Why Ray Dalio is Worried About an AI Bubble
Dalio’s central warning is that transformative AI can still be a terrible investment at the wrong price. Previous breakthroughs—from household electricity to radio, cars, and airplanes—attracted capital because their potential was real, yet “everything has its price.” Investors must distinguish good technology from good investment.
The bubble mechanism begins when paper wealth grows much faster than available cash. A company can raise $50 at a $1 billion valuation, instantly creating nominal wealth, but “wealth is not money”: spending requires owners to sell assets or borrow. Taxes, IPO liquidity, frightened holders, or debt obligations can therefore force selling long before AI’s ultimate value becomes knowable.
Dalio’s bubble indicator puts current conditions in the same order of magnitude as 2000 and 1929. He characterizes 2008 as a smaller, more debt-intensive housing bubble; today’s danger is unusually concentrated across GDP, wealth, equities, bonds, trade, employment, and geopolitics. His portfolio conclusion is blunt: “Don’t do everything AI,” whether that means going all-in or taking a leveraged short.
The sovereign-debt backdrop is the “plaque” that could turn an AI correction into an economic heart attack. Dalio’s approximate arithmetic has Washington spending $7 billion against $5 billion of revenue, carrying debt equal to roughly six times receipts, paying about $1 trillion of interest, and refinancing approximately $10 trillion annually. Meanwhile, foreigners buy about one-third of U.S. debt, and demand from China, Japan, and others is weakening.
A productivity miracle only solves the debt problem if it becomes taxable income quickly enough. Thompson’s bull case is a 10-fold productivity era powered by “geniuses in the data center”; Dalio concedes that sufficient income growth would work, but insists investors trace the whole chain from productivity to wages or profits, government revenue, distribution, and timing. Competition may instead make services cheaper without transferring the productivity gain into enough income.
Dalio’s fiscal prescription is to reduce the deficit below 3% of GDP before the ship “hits the rocks.” He sketches a politically neutral mix—perhaps raising taxes and cutting spending by roughly 4%–5% each—while ensuring AI succeeds and benefits most citizens. Government ownership of AI equity does not itself solve the cash problem: “I’m trying to buy a banana,” he jokes, and a small Anthropic stake is not a dollar.
China’s advantage may lie less in winning the frontier-model contest than in deploying AI as infrastructure. Dalio says Chinese leaders treat AI like electricity, water, roads, or railways, then integrate cheaper systems into an industrial base representing 32% of global industrial production—more than the United States, Germany, and Japan combined. That deployment supports productivity, export competition, economic warfare, and military power.
1. AI amplifies five forces that remain inseparable
Dalio rejects the premise that technology has somehow displaced the other forces governing history: money and debt, domestic political order, geopolitical order, natural events, and human ingenuity. His organizing claim is that “everything is mechanical”—outcomes have causes, and AI changes their intensity without abolishing their underlying relationships.
Debt and politics reinforce each other across long cycles. Postwar settlements can write off obligations and create a relatively clean slate, as after 1945; debts then grow relative to income while wealth and values diverge, producing greater polarization, populism, and internal disorder.
The international cycle is also turning. Dalio argues that the U.S.-built multilateral order—embodied by institutions such as the United Nations, World Bank, World Court, and World Health Organization—is giving way to a world where “power matters,” while climate shocks and a possible “super El Niño in a year or two” add another source of instability.
2. Bridgewater’s original AI encoded causes, rules, and disagreement
Dalio traces AI as a concept to 1956, then describes his own method: articulate decision criteria, backtest them, and computerize them. One rule might combine rising initial jobless claims with rising commodity prices; if growth was high and inflation was rising, the system could test whether short bonds beat long bonds.
The advantages were portability, speed, complexity management, and reduced emotion. Once defined, the same criteria could be applied across countries and markets, turning investing into something like “building a factory that has no workers” or programming a computer chess game.
Dalio compared the system’s proposed trade with his own judgment. When they differed, he says roughly 65%–70% of the time he had forgotten—or failed to program—some complexity; about one-third of the time, the disagreement revealed something new that made him say, “I’d better systematize that.” Over time, human and machine converged.
Thompson credits Bridgewater with a 9% return during the 2008 crash and asks whether its system contained 100 or 100,000 rules; Dalio answers that it was “hundreds.” His democratized version requires no hedge fund: pause after a decision, record the criteria on an iPhone, and test whether those principles would have worked historically.
3. Genuine innovation becomes a bubble when wealth outruns cash
The recurring pattern begins with an unmistakable breakthrough. In the 1920s, people saw household electricity, refrigerators, lights, movies, radio, cars, and airplanes and reasonably concluded, “This is obviously something to bet on.” The technology became real, but enthusiasm did not answer whether its market price was too high.
Companies face an asymmetric choice under radical uncertainty: invest too much, or risk certain defeat by failing to “step on the gas.” Investors then confuse technical promise with financial return, supplying borrowed money and equity without knowing whether future profits can justify the capital committed.
