100 kilrads. Google’s TPUs are surviving current testing conditions. They are not yet proven reliable for the kind of decade-long infrastructure play that the orbital data center thesis requires. Scaling from a test chip to a gigawatt data center also introduces compounding failure risks that have not been tested at commercial scale. High bandwidth memory, which modern AI chips depend on heavily, is particularly vulnerable to cosmic radiation bit-flip events. The components that make current AI chips fast on Earth are the same components most likely to fail in space.

The third is heat dissipation. Space has theoretical cooling advantages because the ambient temperature can reach near absolute zero. In practice, heat dissipation in a vacuum requires radiating the heat away from the chips through physical surface area rather than airflow. Current estimates suggest approximately four tennis courts of radiating surface per megawatt of compute. At gigawatt data center scale that becomes an enormous physical engineering challenge with no established solution at commercial scale. This is the most underreported technical risk in the entire space data center narrative.

None of these problems are fatal but all of them require time and capital to solve. The $2 trillion valuation is pricing the solved version. The engineering reality is somewhere between the current state and that solved version, on a timeline that nobody outside SpaceX can reliably estimate.

The XAI signal

XAI, which SpaceX reportedly acquired in an all-stock deal earlier this year (a transaction that remains unconfirmed until the S-1 is public), has signed an agreement to supply Anthropic with 300 megawatts of compute capacity across 220,000 Nvidia GPUs by end of May. This is the detail that deserves the most attention of anything in this update. To be clear: this is the terrestrial Colossus facility in Memphis. These GPUs are on the ground, not in orbit.

XAI is supplying compute capacity to its direct competitor. Anthropic usage at enterprise companies has reportedly quadrupled. On every observable usage metric, enterprise adoption, coding tools, model performance benchmarks, Anthropic is currently ahead of XAI’s Grok. The Cursor arrangement, widely described as a $60 billion deal, is more precisely a $60 billion option. SpaceX can walk away for a $10 billion partnership fee. One framing is an acquisition. The other is a marketing commitment with an exit clause. The XAI deal with Anthropic is a revenue patch on a business burning approximately $1 billion per month. It confirms that XAI’s revenue reality is significantly weaker than its narrative implies, and that the pressure to generate cash before the IPO is real enough to supply the competition.

In my first piece I noted that the $8 billion EBITDA figure circulating in secondary markets predated the XAI acquisition. The Anthropic deal adds another data point to that picture: XAI’s losses are large enough that the business is willing to generate revenue from any source available, including direct competitors. The S-1 will show what that actually looks like in audited numbers. Until then, the XAI burn rate is one of the three most important unknowns in the entire SpaceX valuation.

What this adds to the Tesla thesis

In my first piece I laid out the Tesla acquisition thesis in full. This update adds two pieces of evidence without changing its speculative character.

Every deal made before the IPO has been about derisking SpaceX and improving what the S-1 will show. The Cursor option gave XAI a customer. The Anthropic deal gives XAI revenue. Both reduce the losses that will appear in the filing. A cleaner S-1 supports a higher IPO valuation. A higher valuation on a 6 to 7% float creates the squeeze mechanics the thesis depends on.

The governance fortress

On May 14th, CalPERS and the NYC Comptroller issued a joint letter formally objecting to the SpaceX IPO structure. These are institutions managing almost $1 trillion in assets. Their objections are very specific: Elon controls 79% of votes while owning approximately 42% of equity through super-voting shares, and the reported S-1 draft includes mandatory arbitration for all shareholder disputes, which blocks class-action lawsuits. They are calling this structure novel and extreme and demanding a one-share one-vote sunset before listing.

The mandatory arbitration clause is the detail that connects most directly to the Tesla thesis. It removes the legal mechanism shareholders would normally use to challenge how management deploys the balance sheet. If Musk cannot be removed from SpaceX and shareholders cannot sue as a class, he can use SpaceX’s capital with total autonomy after the IPO. That is precisely the condition the leveraged buyout of Tesla requires.

The institutional boycott scenario

If CalPERS and major pension funds boycott the IPO over governance, the low float squeeze becomes more violent. The buyers left are retail investors and sovereign wealth funds. Neither group has the same governance sensitivity as institutional asset managers. A boycott by the world’s largest pension funds does not kill the IPO, but it changes who owns it after listing and removes the institutional constituency most likely to push back on management decisions post-listing.

