Two steps sway from the Apocalypse: the AI crash on US stock markets threatens the very viability of interest-bearing Credit

American voters have approved restrictions on data centers

The US stock market recorded its largest daily decline since April 2025 on November 5. The Nasdaq lost 5.5%, while companies in the S&P 500 index collectively lost $2.9 trillion in market value, a record, according to the Associated Press and Bloomberg, cited by Euronews. The chipmakers' index fell 12.2%, its largest daily decline since 2020. Sell-offs continued on Friday in Asia and Europe.

The decline had begun before the vote. In October, the Nasdaq lost 5.3%, while the semiconductor index fell by more than 15%, amid concerns about the cost of investment in artificial intelligence, according to TradeVille.

In the November 3 elections, voters in 41 jurisdictions across 10 states also voted on data centers, the buildings where artificial intelligence runs. In 40 of them, they approved moratoriums, meaning the suspension of construction permits, or other restrictions, according to a TradeVille analysis published by BURSA. Around 40% of planned data center capacity is located in states whose governors have already promised or imposed a pause, according to Cleanview, cited by Euronews.

In The Theory of Interest, I showed that political decisions do not create fragmentation but give shape to pressure that already existed. Investment costs in artificial intelligence were rising, and their recovery was becoming increasingly uncertain. The vote transformed this economic pressure into a political barrier.

What is lost when data centers are lost

Data centers were where new credit was concentrated. That is why what happens to them is transmitted directly to credit.

Major American cloud companies are investing more than $700 billion in 2026, according to Reuters, cited by 24/7 Wall St. These investments are increasingly financed with borrowed money: the share covered by new debt rose from 9% in 2024 to 32%, while the debt of the five major cloud companies reached around $700 billion, according to FactSet. In turn, new credit depends increasingly on this sector: AI-related issuers accounted for around 18% of investment-grade bonds issued this year, according to Goldman Sachs, cited by Risk.net, meaning that nearly one in five dollars borrowed in this way went toward artificial intelligence.

Through moratoriums, authorities are halting the construction of precisely these assets.

A data center that cannot be built or supplied with energy does not generate the income needed to repay the debt taken on for it.

The risk is not the same for everyone.

Major cloud companies have substantial revenues from other activities and can pay their debts even without the data centers that have been halted.

The risk of default is concentrated among companies that live exclusively on artificial intelligence, such as CoreWeave.

Until now, the sector's profit had been the argument that the order of interest-bearing credit was still working. American semiconductors and software generated a return on capital of 28.3% in 2025, according to data from Aswath Damodaran of New York University, and accounted for 16% of the profits of listed American companies, compared with 4% in 2007.

The profit, however, was partly accounting-based.

The cost of a server is not deducted from profit in the year it is purchased, but is spread over the years in which it is used. The more years there are, the smaller the cost deducted each year and the higher the reported profit.

Microsoft and Oracle spread the cost of servers over up to six years. Nvidia, however, releases a new generation of chips every year, and critics consider a chip useful for one to three years, according to Yardeni Research.

If they are right, the actual annual cost of servers is two to six times higher than the amount deducted from profit, and part of the reported profit exists only in accounting.

How state-allocated credit works: China

China has also recently experienced a collapse in artificial intelligence shares.

The STAR50 index, which tracks Chinese artificial intelligence and semiconductor companies, fell 40% in less than four months, Reuters points out.

Chinese credit, however, had already been declining because of the real estate crisis.

Banks are not refusing credit, but demand for credit has remained weak.

In July, the Chinese repaid 340 billion yuan more than they borrowed, according to Caixin.

The state partly offset the decline in private credit demand by directing financing toward the companies it supports.

Listed Chinese non-financial companies earn 6.2% on capital, below its cost of 10.9%, according to Damodaran's data, yet they receive credit at an average interest rate of around 3%, according to the People's Bank of China's monetary policy report. The real estate developer Vanke, which lost 88.6 billion yuan in 2025, is paying its debts with the support of the state-owned company Shenzhen Metro, Caixin notes.

In China, credit has not stopped, but who receives it is no longer decided solely according to the interest-rate criterion.

The state imposes priorities and supports debtors who do not cover their cost of capital.

What the Apocalypse is

In The Theory of Interest, I showed that the order of the world over the past five hundred years is founded on interest-bearing credit.

Interest determines what deserves to be financed.