Dalio’s cleanest example is a company raising $50 at a $1 billion valuation: almost no cash has entered, yet a billion dollars of paper wealth appears. That wealth cannot fund consumption until somebody sells it or borrows against it, so a wealth tax, IPO, debt payment, house purchase, or yacht purchase can force assets onto the market.
Whether the resulting decline becomes a crash depends on ownership. Strong hands can absorb losses; weak or leveraged hands may panic or need cash. Dalio says his indicator places today near 2000 and 1929 in magnitude, while Thompson identifies IPOs, taxes, geopolitical conflict, climate shocks, and rising bond yields as “big pins”; Dalio agrees: “It seems so to me too.”
4. Debt may demand cash before AI delivers its productivity miracle
Thompson’s pushback—worth keeping—is that the boom has already survived tariffs, war with Iran, and other shocks. Why could the same supporting dynamics not persist for three more years? Dalio answers with the smoker who has avoided a heart attack: “Please don’t take comfort in the fact that you haven’t had a heart attack yet.”
His fiscal X-ray shows roughly $7 billion of annual federal spending against $5 billion of receipts, a deficit near 40% of revenue, debt around six times receipts, and interest consuming about half the deficit. Beyond approximately $2 billion of new issuance, about $10 trillion of maturing debt must be resold each year.
Supply is therefore rising as demand weakens. Foreigners account for roughly one-third of U.S. debt purchases, but major buyers including China and Japan have their own constraints; governments also worry that geopolitical conflict could leave their reserves vulnerable to sanctions, as happened to Russia.
Dalio concedes Thompson’s “new treatment” for the patient: if AI raises incomes enough, debt becomes serviceable. But the gain must become household or corporate income, then government revenue, on the required timeline; he cites about $280,000 of national debt per person in the average household. “The miracle of productivity does not carry over to income” automatically, because competition can push prices down.
5. Fiscal repair and strategic AI require an active state
Dalio’s “3% solution” is to reduce the deficit below 3% of GDP so the plaque stops accumulating. He is indifferent to whether policymakers turn left or right—perhaps taxes rise 4%–5% while spending falls similarly—but says both parties instead promise voters that taxes will not rise and benefits will not be cut.
Strategic competition makes pure laissez-faire inadequate. AI can determine economic output, military capability, and “world domination,” so government may need to set direction; however, Dalio says a sovereign equity stake does not automatically create spendable resources. If the government buys a stake, it needs cash, and he says it borrows; the stake itself is not a dollar for buying goods.
China’s model illustrates the alternative: treat AI as broadly available infrastructure rather than primarily a profit center. Dalio argues that its largest advantage may come from embedding simpler, cheaper AI in manufactured goods across 32% of global industrial production—not merely reaching the frontier’s highest capability.
6. Personal adaptation still begins with explicit principles
Dalio is trying to turn his investment knowledge into an AI platform combining education with structured decision-making. He wants someone starting as he did at age 12 to gain access to investing knowledge at a high quality level without an institution or expensive resources—only criteria that can be examined, followed, and improved.
His advice to young people changes in one respect: “You have to be on the crest of the wave, otherwise you’ll drown in the waves.” The deeper principles remain stable—know your nature, experiment, make work and passion one where possible, remember the financial dimension, and treat life as an adventure.
The governing learning loop is “pain plus reflection equals progress.” Principles define both what someone wants and how they will pursue it; mistakes supply the pain, but only reflection converts the experience into a better rule for the next decision.
Full transcript
We have a whole series of dynamics that seem to me like big pins heading toward the bubble.
It seems so to me, too.
Good. So you have a very high concentration of everything you bet on.
In other words, its share of GDP, its share of wealth, its share of the stock market, what it means for the bond market, what it means for trade, what it means geopolitically, for war, and what it means for employment. There’s a lot going on about this one.
1. Why the bubble hasn’t popped yet
Hi, I'm Nicholas Thompson, CEO of The Atlantic, and this is the most exciting thing about AI. A few weeks ago, I had Azim Azhar on the show, and we were talking about one of the biggest questions in AI. Are we in a bubble? Technology is amazing, but that doesn't mean the economy is working. And if you listened to that episode, you know that Azim's response was, “No, no, it's not a bubble.” We worked through a whole series of arguments. I seem to have proven that this is so. He rejected every one of them. But, of course, not everyone shares Asim's views. One notable skeptic is Ray Dalio. He is the founder of Bridgewater, the world's largest hedge fund. Ray is known for predicting the 2008 financial crisis, which meant that when the housing bubble burst and unleashed economic chaos, Bridgewater's investments actually rose in value, as they have done virtually every year since the fund's inception. He left the firm last year. He now shares his insights through his newsletter, Principles of Perspective, as well as through books such as How Countries Fail. If you want to gain sound knowledge about markets and investing, and what happens when an asset is overvalued, Ray is a good person to get to know. For the past year, he has been sounding the alarm about artificial intelligence. He said it was a bubble, but not for the reason you might expect. So I caught up with Ray to hear why and how we can avoid another cycle like 2008. This podcast is from my team on the business side of the Atlantic and is independent of the Atlantic editorial team. Here's my conversation with Ray Dalio.