Watch how many major institutional names appear in the S-1 book when it drops. If the allocation is dominated by sovereign wealth and retail rather than the major pension funds and endowments, that tells you more about the post-listing dynamics than the valuation does.

Google versus the field

Alaska regulatory filings confirmed Google held 6.1% of SpaceX as of year-end 2025. Following the reported SpaceX-XAI merger in February 2026, that stake is estimated to have diluted to approximately 5.0% of the combined entity. At $2 trillion that is still worth around $100 billion. Google also has a decade of TPU development, proven radiation hardening at small scale, and a core business in cloud, advertising, and enterprise software that generates real revenue regardless of whether orbital data centers work on any particular timeline. Meta has no space plans. Microsoft has no orbital compute strategy. Amazon has Blue Origin but has not demonstrated the same launch cost trajectory. The space bet is a call option on top of a functioning profitable business. For SpaceX, the space bet is the business. That asymmetry is worth holding in mind when the S-1 arrives.

What has not changed

The S-1 has still not dropped. The Morgan Stanley conflict, the NASDAQ float waiver, the low float mechanics, and the $500 billion valuation move on narrative alone are all unchanged. The audited XAI burn rate, the real Starlink ARPU, the debt load across the ecosystem, and what the $75 billion raise is officially earmarked for are all still unknown. The governance objections from CalPERS and the NYC Comptroller are now on the record.

Everything before the filing is still a pitch.

Before you decide I'm an idiot, you should know that I spend my days trying to put more wind turbines on the grid. I'm probably still an idiot ( aren't we all? ), but you should know that before reading the rest. I believe in wind's necessity. But I have come to think that the industry I work in has been used as a rhetorical cudgel to beat down the only partner that can actually help it succeed. That's an uncomfortable thing to say out loud. It's also, I think, where the evidence points.

The question, as it has been framed for most of the past half-century, goes roughly like this: nuclear or renewables? Pick a side. Argue your corner. The framing assumes that choosing one means abandoning the other, and that assumption has quietly done enormous damage. It shaped legislation, redirected capital, and left several major economies dependent on an energy source whose geopolitical risks nobody properly priced until a war in Eastern Europe made those risks impossible to ignore.

Here is what actually happened, and what it cost.

What Nuclear Waste Actually Is

Most people’s mental image of nuclear waste comes from The Simpsons. Glowing green liquid in yellow barrels. Oozing, toxic, ready to leak into the nearest river and produce something with extra limbs. It’s a vivid image. It’s also almost entirely wrong.

High-level nuclear waste is a solid. It looks like small ceramic pellets, about the size of your fingertip, packed into metal rods, sealed inside dry casks that weigh up to a hundred tons and are built to survive a direct hit from a freight train. There is nothing to ooze. All of the high-level waste ever produced by every nuclear plant in the world could fit inside a space roughly the size of a football field. This is what extraordinary energy density looks like from the waste end: the fuel is so efficient that there isn’t much left over.

The long-term storage question, the one that gets raised whenever nuclear waste comes up, has had a demonstrated technical answer for decades. Deep geological disposal, burying sealed casks far below any water table or biosphere, works. We know it works because nature ran the experiment two billion years ago in what is now Gabon, where a natural uranium deposit sustained nuclear reactions for hundreds of thousands of years and the resulting waste moved less than ten metres from the site. Finland is already building a permanent repository. Sweden has approved one. The technology exists, the geology is understood, and the engineering is not exotic. What has been missing is not a solution. It is the political will to build one.

Measured in deaths per unit of electricity generated, nuclear sits between wind and solar. Coal kills at a rate so much higher that stating the comparison feels almost unfair. One in five deaths globally can be attributed to burning inefficient fuels. That number doesn’t generate evacuation orders or HBO miniseries, so it doesn’t generate the same political response. The fear of nuclear wasn’t irrational; accidents happened, radiation is real, and the dread of invisible hazard is a well-documented feature of human psychology. But that fear was never seriously compared against what the alternatives were actually doing to people. That gap in the comparison is where things start to go wrong.