Anyone who borrows to earn from a difference in price between two markets or in returns between two sectors earns only if the difference covers the interest and the costs of the operation.

In the book, this operation is called financed arbitrage.

Credit is not granted only for arbitrage.

It is also granted for productive investment, consumption, refinancing old debts or government expenditure.

For investment and arbitrage, interest is the threshold: a project that brings in less than the interest is no longer worth undertaking with borrowed money.

In the book, I showed that financed arbitrage is not a marginal activity but the engine that moves capital. Through it, interest determines which economies and which sectors receive the money.

I also showed that, as profit differentials narrow, credit changes its function.

It no longer directs capital toward production but supports demand, refinances existing debts, sustains asset prices and balances budgets.

This is what happened after 2009, when debt moved from companies to the state.

Therefore, when financed arbitrage becomes blocked, production does not stop, but credit allocation loses the principle that organized it.

This order operates only between two limits: too much integration and too much fragmentation.

When the world integrates too much, prices equalize and there are no longer differences from which to earn through arbitrage.

When the world fragments too much, the cost of moving between economies exceeds the gain.

In the book, I described three bottlenecks of financed arbitrage:

The barrier bottleneck (B) occurs when legal and political barriers raise the cost of moving faster than the gain increases.

The lack-of-profit bottleneck (A) occurs when profit differentials no longer cover interest and costs.

The debt bottleneck (C) occurs when debt payments can no longer be covered by the income of those who owe them.

Interest-bearing credit has gone through many crises: 1929, the 1970s, 2000 and 2008.

Each time, the way out came through one of two paths: cheaper credit or the emergence of new profit differentials.

In the book, the second path is called self-regeneration: when differences are exhausted, fragmentation creates new ones.

Self-regeneration, however, is not guaranteed. It works only if the profit differentials created by fragmentation increase faster than the cost of moving between economies.

This is where the meaning I give to the Apocalypse begins: the moment when credit can no longer be directed according to interest, and the decision about who receives the money may pass to the state.

It is not the end of credit: in China, credit continues. It is the end of the order in which who receives the money was decided according to the interest-rate criterion.

The first bottleneck, through barriers, has begun

When price and return differences between countries narrow, credit moves toward differences in returns between sectors.

Artificial intelligence was now the largest sectoral differential among them: a return of 28.3%, more than three times the average return of American non-financial companies, at 8.7%.

That is why new credit became concentrated there.

Through moratoriums, voters did not equalize this gain but raised the cost of reaching it.

Each data center now has to go through approvals, delays and the risk of being halted.

Around 30% of the additional capacity planned for 2028 is at risk, according to a UBS analysis cited by Finimize.

For halted projects, the barrier bottleneck has already occurred: no gain can cover the cost of a data center that cannot be built.

For the artificial intelligence sector as a whole, it is not yet known whether the bottleneck has occurred. There are no public figures on how much moratoriums have added to data center costs, so that this can be compared with the sector's 28.3% return.

In the book, I showed that barriers do not only raise construction costs but also interest rates.

Anyone lending money for a project that may be halted demands a higher risk premium.

According to Goldman Sachs, AI-related issuers were already paying a premium of around 1.15 percentage points above government bonds in September, compared with 0.78 percentage points for the rest of investment-grade bonds, as reported by BNN Bloomberg.

For companies that live exclusively on artificial intelligence, the premium has risen much more. Insurance against CoreWeave's debt default reached more than 10% a year of the insured amount in October, and its bonds maturing in 2032 were trading at 78% of face value, Reuters reports.

Some creditors no longer want this risk at all.

The Dutch bank ING is suddenly putting up for sale a portfolio of around $500 million in data center loans, according to Bloomberg, cited by Euronews, meaning it prefers to sell the loans rather than hold them until maturity.

Some projects can move to other states, but the debt taken on for those that have been halted remains, and the income from which it was supposed to be repaid becomes uncertain.

Restrictions can, however, increase the return of approved projects, because capacity supply falls.

Money has begun to move.

In October, energy-sector shares rose by around 13%, while healthcare, utilities and consumer staples shares rose by around 10%, TradeVille also points out.

It remains to be seen whether any of these sectors will generate a return above interest comparable to that of artificial intelligence.

If not, new credit will no longer have anywhere to go according to the interest-rate criterion.

The second bottleneck, through lack of profit, is two steps away

How close American companies are to the lack-of-profit bottleneck can be seen from the difference between what they earn on capital and what they pay in interest. I call this difference the reserve.