Ray Dalio, welcome to the hottest thing in AI.
I am passionate about the most interesting things in AI. This is what we will try to discuss.
2. The forces that form bubbles
Okay, let's just start your grand theory, the five forces that govern history, debt, domestic politics, world order, natural disasters, technology. You describe a story with its ebbs and flows. My first question is: are we at a point where technology, the fifth force, is becoming more powerful than the other four, or are they in the same balance they have been in historically?
No matter how you look at it, they are interconnected.
Yes. And I just want to touch on this, okay?
Please.
The first force is money. Debt—dealing with expenses and capital—is very important in AI. Is this a bubble? Will this work or what? So where are we with this first force? Do we have a debt bubble? What will happen?
Okay, that’s a lot of questions, isn’t it? Every listener will know the answer to this conversation. That’s why I wrote this book. Because it’s mechanical. It’s still mechanical.
When you’re in debt, if you hold a bond, you expect to get your money back. And if you’re a debtor, you have to pay, right? That doesn’t change with artificial intelligence.
You answered all my questions, Ray.
I wouldn’t think about it that way. Secondly, the force’s mechanics are mechanical. My main point is that everything is mechanical. Everything that happens has reasons that make it happen, and if you understand the mechanics, then you understand how it works. That’s important.
Artificial intelligence doesn’t change that. It will influence it. But the second force is politics—internal order. There’s a political order, and there’s a social order. The more you move left and right, the more conflict you have.
If you have a debt problem, then you have a problem. These things get into big cycles. There’s a debt cycle, and there’s a political cycle. When you come out of a conflict, you write off the debts, a war starts, and you start everything from scratch, like it was around 1950 or after 1945. You have a blank slate, and you start the cycle.
Then debts grow relative to income, and there’s politics, which is related to the wealth gap. This second force—the gap in wealth and values—means that when this cycle happens, it leads to more disorder because there’s more polarity, populism, and so on. So this is the second force, very much related to the first force, which is debt and money.
And we’ll move on to another—the third force, geopolitics. In other words, the second force is how we treat each other within the country.
Yes.
The third force is how countries behave toward each other. There is a world order, and that has a big impact, doesn’t it? We’re moving away from what was once a multilateral world order created by the United States after the war in 1945.
It was almost a copy of democracy, but there were the United Nations and these multilateral institutions, the World Court, the World Bank, and the World Health Organization. It’s over, isn’t it? We’re now moving toward what the world was like before, where power matters.
Yes, and this leads to more conflict. So that’s the third one.
Fourth, in no particular order, is natural disasters.
We must realize that more people have died from natural disasters throughout history, and nature matters. Natural disasters, and of course climate, are a problem. Probably, as we approach a super El Niño in a year or two, we’ll see that it matters.
Then the fifth force is human ingenuity and, in particular, new technologies. AI matters, but it’s very closely tied to these other forces. So we can’t look at it in isolation, can we?
How much money do they get? Do they make money? Which ones are political, and which ones are geopolitical? I’m sure that when we talk about AI, we’ll talk about China. We’ll talk about what this means in terms of economics and war.
These are the five forces, and they tend to operate cyclically because the first three are very cyclical.
That’s a good answer, Ray, because I wanted to answer each sentence with a different question. But let’s do this: I want you to go back in time a little, and then we’ll get to where we are in the debt cycle, where we are in the political cycle, and how AI will impact that.
3. The rules-based system Ray used to run Bridgewater
You remember Bridgewater, the most successful hedge fund in America, and you ran it for 40 years?
About 40 years. I controlled it for 40 or 50 years, and I probably ran it for 48 years.
What’s interesting and relevant to this conversation is that you’ve been extremely successful using a form of artificial intelligence and, obviously, a form of high technology, but very much a rules-based system.
As I understand it, given the way you set it up, you said, “These are the principles. These are what I believe in. These are the rules.” Explain how Bridgewater used AI, and let’s talk about some of the differences between that and how AI works now.
The term AI came about, and the concept emerged in 1956. It had different forms. When I created Bridgewater, I thought, “What are my criteria for making decisions?” I would define them, go back and test them in retrospect, and then computerize them.
I’ll give you an example of a rule: if growth, as measured by initial jobless claims, was rising, and then I combined that with commodity prices as a measure of daily inflation, if growth was high and inflation was rising, would it be a good decision to have short bonds instead of long bonds?
So you define these rules, you define these systems, and you build them.
Continue.
I learned about the possibilities of doing it this way. First of all, if you have these rules, you can apply them everywhere. For example, if I were thinking about what I was going to do or what country I wanted to invest in, I could apply these criteria everywhere.
Complexity could be handled much better. The speed was much better. Fewer emotions influenced decisions, so I could make much better decisions.
It was more like defining criteria, researching them, programming them, and building a factory that had no workers. The old way, which I think was wrong for everyone, was to weigh things in your head and then think about what they were. Instead, you define them, program them, and test them.