How Fear Became Regulation Became Economics

The death of nuclear construction in the United States didn’t happen because of Chernobyl. It was already well underway by the time Chernobyl happened. The mechanism was quieter than a disaster. It ran through a regulatory principle, written into a document in 1975, that made building nuclear plants financially impossible without anyone quite intending that outcome.

The principle was called “as low as reasonably achievable.” The idea was that any reduction in radiation exposure, no matter how small, had to be pursued if it was technically possible. Underlying this was a model called linear no-threshold, which holds that there is no safe dose of radiation, and that every exposure, however tiny, adds proportionately to your lifetime cancer risk. Under that model, “as low as reasonably achievable” has no floor. It’s permanently open to interpretation, which meant that anyone could challenge any aspect of a nuclear plant’s design, at any time, on radiation grounds.

Between 1973 and 1978, roughly one new rule governing nuclear construction was issued every single day. Plants designed in 1971 spent years retrofitting to requirements that hadn’t existed when they broke ground. A single month of construction delay cost around $44 million, according to Congressional Budget Office estimates. Projects that expected to take five years took twelve, fourteen, seventeen. Some were cancelled entirely. One utility went bankrupt, the first since the Great Depression. The Nuclear Regulatory Commission acknowledged in 2021 that the linear no-threshold model may not actually be scientifically valid. That acknowledgement arrived about forty years too late to help.

By the 1980s the economics had broken. Cost and timeline had become unpredictable enough to make private financing impossible. Three Mile Island caused zero fatalities. Fukushima is more complicated, and worth getting right. No deaths from acute radiation exposure occurred. One worker later died of lung cancer attributed to radiation, though experts noted the probability his cancer was actually caused by his exposure levels was low. The confirmed death toll from Fukushima stands at around 2,300, but these were evacuation deaths: elderly people displaced from hospitals, suicides, people who lost access to medical care during prolonged shelter stays. The earthquake and tsunami that triggered the accident killed over 19,500 people in the surrounding region, which makes clean separation difficult, but researchers have noted that Fukushima’s evacuation-related mortality was higher than in nearby prefectures hit harder by the tsunami, which they attribute to the particular stress of nuclear evacuations. The honest version of this story is not that Fukushima was harmless. Nor is it simply that the harm came from fear rather than radiation, because that framing skips something important. People were told nuclear was safe. Then a reactor melted down on live television. The fear that followed wasn’t irrational; it was the predictable response to a trust that had been broken, by an industry that had spent decades insisting failure was essentially impossible. When you tell people something can’t happen and then it happens, they don’t parse your subsequent actuarial tables very carefully. The tragedy is that the response, measured in policy rather than in immediate deaths, may have caused more harm than the accident itself. Neither accident was used proportionately; both were used to justify conclusions that went considerably further than the evidence supported. The engineers retired. The supply chains dissolved. The knowledge of how to build these plants efficiently faded because the industry had simply stopped building them.

The European Dependency

Europe went through a version of the same shift, with one extra variable. There was a cheap, apparently reliable substitute available to fill the gap: Russian natural gas, piped in at prices that made every alternative look expensive. Nuclear phase-outs were legislated. Plants were taken offline before their engineering lifespans were up. Wind and solar expanded. The political discomfort around nuclear faded. Russian gas covered the baseload that intermittent renewables couldn’t reliably supply.

For a while, the arrangement appeared to work.

February 2022 made the assumptions visible.

When Russia invaded Ukraine and the gas supply was disrupted, Europe discovered that the bridge it had been standing on was a strategic dependency it had never properly priced. The scramble that followed, emergency LNG contracts at spot rates, coal plants brought back online, industrial facilities curtailing output, was the deferred invoice for a deliberate decision made years earlier: close nuclear capacity, replace it with imported gas, trust that the geopolitical conditions would hold. The geopolitical conditions did not hold.

The consequences are still arriving. European energy costs are now structurally higher than those of major competitors. Energy-intensive industries including chemicals, steel, aluminium and fertilizers have contracted, relocated, or both. Several of the plants closed before 2022 were shut down on policy grounds while still within their operational lifespans. That capacity would have mattered. It was removed before it needed to go.