In 2024, American non-financial companies earned 8.7% after tax on their entire capital, according to the US Bureau of Economic Analysis.

The Baa interest rate, paid by companies with the lowest rating in the investment-grade category, was 6.47% in September, according to Moody's.

The reserve is therefore 2.23 percentage points.

The reserve measures the tension between capital's return and the cost of credit.

It is not a mechanical threshold.

The reserve compares the average return on existing capital with the interest rate, so it does not directly show how much new investments generate.

Since 2004, the reserve has been negative only in 2008, at -0.45 percentage points, and in 2009, at -0.90 percentage points, precisely the years in which American credit stopped.

Since 2010, it has remained positive, between 1.86 and 6.41 percentage points. A negative reserve and the halt in credit therefore occurred together, which does not mean that one caused the other.

In the book, I described two thresholds.

Below the first, the return no longer covers interest and costs, and financed arbitrage stops.

The second adds a minimum safety margin.

Between the two, the return still covers interest, but the remaining margin is too small. A small increase in interest or a small decline in profit cancels it out. To earn the same amount, borrowers take on more debt and more risks. The system works, but each shock brings it closer to a bottleneck.

That is why the system weakens before the reserve reaches zero.

This is what happened before the 2008 crisis. The reserve fell from 1.54 percentage points in 2005 to 1.32 in 2006 and 0.82 in 2007. It was still positive, but the remaining margin could not cover the decline in profits and the increase in interest in the following year, and in 2008 the reserve became negative.

First step: losses move from the stock market into corporate credit. When share prices fall, companies find it harder to raise money by selling new shares and have to borrow more. At the same time, creditors consider the risk of not being paid greater and demand higher interest on bonds and loans. The weakest companies end up unable to pay their debts.

Signs have appeared beyond artificial intelligence as well. The difference between the interest rate on high-risk corporate bonds and that on US government securities rose to 3.4 percentage points on November 5, its highest level since the beginning of 2025, according to Saxo.

American banks, however, have not tightened their lending conditions.

In the Federal Reserve's October survey, conditions for loans to large companies remained practically unchanged. Only demand declined, because companies are investing less.

In the book, I showed that this is what financing looks like when the economic space fragments.

The creditor becomes more selective, and the projects worth undertaking with borrowed money become fewer.

Second step: the profits of all companies decline. The step has two parts:

a) a decline in artificial intelligence profits and

b) a recession in the rest of the economy.

For now, share prices have fallen, not profits. Third-quarter profits actually increased by 29.5%, and 83% of the S&P 500 companies that had reported by October 23 exceeded estimates, according to TradeVille.

Artificial intelligence profits account for 16% of the profits of listed American companies.

If this sector's return falls from 28.3% to the median of the other sectors, 14.8%, its profit declines by around 48%, and total profit by around 7.6%. If the profits of all non-financial companies decline by the same proportion, their average return of 8.7% falls by around 0.66 percentage points. This figure is an approximation, because the two groups of companies are not the same.

After the elections, restrictions on data centers made this decline more plausible.

A recession like the one in 2009 would add a decline of 0.9 percentage points: that is how much the return of American non-financial companies fell between 2007 and 2009, from 7.3% to 6.4%.

The scenario assumes that this decline affects the rest of the economy, in addition to the loss from artificial intelligence.

After the two profit declines, the reserve would remain at around 0.67 percentage points.

For it to disappear, the Baa interest rate would have to rise by around 0.7 percentage points. For comparison, it has risen by that much from 2024 until now.

Neither step is sufficient on its own.

Together, however, they can produce the lack-of-profit bottleneck.

The third bottleneck, through debt, is approaching at the state level and in China

American companies are not overindebted.

Debt payments take up 36.9% of their income, compared with 44.6% at the end of 2008, according to the Bank for International Settlements.

Debt has moved elsewhere.

The issuance in 2009 did not reduce it but moved it.

In the US, the state took it over from companies and households: its debt rose from 62% to 117% of GDP.

Globally, emerging economies took it over, primarily China, where debt doubled.

Overall, the debt of economies reported to the Bank for International Settlements increased from 198.5% to 242.5% of GDP, according to its data.

The American state can still borrow, but at a higher cost.

In 2009, the interest rate on its 10-year bonds averaged 3.3%, while now it is around 4.8%, Saxo also shows, on nearly twice as much debt.