It was like creating a computer chess game. These are my criteria and how I would play chess. The system would deal with the world, and then I would deal with the world, and I would reconcile those two things.
What move would it make? What move would I make? If they were different, I would ask why the moves were different. Probably 65–70% of the time, it was because I had forgotten something or couldn’t program that complexity. Maybe a third of the time, it was something else that made me think, “I’d better systematize that.”
Over time, they became one.
Because you learned something new about commodity prices or the bond market?
That’s true. In other words, the criteria are still there, the information is still there, and there are still cause-and-effect relationships. Causes arise before effects.
Whether it’s the most advanced artificial intelligence or any other kind of thinking, if you know these cause-and-effect relationships, you have to understand them.
Right. So you build these rules, and you very famously got 9% returns during the 2008 stock market crash. You understand where things are headed in a way that no one else, or very few people, do.
How many rules do you systematize in the Bridgewater system? Was that 100 rules? Was that 100,000 rules?
I would say that there are a certain number of markets, and then there are a certain number of rules for each market that are added. I would say it's hundreds, but I don't want to make it abstract. I want to convey this to every person.
It's okay if you just say, “I'm going to pause. I have made a decision.” You decide to invite me to this show, whatever it is, and you stop and think, “Why did I make this decision? What are my criteria?” You record them on your iPhone or something, and you start thinking about it. You will get the criteria that you use.
If you do that, you can systematize your decision-making process, and that's certainly where we are now. If you do a good job and know how to do it, you will also test these rules to see how they would have worked in the past, to help ensure better principles. I call them principles: What are these principles that, if I do this, I will get a better result than if I did that? This is what we are going for in AI.
So, back to the story of Bridgewater. Bridgewater succeeds. You can predict when the bubbles will burst. You can predict what will burst the bubble, right?
4. Assessing the current tech bubble
Knowing that there is a bubble is one thing. To know what will burst the bubble, or to predict, guess, or estimate, in 2026, what stage of the cycle are we in now, especially when it comes to the issue of the tech bubble? And then, very related to this, if we're in it, what I heard from you is: What could puncture it?
The way I want to convey this, and try to convey this in the book, is what the cause-and-effect relationships are. How does the mechanism work?
There are basics, like: If the government has a lot of debt, how is it going to pay it off? If companies take on a lot of debt and sell equity, they have to raise capital to do these things. There is a mechanism. There is a certain supply and a certain demand that goes into this.
Now you look back and answer your own question and say that it's a bubble question, not an AI question, because AI can do that too. It just has to be reproduced.
Throughout history, and I think it's still true today, it's good that amazing inventions come along that fundamentally change the way the world works. If you look back at this, there are mechanics to it that mean people want to invest in these things because of what they believe in and what their experience is.
Like in the 1920s: For the first time, electricity appeared in homes, and you had a refrigerator and light. First came movies, then radio, and then cars and airplanes. Everyone looks around and says, “This is obviously something to bet on.” Then everyone wants to bet on it, and it becomes a reality.
But everything has its price. Is this price too high? Then there is the need for money. The way it works is this: People borrow money, and this goes through a certain cycle.
There is what companies do, and there is what people and investors do. Companies say, “I need to compete for capital. I don't know if I'm investing too much or not enough.” There are 2 options, because this is very unknown. First, “I'd better invest as much money as possible, because I'll definitely lose if I don't step on the gas.” Then there are those who can reinvest.
No one understands this. Nobody understands this, and it's so hard right now, because you could be 1,000 times bigger with your annual profit rate at the end of the year than you are at the beginning. It is very difficult in this range.
That's right. This is difficult.
Please distinguish between good technology and good investment. People classically don't do that. If it's good technology, we don't know if it's right.
Then they say to themselves, “I can borrow money,” or companies can borrow money. There are limits on how much money you can borrow.
5. How bubbles pop
What happens is that you also have another dynamic where wealth increases relative to money. Let me give you an example. You can make a proposal today or anytime: “I'll collect $50 at a $1 billion valuation.” I only put in $50, but suddenly you're worth $1 billion. This company is worth $1 billion. It may not be worth $1 billion, but now you have a lot of wealth.
And that's why there's a lot of wealth around. Wealth is not money. You cannot spend wealth. To spend, you either need to sell wealth to get money, or you have to borrow, which means you'll still have to find the cash at some point.
When you find yourself in that situation, you see very classically that this particular wealth increases relative to money. Then the bubble bursts.
A bubble grows. The question is: Do we define this as a bubble? Many people think, “Will it pay off in the end?” But the end never comes. Twenty-five years later, it's the end. The question is whether there is a crash or whether something falls in the interim.
This happens when they want to sell wealth. Let's say, for example, if you had a wealth tax. This is what California offers: A tax on billionaires. Everyone must pay it. They don't have cash; they have to pay it.
That's the mechanics. You have to find cash, and that's what bursts the bubbles.