This pattern appears, with local variations, across much of the continent. A sequence of individually defensible decisions whose cumulative result was a vulnerability that only became fully visible when the external condition underpinning it was removed.

The Countries That Kept Going

Not everyone made the same choices, and the contrast is worth looking at. France built nuclear aggressively and ran a partially closed fuel cycle for decades, achieving some of the lowest carbon electricity in the developed world. Sweden combined nuclear with hydropower to similar effect. South Korea kept building continuously and, unlike countries that stopped, kept the industrial learning curves running. Its construction costs stayed manageable while Western costs inflated dramatically, largely because competence in complex engineering tends to erode when you stop practicing it.

China, which had almost no commercial nuclear capacity twenty years ago, has been starting more nuclear projects annually since 2008 than the rest of the world combined. It is also commissioning an experimental thorium molten-salt reactor, a design that addresses several of the core criticisms of conventional technology, representing a genuine advance rather than a marginal improvement.

These countries didn’t ignore the risks. They looked at those risks and then looked at the risk of not having reliable, low-carbon baseload power, and they weighted the second one more heavily. The outcomes diverged accordingly.

One note on a related idea worth handling carefully: nuclear reprocessing, sometimes called nuclear recycling in recent coverage, is genuinely interesting in theory. Spent fuel still contains usable material, and a closed fuel cycle could in principle reduce waste volume and radioactive lifespan. But the technology is harder, more expensive, and carries more proliferation risk than the optimistic framing usually acknowledges. The same chemical process that extracts reusable material from spent fuel is the same process used to produce material for nuclear weapons, which is part of why the countries currently doing reprocessing at scale are largely those with existing weapons programs or ambitions. The case for nuclear in the near term rests on existing dry cask storage being safe, which it demonstrably is, and on conventional reactor construction resuming at scale, which is the tractable problem. Reprocessing is a different and harder conversation.

The Physics of the Mix

Wind turbines produce power when wind blows. Solar panels produce power when the sun shines. These are not limitations that better engineering will eventually solve; they are physical properties of the energy sources. Battery storage is improving fast, but storing enough electricity to cover multi-day weather events at grid scale remains an unsolved problem in most places. In northern latitudes, solar generation in winter can be a quarter of what it is in summer, and the gap between peak demand and available supply tends to widen exactly when filling it becomes most expensive.

Grids need some source of power that runs continuously, independent of weather and season. Geothermal can do this but is geographically constrained. Carbon capture on gas has been “almost commercially viable” for decades without quite getting there. Nuclear runs regardless of conditions, produces no operational carbon emissions, and has a lifecycle carbon intensity lower than most solar configurations once you account for manufacturing energy inputs.

There is also an embedded cost to renewable infrastructure that rarely gets factored into the comparison. Offshore wind installation requires diesel vessels, steel, concrete, copper cabling and rare earth elements, most of which are currently produced using fossil fuel energy. The roads to onshore wind farms, the foundations for offshore platforms, the manufacturing energy for panels: none of this is carbon-neutral. Renewable energy is clean in operation. Building it is not, and that construction period runs for decades. Wind turbines have operational lifespans of roughly twenty to twenty-five years under conventional assumptions, though newer technology and component replacement programs are extending that figure. A nuclear plant runs for sixty to eighty years. These are not equivalent material commitments over time.

Wind and solar belong in this grid. So does nuclear. The framing that positions them as competitors misreads the problem. The grid that decarbonises most effectively uses all available low-carbon sources: wind, solar, hydro, geothermal and nuclear, each contributing what its physics allow.

The Political Time Horizon

There is a simple reason why this fairly obvious conclusion hasn’t translated into obvious policy.

A nuclear plant authorised today in a Western country takes somewhere between fifteen and thirty years to produce its first unit of electricity. Once it’s running, it operates for sixty to eighty years. The entire capital investment is front-loaded. No politician who authorises a plant today will be in office when it opens. The ribbon-cutting belongs to a different government. Given that elections happen every four or five years, the incentive to make a decision whose payoff arrives in three or four administrations’ time is close to zero.