In China, the debt bottleneck is closest. Companies and households spend 19.1% of their income on debt, more than ever before, according to the Bank for International Settlements.

If the losses from artificial intelligence were taken over by the state again, as in 2009, they would be added to this debt.

How the three bottlenecks are connected

In the book, I showed that the bottleneck does not come from a single place but from the way the three bottlenecks sustain one another.

Barriers raise the interest demanded from artificial intelligence companies.

Higher interest reduces the reserve between return and interest, thus bringing the lack-of-profit bottleneck closer. The same interest increases payments on new and refinanced debt, thus bringing the debt bottleneck closer.

Debtors who pay more on old debts have less money for investment, and credit demand falls.

That is why a bottleneck that begins in a single sector does not necessarily remain there.

The 2008 remedy has less effect

In The Theory of Interest, I showed that when interest rates fall, more arbitrage operations become profitable for a while. Among them are operations that earn only slightly above interest, and a small increase in interest or a small decline in earnings makes them unprofitable again. That is why the system becomes more fragile.

This is how the 2008 remedy worked. Central banks issued money, bought bonds with new money, interest rates fell, and companies and states were able to borrow cheaply again.

The remedy worked for three reasons.

First: prices did not rise because of the new money, since the recession had brought American inflation down from 5.5% in July 2008 to zero in December, and oil had become cheaper.

Second: most of the new money remained in banks' accounts at the Federal Reserve, because credit demand was weak.

Third: China was buying American bonds, and the American state could borrow cheaply.

Now, American inflation is 3.7%, and excluding energy and food it is 2.6%, according to the US Bureau of Labor Statistics.

The Federal Reserve is doing the opposite of issuing money.

It raised interest rates in September and kept them unchanged on October 28, when two committee members voted for another increase, while the market estimates a probability of around 85% that interest rates will rise in December, TradeVille reports in BURSA.

In a deep recession, demand would fall, and inflation would fall with it, as in 2008.

Then the Federal Reserve could lower interest rates again.

The remedy, however, would come after the damage had occurred, not beforehand to prevent it.

No recession would bring all inflation down, because part of it does not depend on demand. Brent oil fell after the elections to around $100 per barrel, according to Saxo, but costs around half as much again as in November 2025, because the Strait of Hormuz remains closed. As long as oil remains expensive, the Federal Reserve would find it difficult to lower interest rates, even in a recession.

In the book, I also showed a deeper limitation of the remedy. The central bank can lower interest rates, but it cannot create profit differentials where they have been exhausted.

In 2009, cheap credit had somewhere to go: toward emerging economies, primarily China, and later toward artificial intelligence.

Now, a lower interest rate would find fewer profit differentials to exploit.

The buyer from 2009 has withdrawn. China reduced its holdings of American bonds to $590.7 billion in August, the lowest level since 2008 and 55% below the 2013 peak, according to US Treasury data cited by SCMP, and is keeping its reserves in gold, which it has been buying for 23 consecutive months, according to Kitco. Without this buyer, a new issuance would make credit cheaper for the American state less than it did in 2009.

An issuance that raises prices shifts losses onto everyone who holds money.

In the book, this monetization is one of the mechanisms that takes the place of arbitrage after the bottleneck.

It is already a form of the transition to the state.

Why the system no longer regenerates itself

A negative reserve has occurred before, and credit recovered.

In 2009, the reserve was -0.90 percentage points, and by 2012 it had reached 3.47 percentage points.

Of this 4.4-point recovery, around 2.4 points came from the decline in the Baa interest rate, from 7.3% to 4.9%, amid monetary easing.

The rest came from an increase in company returns.

The cyclical adjustment was therefore made mainly through cheaper credit.

Now, cheaper credit would come later and would have less effect.

The second way out, self-regeneration, works in the opposite way from the past.

Moratoriums are also a form of fragmentation, but they do not create a new difference. They close a difference that already existed within the same economy, thus preventing self-regeneration rather than producing it.

Fragmentation between blocs also creates new differences, for example in oil, since the Strait of Hormuz has been closed.

In the book, however, I showed that fragmentation simultaneously raises the threshold: barriers increase the risk premium, and therefore interest rates.

If the threshold rises at least as fast as profit differentials, fragmentation no longer restores the gain.

The self-regeneration bottleneck becomes visible when risk premiums rise faster than price differences.

For fragmentation between economic blocs, public data do not yet allow this comparison.