There are a few more things. Is it in strong hands or weak hands? In other words, can the owner take a loss and live through it, or do they find themselves in a position where they get scared and sell everything? Or do they need the money because suddenly they thought they were rich, but then suddenly they're twice as rich?
What do they do? Are they buying more? Do they sell? This is how the dynamics happen. That's exactly how bubbles work.
Good. Fantastic. Thank you very much. It seems we are quite close to a series of moments that could lead to the bubble bursting. We have potential wealth taxes being discussed. Anthropic is about to go public. All this wealth will turn into money. People will have to buy stocks.
Well, no, it won't turn into money from the start. There will be a process. Someone will want to buy, turn it into money, because they want to buy this house or that apartment.
An Anthropic employee will sell wealth to get money to buy a house or a yacht, or donate it to charity. Someone will have to get money to buy wealth. They might have to sell their other wealth, right? We don't know what will happen.
So you have all these triggers. We saw with the [?] fund last July that you have very avid investors who are extremely involved in credit games. At that point, Leopold Ashburner had to borrow a lot of money to cover a bunch of bets.
We have a bunch of dynamics that seem to me like some big pins heading towards the bubble.
It seems so to me too. You have a very high concentration of everything that is being bet on—in other words, its share of GDP, its share of wealth, its share of the stock market, what it means for the bond market, what it means for trade, what it means geopolitically for war, and what it means for employment. There is a lot to this.
Going back to your strengths, I'm simply saying that the American political system is not in the best shape. Domestic politics don't seem to be going very well. China's geopolitics is not very good. There could be a potential shock to the system with a move against Taiwan. Xi Jinping is in his 70s and wants to do this in his lifetime. Who knows what will happen? Maybe it's better to do this before Trump leaves office. We don't know, do we?
Climate is another unpredictable circumstance. Disasters can happen, right? There are all sorts of forces that make this moment dangerous. Rising bond yields, all sorts of problems at the moment.
I think there are 2 questions: Is it a bubble? I'm saying that I created a bubble indicator that has all of these things. This is a bubble measured similarly to the bubbles of 2000 and 1929, in the same order of magnitude, taking this mix.
Let's say I'm wrong. Is this a smaller or bigger bubble than 2008?
2008 was a smaller bubble.
It had a lot more debt than 2000, in terms of this dual-bubble dynamic. But it was housing. In 2000, there was a bigger bubble than in 2008, which was a big bubble because of debt. The topic was different. The theme in 2008 was houses. It was real estate, and it was structured that way. It was a bubble that was going to burst.
But if you take the enthusiasm, this whole stock-market thing and everything that goes with it, there was a bigger bubble in 2000 and 2029 as well.
The point I'm trying to make is that these conditions exist. We know what they are, and we also know that everything is one. Everything is AI.
What a person should do is not do everything in AI. Whether you want to bet on this or bet against it, don't do one thing that is very volatile and leverages someone you're betting on.
Let me ask you a few questions that refute this hypothesis. One of them would be: Ray, we've had this bubble situation for a while now, and we've had a bunch of exogenous shocks, right? We had tariffs, we had a war with Iran, and we had a situation with situational awareness. It seemed like none of this even came close to bursting the bubble.
We’ll just survive the man-made [?]. We had the SpaceX IPO, right? We had a lot of wealth and money. We will survive the anthropogenic IPO, the open IPO of artificial intelligence.
Why can’t the dynamics that supported the economy last year sustain it over the next 3 years? I would ask every person who asked this question to first understand the mechanics and look at the graphs.
So, let me call it, let’s say, a debt bubble. Okay, let me start with the bond market. Okay, it’s the other way around. There is an asset and a liability. An asset is a bond; a liability is a debt. Good?
The mechanism that I want to tell you about—let’s start with the U.S. government, but we can go further—is that it works the same for the government as it does for us with debt. That is, you take it on yourself, you have to return it, and there are conditions. The only difference is that the government can print money.
What happens very simply is that, when money and credit are borrowed and they earn more than the cost of servicing that debt, then you have a healthy system. Securing debt and leverage is good if it generates the revenues needed to service it and the world becomes a better place. You can see this in the numbers by watching how much debt service adds to expenses.
When debt service increases relative to spending—for you or for the government—that means your ability to spend is less because you have a debt-service payment. That’s like a buildup of plaque in the circulatory system, and that’s what causes an economic heart attack.
You can watch this. I’ll take a quick look at these numbers now. Secondly, there is supply and demand. In other words, every year, if you have a deficit, that means you have to sell your debt. This means that there is a certain amount of it and there are certain buyers of this debt, and the question is whether they want to buy it.
So either of these 2 things could be causing the problem. Yes. You can monitor it just like your cardiologist can monitor this plaque building up in your arteries.
Okay, here are the numbers. The United States government spends about $7 billion a year and receives about $5 billion a year. So a $2 billion deficit means it’s overspending by about 40%. And that has been causing these deficits for a long time.