This isn’t a failure of individual courage. It’s a rational response to how political incentives are structured. Long-horizon infrastructure commitments require a political culture that can sustain them across multiple governments, and that culture existed briefly in the postwar decades when the original nuclear buildout happened. It largely stopped existing once the political cost of nuclear association rose and the regulatory environment made the economics look unpredictable.

The compounding effect is significant. Plants that should have been authorised in the 2000s, to be operating now, weren’t authorised. Plants that should be authorised today, to be operating in the 2040s, face the same friction. Every year of delay is another year of fossil fuel dependence and atmospheric carbon accumulation that doesn’t reverse. The bill doesn’t stay flat while we debate; it grows.

The Most Uncomfortable Part

The standard version of this section blames fossil fuel lobbying, and that’s not wrong. Companies whose reserves become stranded assets in a low-carbon energy system have obvious commercial reasons to slow the timeline. That dynamic is real and worth naming.

But the more uncomfortable truth is that the most effective opposition to nuclear over the past fifty years didn’t come from oil companies. It came from environmental movements. The groups most loudly committed to saving the planet were, in many cases, the ones who successfully dismantled the technology most capable of helping them do it. They campaigned against nuclear on safety grounds. They won. And the gap left by nuclear was filled not by wind and solar, which weren’t yet at scale, but by gas. Decades of gas. With all the carbon that implies.

This isn’t an accusation of bad faith. The fear was genuine, the concerns about accidents and waste were sincerely held, and in the context of the 1970s and 80s they were not unreasonable. But good intentions and correct outcomes are different things. The environmental movement’s victory over nuclear is one of the more consequential own goals in the history of climate policy, and the fact that it came from the right side of the argument makes it harder, not easier, to look at directly.

Fossil fuel incumbents are a secondary character in this story. They were the beneficiaries. The protagonists were people who thought they were doing the right thing.

The Forcing Functions

Nuclear is coming back. The mechanism isn’t a shift in public opinion or a sudden policy enlightenment. It’s more immediate: data centers need continuous power at scale, and the grid can’t deliver it from clean intermittent sources alone. AI training requires electricity that doesn’t take breaks. The economics of GPU clusters collapse if they sit idle waiting for the wind to pick up. Technology companies that had made public commitments to clean power around the clock found themselves facing an arithmetic problem with only one available answer. Microsoft is restarting a reactor at Three Mile Island. Amazon has committed substantial capital to small modular reactor development. Google has signed power agreements with nuclear developers. These are operational decisions by institutions with detailed cost models and strong commercial reasons to get the answer right.

Small modular reactors, factory-built units closer in scale to naval submarine reactors than to conventional large plants, offer a potential path around the cost and timeline problems that made large nuclear economically difficult in Western countries. If they can be standardised and mass-produced, the learning curve economics that kept costs manageable in countries with continuous construction programs become available to markets that lost those programs decades ago. The technology isn’t commercially proven at scale yet. The volume of capital moving toward it suggests that several large, well-resourced actors think it will be.

China’s buildout has added a competitive dimension that’s hard for Western governments to ignore. The US ADVANCE Act of 2024, passed with bipartisan support, is the first significant reform to the American nuclear regulatory framework in decades. Several European countries that legislated phase-outs are reconsidering. These reconsiderations are happening under pressure rather than by design, which is more or less how most large-scale policy reversals work.

The Whole Bag

Around 80% of global primary energy still comes from fossil fuels. The entire low-carbon share, hydro, nuclear, wind, solar and geothermal combined, is roughly 16%. The gap between where we are and where we need to be is not one that any single technology closes. It requires every available tool, deployed as fast as the engineering and the capital allow.

There’s a principle in investing that maps onto this cleanly. When outcomes are uncertain, when the downside of being wrong is severe, and when no single asset performs well under all conditions, you diversify. You buy the index. You hold the whole bag. A strong view on which asset is best doesn’t change the calculus; when the stakes are high enough, the cost of being wrong is not something you can absorb. The investor who concentrated entirely in one stock and was right did better than the index investor. The investor who concentrated and was wrong lost everything. When the downside is catastrophic and the outcome is genuinely uncertain, diversification isn’t a hedge against conviction. It’s the only strategy that makes sense.