Therefore, both paths through which interest-bearing credit emerged from past crises, cheaper credit and the emergence of new differences, are now more difficult to use at the same time.

A cyclical adjustment means that credit declines and then recovers through one of the two paths. If neither works anymore, the reserve is no longer restored, and credit can no longer follow the interest-rate criterion. Its allocation then becomes available to pass to the state. If it passes, it returns to the market with greater difficulty than after 2010. This is how a cyclical adjustment can become the end of the order of interest-bearing credit.

Why the state

In The Theory of Interest, I showed that, after the bottleneck, capital allocation passes to mechanisms that no longer depend on arbitrage: administrative decisions, fiscal coercion, capital controls, directed credit or monetization.

The other paths are bankruptcy, in which debts are reduced through losses; restructuring, in which debt terms are changed; and equity, through which the person providing the money shares the risk with the person using it.

In 2008, the American state had already made a first partial transition.

It chose which banks would survive: it allowed Lehman Brothers to go bankrupt and rescued AIG.

It placed Fannie Mae and Freddie Mac under its control, and they have since guaranteed most new mortgage loans.

The Federal Reserve bought around $1,250 billion in mortgage bonds, thus deciding that the new money would go toward housing.

General Motors and Chrysler were rescued with public money.

Banks continued to decide whom to give each loan to, but the state decided who remained in the market, which sectors received money and how much money cost.

The transition of 2008 could be partly reversed after 2010, because issuance restored the reserve and the market resumed allocation.

I believe a new transition to the state is the most likely of the paths, for three reasons.

First: the precedent of 2008.

Second: new credit became concentrated in a single sector, artificial intelligence. When a potential loss is concentrated among a few large debtors, the state intervenes more easily.

Third: if the remedy has less effect and self-regeneration no longer works, the reserve is no longer restored as quickly, and the transition to the state becomes harder to reverse.

In China, we can see where a transition that is no longer reversed leads.

What the transition of credit to the state brings

Politically, the decision about where money goes passes from banks and the market to the government.

Access to credit no longer depends on how good the project is, but on proximity to those who make the decisions.

Through voting, voters end up deciding the fate of entire sectors, as was seen on November 3.

Financially, interest is set by decision, not according to the risk and return of the borrower.

For the state to be able to provide cheap credit, deposit interest rates are kept low, and savings are prevented from leaving the country through capital controls.

Economically, companies that do not cover their cost of capital survive, but are supported because the state considers them strategic.

Investment accumulates in politically chosen sectors: defense, energy, chips.

The return on capital falls, because the choice is no longer made according to profit.

Socially, losses pass to those who did not take part in the transaction: depositors, taxpayers and, through inflation, everyone who holds money.

Society is divided between those who receive credit from the state and those who pay for it.

Morally, the old order had a rule: the borrower is responsible, and the contract is the measure.

In the transition to the state, the decision-maker no longer bears the loss. Interest can no longer be justified as the price of risk, because others bear the risk.

So far, I have measured how close the order of interest-bearing credit has come to its own limit. What comes after it can no longer be measured, because it depends on how society is organized: who bears the risk of financing and who bears the loss.

In the book The Constitution of Freedom E (BURSA Supplement, 2026), I described a different human organization, in which risk remains with those who decide on financing.

In the order of interest-bearing credit, capital is paid through an obligation that is owed whatever the outcome of the financed activity.

The lack-of-profit bottleneck and the debt bottleneck arise from this: credit stops when the result no longer covers the obligation. The barrier bottleneck also stops projects financed with equity, but in that case the loss remains with the person who provided the money, without a debt that still has to be paid.

Article 87 of The Constitution of Freedom E prohibits remunerating capital through obligations that accumulate independently of the result.

Article 88 requires risk and the outcome to be shared proportionally between the person providing the money and the person using it, without losses being shifted onto others or onto the community.

The rule concerns the form of the contract, not the destination of the money. The state does not decide who receives capital.

It only decides that risk cannot be shifted onto those who did not take part in the transaction.

In this organization, the person providing the money earns when the financed activity earns and loses when it loses.

There is no longer a fixed interest payment that the outcome cannot cover.

State-allocated credit meets none of these criteria: the obligation remains fixed, and the loss passes to others.

If artificial intelligence companies can no longer earn enough to sustain interest-bearing credit, and the state takes over its allocation, the question remains: what order will take the place of the one founded on interest-bearing credit?