The amount of debt it has is 6 times the amount of money coming in. Right. And not only do you have that, but your interest on that debt has now become half of that deficit. That’s about $1 trillion. That's $2 billion you have to sell, about a trillion is the interest rate.
But that is the smaller number. A larger number is that every year you also have to sell about $10 trillion that has a maturity date. That means you have to resell it. Now you can see it, and you can predict it.
So that’s the answer to your first question. It’s like someone who has had a lot of problems and says, “I haven’t had a heart attack yet. I eat fatty food. I’m enjoying this. I haven’t been outside this week.”
You can show them a sign and let them decide what they want to do. I can make predictions year after year about what it looks like, and I can say, “You’re probably going to have a heart attack.” Please don’t take comfort in the fact that you haven’t had a heart attack yet, because you can go back and do these numbers and you’ll see where they are.
I’m also telling you that the supply-and-demand situation is a problem. Demand is decreasing. One-third of our debt purchases are from foreigners. The largest foreign buyers were China and Japan.
For various reasons, foreigners are worried about the size of the debt. They’re also worried about the conflict that exists in the world and whether they can be sanctioned, meaning they can’t make these debt payments, the way Russia was sanctioned and so on.
That weakens foreign demand. Japan, for its own reasons—and I won’t digress—and other countries think that they need it, so it becomes a dynamic. We have a supply that goes like this, and we have weaker demand. So we have both of these 2 classic problems.
I don’t believe myself. You don’t have to believe me. I’m just saying: you can look at the X-ray and decide for yourself. Does this make sense?
The analogy, Dr. Dalio, is that one of the things AI optimists would say here is: you spend $7 trillion, you get $5 trillion, you have a decrease in demand for debt purchases, and you have extraordinary maintenance costs. But we’re about to enter a whole new era where productivity will increase not by X percent, but by 10 times, right?
We’ll have geniuses in the data center. We’ll have a whole new world of work, and we’ll be able, thanks to increased productivity, to get out of it. The reason it doesn’t fit into your model is because it never happened.
There has never been such an increase in productivity and technological advancement as this. So that’s exactly what I would tell the doctor. I would say, “Dr. Dalio, I understand that this is bad. I have plaque buildup on my teeth. I understand that I haven’t been outside. I smoke cigarettes, right?”
“But there is a new treatment method. It is in development. It’s about to be approved by the FDA. I’m fine.”
Okay. Just in terms of the mechanics, if incomes go up so much that it becomes easy for the government and others to pay off this debt, then of course you’re going to get the results that you’re talking about. Now we’re dealing with the actual probability of that, the likelihood of that, and what’s going to happen next.
First of all, one of the problems that’s happening is that it’s creating very large disparities in wealth. With these disparities in wealth, the question is how they’re going to play out in terms of inequality in wealth. We can expect something, but nobody knows for sure.
There are those who own stocks, and there’s what we were talking about regarding wealth and all that. So, first of all, the 3 questions are: How does this transfer of wealth happen?
6. Why there might be a crash even if AI ushers in unprecedented productivity
The average household has twice as much national debt, part of which is about $280,000 per person in the household. So how do they get it with this card? You need to go through this if you are getting a productivity miracle, which is converted into income.
This must be converted into sufficient income to cover this over the required time. Then you have to make sure: How does the government get this for its revenue, or do others get this? Is it because of taxes? Will this be resolved politically? How quickly will this happen relative to the time horizon that needs to occur for this to happen?
You need a productivity boom, and then you also need a corporate tax boom. You need to increase your income. So when will we get the income? Where will we get this money over time? How much will it be?
Please go through this line by line and see what it is year by year and what you think will happen to get there. By the way, the miracle of productivity does not carry over to income. Good, because there is always competition, and there are always means by which it becomes cheaper. There is price competition, and so on.
But we can agree on the mechanism of this. What I would say is that, at least when I look at it, it’s a scary picture. So maybe let’s not ask, “Where are we going to be in 5 years? How is this distributed? How does this all work, and will we be there?”
If you want to bet on it, you can bet on it. When I look at it, I think, at the very least, we’d better worry about it.
Let me ask you another question that I heard from you, which I think is fascinating. OpenAI—maybe last spring—there was some talk that perhaps the government should have a stake in OpenAI, maybe a 10% stake, like it has in Intel. Perhaps there should be a sovereign wealth fund where the United States government takes stakes in large artificial-intelligence companies.
You like the idea, right?
I want to make it clear: we are entering a new world order. We’re in a new world order, okay? In such times of international competition, believing in a completely free-market system that governs everything has never been the case, and it is not necessarily the best way to go.
We’re very free-market. I’m very pro-free-market. But if you make it completely free, then here’s what happens: consumers buy what they want to buy. You can buy bags for $1,200 or something, and those resources might be inefficient in competing in a world where, if you win a battle with artificial intelligence, you’re not only winning economically, but you’re also fighting for military power or for world domination, and so on.
So government intervention, in a certain way, is necessary. You can’t be too clear about the forms of leadership. How to do this and how to achieve this balance is a very complicated thing.