The energy transition is that problem. We don’t know with certainty which combination of technologies will prove most scalable, most affordable, and most reliable across all geographies over the next fifty years. Battery storage might improve faster than current trajectories suggest. Small modular reactors might or might not work at commercial scale. Geothermal might become deployable anywhere rather than only where geology cooperates. Each of these outcomes is plausible. None is certain. What’s certain is that the cost of the transition failing, of fossil fuel dependence continuing while atmospheric carbon keeps rising, doesn’t reverse.

Looked at from this angle, the decision to phase out nuclear while renewables were still developing wasn’t bold. It was a concentrated bet placed under the impression that the outcome was already known. The European gas dependency was a concentrated bet on a single supplier. The regulatory dismantling of the US nuclear industry was a concentrated bet on a safety model that its own regulators later admitted might not be scientifically valid. In each case, when the assumption underlying the bet changed, the cost landed on populations rather than on decision-makers.

Leaving fossil fuels behind is already one of the harder things our civilisation has attempted. The dependencies run deep, the political economy of transition is hostile, and the timeline is shorter than the scale of the problem would suggest is reasonable. On the question that determines the physical conditions under which everything else happens, running a concentrated portfolio when a diversified one is available is a choice. And the consequences of that choice are already visible in the data.

Please buy the whole bag. We cannot afford to fuck this one up.

Image credit: NASA / ISS069, August 2023.

The document nobody read

For eleven weeks, the dominant narrative has been that the Iran war is impulsive, reckless, and strategically incoherent. Trump bombed Iran on a whim. He thought the regime would collapse. He was wrong. The war dragged on. None of this makes sense.

There is a publicly available document on the US Department of Defense website called the National Defense Strategy. It is not classified. It is not leaked. It was published, distributed, and is available for free download by anyone with internet access. It outlines in clear language what the United States intends to do to maintain global dominance over the coming decade. When you read the war in Iran against that document, almost nothing about it looks impulsive. It looks like a plan being executed with varying degrees of competence.

This piece maps the strategy, tests it against what has actually happened, and identifies where the plan appears to be working and where it has run into structural resistance it did not fully anticipate.

You can read the document here

Why the strategy existed at all

To understand why the document was written, you need to start in February 2022 when Russia invaded Ukraine. The invasion itself was not the problem from a dollar system perspective. The American response was. The US sanctioned Russia, removed it from the SWIFT payment system, and froze over $200 billion in Russian sovereign assets held in Western financial institutions. These were legal actions under existing frameworks. But they demonstrated something that every central bank in the world immediately noticed: the dollar system, which had always been presented as politically neutral infrastructure for international trade, was in fact a political weapon that could be deployed against any nation that held reserves in it. The guarantee of seamless, neutral exchange had been broken.

The consequence was predictable. Russia, China, and Iran began building the architecture for an alternative. Yuan clearing banks opened in London, Dubai, Singapore, Hong Kong, and Switzerland. Iran began selling oil to China in yuan. Russia and China signed commodity deals denominated outside the dollar. The Eurasian land corridor, rail and pipeline infrastructure infrastructure connecting Russia, Iran, and China, began to take shape. If that corridor became operational at scale, American sea power would be partially negated as an enforcement mechanism. The nations of the world would have a credible alternative to the dollar system. And the dollar’s status as the world’s reserve currency, which creates structural global demand for US debt, would begin to erode in a way that would be very difficult to reverse.

The National Defense Strategy is the American response to that trajectory.

The four-point plan

The strategy has four objectives. The first is to secure the Western Hemisphere as an exclusive American sphere of influence. The document frames this as a reassertion of the Monroe Doctrine, the nineteenth century principle that the Americas are an American domain, not subject to interference by outside powers. In practice it means Venezuela, which holds the world’s largest proven oil reserves, was taken. Cuba is next according to public statements. Canada, Greenland, Mexico, Colombia, Nicaragua, and Honduras have all been named explicitly in public communications over the past year. One third of US naval assets are currently parked in the Caribbean. The message to China and Russia is clear: this hemisphere belongs to Washington. If you want access to its resources, you pay a toll or you are blocked.

Venezuela was not only about oil. The Orinoco Mining Arc contains rare earths, coltan, lithium, and strategic minerals that Chinese companies had been extracting and processing at source for over a decade, routing them into Chinese-controlled supply chains. China controls approximately 91% of global rare earth refining capacity. The US was over 50% import dependent on China for materials essential to defense systems, the same materials that go into missile interceptors, AI infrastructure, and precision guided weapons. Russian military advisers, Iranian Revolutionary Guard operatives, and Chinese extraction companies were all operating simultaneously inside the Western Hemisphere. Oil was part of the Venezuela story. Rare earths, strategic minerals, and the removal of rival military presence from the American security perimeter were the rest of it.

The second objective is to restructure alliances. NATO pays more for its own defense. European nations fight Russia in Ukraine with American weapons and financing but with European soldiers. Japan and South Korea take primary responsibility for containing China in East Asia. America provides the strategic architecture and the weapons supply chain, but reduces its direct exposure. This explains the logic of Trump’s NATO demands, which appeared to many observers as an attack on alliances. From a National Defense Strategy perspective, it is a deliberate rebalancing of who absorbs the cost.

The third objective is to economically strangle China by controlling maritime chokepoints. China receives 50 to 60% of its energy from the Gulf. It relies on the Strait of Malacca for access to Southeast Asia, India, Africa, and the Middle East. The first island chain, South Korea, Japan, Taiwan, the Philippines, Malaysia, blocks China’s direct access to the Pacific. Control the Strait of Malacca and you can tax or blockade 90% of China’s energy trade. Disrupt the Gulf and you cut off the supply before it even reaches the Malacca. The Iran war does both simultaneously. A decisive military defeat of Iran is secondary to the primary objective: keeping the Strait of Hormuz disrupted long enough to demonstrate to the Gulf states, to Europe, to Asia, and to China itself that the US controls the global energy supply chain.

The fourth objective is to reinvigorate the American defense industrial base. The Pentagon recently visited Ford and General Motors and told them the US might need them to stop making cars and start making munitions. This was published. The profit margins on weapons for the Pentagon are substantially higher than on consumer vehicles. Companies like Ford and GM have every financial incentive to comply. A sustained war in the Middle East, even one without a decisive outcome, serves the purpose of justifying and accelerating this industrial shift.

Why Iran did not collapse

The strategy assumed Iran would respond the way Venezuela did. The rapid collapse of the Maduro administration, widely attributed by analysts to precision special operations, served as the proof of concept. Decapitation strike, regime collapses, war ends. The logic appeared transferable.

What the strategy did not fully account for is topography and institutional design. In 2003, the US invaded Iraq and won in approximately two weeks using shock and awe, a rapid decapitation of of the central command structure. This worked because Iraq is flat desert. There is nowhere to hide. Air power is decisive in flat desert. Iran is not flat desert. Iran is a mountain fortress. Its terrain is almost entirely mountainous, with the Persian population concentrated in the interior and ethnic minorities in the borderlands. Underground missile bases are built into the mountains. They cannot be destroyed from the air without targeting the mountains themselves.

More importantly, Iran’s military command structure was designed specifically to survive American decapitation strategy. The IRGC operates under what analysts describe as a mosaic command structure: 31 provincial commands, each with its own leadership, supply chain, financing, and operational independence. There is no single head to cut off. When you kill the leadership in Tehran, the 31 provincial commands continue operating. They were designed to do exactly this. American surveillance technology can identify and target leadership. The Iranians solved that problem by distributing the leadership across the entire country and building in operational independence at every level. The ceasefire delays that followed early strikes, when missiles continued firing after ceasefires were announced, were not communications failures. They were the mosaic structure working as intended. There was no central authority that could instantly stop all 31 commands simultaneously.

Iran’s counter-strategy

Iran does not need to win militarily. Its strategy is a war of attrition, what strategists call a game of uncle. Neither side has the capacity to destroy the other. The question is which side accumulates enough pain to force a political settlement on the other’s terms.

Iran’s leverage is Hormuz. By closing the Strait, it does not need to defeat the US military. It needs to keep the global economy under enough pressure that Western governments, facing food inflation, fertilizer shortages, and energy costs, pressure Washington to end the conflict. Iran is also financing the war through the Strait itself, reportedly charging approximately $2 million per vessel for transit through the zone it now controls, which its navy has stated extends to Fujairah. The blockade has become a revenue stream.