I also support Trump’s policies. I’m in Connecticut, and I support Trump’s policies. I believe this way—I told my son about it this morning. He asked if we could move to Connecticut so he could get some of Dalio’s funds into his account.
The point is, it depends on how. It depends on what, in terms of giving people a share.
But also let them know that it's good. Sometimes you need the government to set the direction. Free-market people, as I usually tend to say, are very cyclical. This is part of this big cycle. When you get into it, you need a government—you hope for a smart government that does this well and provides things like that.
So when you say I'm for it, I just wanted to make it clear what I think about it. But there is also the advantage that if the government owned 10% of Anthropic before Anthropic went public, that means only 90% of the wealth is converted into money, right?
Don't you need to borrow money somewhere? Where does the government get its money from?
It borrows it.
Well, that depends on whether the government pays Anthropic for the 10% stake or steals it. You can make the government steal it. That's what you're talking about.
I'm not saying that I—
Well, let's use it in support of the idea in general, and definitely not in support of theft. Governments, I think, don't steal. You can make them take where they are—
They could take government property.
They could take it. That's what you're—
Just say, for the listeners, I'm not advocating for the government to steal these companies. I'm just asking about—
They're taking them away. There are sovereign wealth funds in the world where governments buy shares in companies, which reduces the amount of wealth that needs to be converted into money.
So, if I were—okay, I'm just a mechanic, right? I'm just trying to say, is the government getting cash? Is this a cash thing, where someone takes something from someone else? I'm just a mechanic who runs what I like about your principles and your mechanics.
So let me ask you this. If Dr. Dalio gets appointed to lead, and there are midterm elections, and Trump says, “You know what? For the next 2 years, I will do everything I can to ensure that America's financial sector does not implode, and that we in the United States are as well-positioned as possible to become an AI economic superpower compared to China. Mr. Dalio, you are now in charge of American economic policy. You have no obstacles. You can do whatever you want. What are you doing?”
I'm still dealing with the political system, right?
No. The political system—you do whatever you do. Everyone agrees.
I know, but I'm asking this question for this reason. When I went to Washington, I knew leaders on both sides, and almost everyone agrees with this explanation in the book and the mechanism. And there's one thing, by the way: you're very guarded, and I've spent a lot of time with you, and I have no idea who you would vote for or who you voted for in the last election.
7. Solutions to the government’s debt crisis
Right. I just want to tell you that I'm apolitical. We can discuss what that means, but I'll try to answer your other question first.
What I have written here is what I call the 3% solution. The 3% solution is that you need to reduce the deficit to less than 3% of GDP so that it doesn't get worse and create even more plaque.
Okay.
I don't care if it's on the left or the right. Of course, I have my own opinions and preferences, but I think the argument gives you an extraordinary diagram here at the end of all the different levers you can use to achieve this.
Yes. Okay.
What I'm saying is, it's like a ship heading for the rocks. We can agree that you're heading for the rocks on a ship, and I don't care if you turn left or right. I'm just saying, take that 3% from it.
This is undeniable. Scott Bessent thinks it's 3%, but others think it's 3%.
Okay.
I said basically I'll do it proportionally. I don't know—raise taxes by 4% or 5% on whatever, and cut your spending by 4% or 5%, and something like that. The main thing is that you get there; otherwise, you will hit the rocks.
And everyone agrees, but the response I get from both sides is this: you have to understand that in order to have a mandate to be in government and have a vote, I have to make a few promises. “I will not raise your taxes or cut your benefits.”
Right.
So when I'm asked a question, do you give me the authority to say, “Fuck the political system”?
You have the authority, because otherwise you wouldn't be able to do this. So I give you the authority to say, “Fuck the political system,” and I want you to design your policies with maximum consideration of how they will affect AI and America's position in the strategic AI competition.
I think you should make the country good for everyone.
Yes.
Taxes should in some way be a means by which the majority of people benefit. Otherwise, we won't have a system that works well, and AI has to succeed. In different ways, there are probably going to be some taxes associated with AI. It doesn't have to be just with artificial intelligence; it could be in different ways—ways in which we're not worried about the explosion or the unemployment issues.
Yes. Where you have to win this war.
To win this war, you have to achieve it in such a way that it becomes beneficial for the majority of the country. There will be some taxes associated with that, and I don't want to get into the technical details. There are a lot of technical details about taxes—exactly what, the timing, and how it works—but it has to be.
The problem is that there is still not enough money. Can companies involved in artificial intelligence be taxed? Could this have a significant impact?
They don't make money.
But they make enough money to pay a certain amount of taxes.
I'm just saying—let's go back to our previous point. They can pay their taxes through the equity of the U.S. government.
Okay, but it's still not money.
Okay, but you actually then have something like—
I think you're right.
No, I didn't solve the problem, right?
Good.
I'm just trying to say, if you want to do it like wampum, then you can take the stock and go to some guy who's having trouble paying his bills and say, “I'm going to give you the stock.” If someone can trade the stock for food at the store—
Okay, okay. I'm trying to buy a banana and I don't have a dollar in my pocket, but I have a small stake in Anthropic, and that's fine.