The US response has been to blockade Iran’s blockade, establishing its own control zone in the Arabian Sea to cut off Iran’s oil exports through Kharg Island, which handles approximately 90% of Iranian oil. Two competing control zones now exist simultaneously in the same body of water. The US strategy assumes Iran’s economy breaks under this pressure before Western consumers do. Iran’s strategy assumes the opposite. Both assumptions cannot be correct. The fertilizer and food price signals already in the system suggest the timing is tighter than markets are pricing.

The China variable

China has been the most interesting actor throughout this conflict precisely because it has maintained optionality on all sides simultaneously. It buys Iranian oil in yuan, maintaining Iran’s economic lifeline. It has been building yuan clearing infrastructure across the global financial system. It signed 24 trade and investment deals with the UAE in the same week the UAE threatened to price oil in yuan and sell US assets. It controls the rare earth minerals and tungsten the US military needs to rebuild its depleted missile stockpiles.

China is also apparently willing to negotiate with Washington. Trump is reportedly travelling to Beijing in mid-May. The expected framework: China continues buying US Treasuries to finance American debt, and in return gets access to Western Hemisphere energy and the American consumer market. China played this same role during the Cold War, maintaining ties with the Soviet Union while negotiating separately with Washington. The posture is familiar.

A US-China arrangement that reduces China’s financial support for Iran, even implicitly, changes the timeline of Iran’s capacity to sustain the conflict. This may be the negotiation that actually determines when and how the war ends, with everything happening in Islamabad and in the Strait as secondary pressure on that central deal.

The two clocks

Both sides are racing against different clocks, and understanding both clocks is the key to reading what happens next.

Iran’s political leadership believes it can wait out Trump. The midterms are in November 2026. A Democratic win in Congress constrains the presidency and limits the executive’s capacity to sustain military operations. In the worst case for Iran, Trump is gone by January 2029. Iran needs to survive economically and militarily until one of those events creates a political opening. The IRGC does not share this calculation. It is fighting a religious war with no endpoint short of total victory, but the political leadership is operating on a patience strategy timed to American electoral cycles.

The US needs to lock in structural changes before that window closes. Chokepoint control, Western Hemisphere energy dominance, defense industrial base reinvigoration, and the China deal all need to be sufficiently advanced that a change in administration cannot simply reverse them. Trump achieved this with China tariffs in his first term. Biden inherited them and kept them because the structural logic was sound regardless of who was president. The same playbook applies here. Create facts on the ground that are difficult to undo.

The bond market is the third clock that neither side controls. Every time the 10-year yield has approached 4.4% during this conflict, the US military posture has softened. That has happened at least three times since March. Iran has noticed. The financial cost of the war is approaching the threshold where the strategy becomes self-defeating, where maintaining the conflict costs more in Treasury market dysfunction than the strategic gains are worth. The constraint is financial, not military. And the National Defense Strategy, for all its clarity about chokepoints and industrial bases, has no answer for a bond market that prices the cost of the war faster than the battlefield moves. You cannot fight a war with paper assets indefinitely. Eventually the bond market tells you when it is over.

What this means

The war in Iran has structural logic behind it. Read against the National Defense Strategy, almost every major decision of the last eleven weeks maps to one of the four objectives. The topography of Iran, the mosaic command structure of the IRGC, the resilience of the Iranian economy under pressure, and the bond market tripwire have all pushed back against the plan in ways that extended the timeline significantly beyond what the decapitation model assumed. The plan ran into structural resistance it underestimated. That is different from having no plan at all.

The plan is also not obviously failing. China may be moving toward a deal. The Gulf states have been brought to the negotiating table. Western Hemisphere energy dominance is being assembled. The defense industrial base is being reinvigorated. The Strait of Hormuz has demonstrated, for anyone paying attention, that American sea power remains the dominant enforcement mechanism for global trade. Whether you consider that a success or a catastrophe depends largely on where you sit in the global economy and how directly you have been affected by eleven weeks of disrupted supply chains.

What it is not, is random.