So we still need to get back to how this is going to work, right?
Good.
Good. So you implement this policy.
8. The difference between how China and the US view AI
Let me ask you about strategic competition with China, because you spend a lot of time in China. What are people studying AI in the United States getting wrong about how China thinks about AI?
The main things to think about are how China thinks about AI and how Chinese leaders think about it differently.
First, they view AI, generally speaking, like electricity, water supply, or infrastructure—building roads and railways—as a means by which profit is not the primary goal. Profit is good. It's a matter of rationing, but the main goal is the efficiency of everything. If so, then everyone should have AI, just like everyone should have electricity.
The cost of this is a secondary consideration. The impact on productivity is part of that. Then they think about how they compete and use AI to compete in global markets.
So the second of those three, I would say, is how they're implementing AI into industrial goods and so on. This model of frontier capability, achieving the highest levels of capability, is not where the greatest advantage lies. They of course want to do it in the best way possible, but the advantage comes from using it and improving the standard of living, including through simpler, cheaper versions of artificial intelligence in manufactured goods.
The Chinese now own 32% of the world's industrial production—more than the United States, Germany, and Japan combined, in terms of volume of manufactured goods. They are very competitive in global markets. They think, “How do I compete with this and use AI in this way?”
They also know that this is big. Whether it's used for economic warfare or military warfare, it's something you have to have to be effective.
Let me finish this, Ray. I used a digital Ray. This is an AI avatar. You can ask him questions. I asked him to help me prepare for what I would ask you in this interview.
What are you doing to make AI more accessible to people, to investors, and to others?
I'm at a point in my life where my top priority is to pass on everything I have that could be valuable. To some extent, it's these ideas, and to some extent, it's philanthropy or something related to that.
AI in investing—I want to make it available to anyone who has the opportunity to make the most of it the way I did, and thus create a means or a vehicle through which there is a combination of education and the enormous opportunities that can come from AI, with an AI platform that can be effective.
And how will this work?
There will be an artificial intelligence platform that everyone can use to structure their—
That's what I'm hoping for, yes. Well, let me ask you one more question.
So, you wrote this wonderful book, Principles. It's not just the principles you used for your investments. These are principles for how to live, right? That's one of the things I like about it.
9. What Ray tells young people entering the AI era
Tell me, what advice would you give to a young person? My children, let's say—they're 18, 16, and 12 years old. What advice would you give them about the world they are going into that is different from the advice you would have given someone 15 years ago?
I don't think it's much different. In terms of how it's different, I would say you have to be on the cutting edge of artificial intelligence.
You have to be on the crest of the wave; otherwise, you'll drown in the waves. But I would say the advice is the same for all of life. Everyone has their own principles and makes their own decisions, but there are certain things that are true.
First of all, know your nature. We all have our own nature. You might have psychological profiles indicating whether you are adventurous, introverted, or extroverted. What is your nature? We all yearn for something. We want something, and we have certain tendencies. Find a way that fits that.
You found your way, I found my way, and it suits me. On the way to that path, if you know your nature, experiment and learn. The number one principle, or the main principle, is learning. Learning comes from mistakes, and so I have a principle: pain plus reflection equals progress.
As you go through these things, live your best life, however you define it. Make your work and your passion one and the same, if you can. Don't forget about the financial part. I have a principle: make your job and your passion one and the same, and don't forget about the money part. But don't confuse passion with work; combine those things if you can.
This is an adventure. Go on an adventure. Principles define what you want in your life, and they determine how you will achieve it. I don't see any difference between work principles and life principles. You still have your principles, and they are what make you successful in what you strive for.
Right. And when you talk to your kids—or anyone who talks to kids—there are these general principles that are probably good advice for them.
Yes, I think that's right.
The last question I ask all my guests is: if you had an unlimited budget and could support anything in AI or create anything in AI, what would you do?
10. What Ray would do if he had an unlimited budget to invest in AI
What I'm doing is trying to create artificial intelligence. I have a big budget, and I'm trying to take what I know about investing and make sure that everyone can access it at such a quality level that they can use it from an early age.
I started when I was 12 years old. You don't need to have an institution, and you don't need to have a lot of expensive resources. You just need to have criteria in your head that you think through, and then you can follow them.
I try to pass on my own abilities, using what I have learned. Then I do my other things—philanthropic things and stuff.
Yes. Thank you. In this interview, we had some pain and some reflection, and I think a lot has changed. Thank you very much, Ray Dalio.
Thank you.
It was Ray Dalio. If you enjoyed this episode, you might also like my conversation with Azim Azhar three weeks ago or my episode with Erik Brynjolfsson from Stanford from May of this year. And if you want more of this podcast, please subscribe wherever you watch or listen. If you watch on YouTube, leave us a comment. We'll probably see him. If you listen on Spotify, Apple or another platform, leave us a review. Thank you for viewing “The Most Interesting in Artificial Intelligence.”