4:16
Greetings. Welcome to another episode of ColdFusion. Imagine: you are 24 years old, today is your wedding day, and you are going to marry the person you love. You invited friends, family, and Silicon Valley's most prominent figures to Carmel, California, for a multi-day celebration. There's a Tuscan-style villa, a spa in the middle of the forest, and even a pre-wedding conference with panel discussions. But you have another reason to celebrate. Less than 2 years ago you started a hedge fund with $225 million and now you manage $45 billion. But there is one problem. While the guests are arriving and you are preparing to say your wedding vows, your biggest investment is crumbling. Goldman Sachs, JP Morgan and Bank of America are not at all thrilled.
They demand more money to cover your losses, and in a matter of days, approximately $35 billion of the empire you built will be gone. You may not think $35 billion is a lot of money, but it's important to understand how big it is. So, if I gave you $1 million every day, how long do you think it would take to spend it? Some of you might say: maybe a year, maybe 10 years. But no. How about 96 years? Imagine: $1 million spent every day for almost a century. That's a huge amount for a 24-year-old to manage, let alone lose in such a short time. This is a true story that recently happened to a young man named Leopold Aschenbrenner. For
him, it was a truly " hell of a" wedding week. Leopold Aschenbrenner Leopold made his reputation by convincing the world that he saw the future before anyone else. His hedge fund was literally called Situational Awareness, and somehow it became the biggest crisis of his career right when he was supposed to be having the best week of his life. So how did a 24- year-old with almost no trading experience start managing tens of billions of dollars? And how did a hedge fund that was generating over 1000% profit lose about $35 billion in 2 weeks? And what does all this tell us about the current state of the markets? Let's figure it out.
The future can be seen first in San Francisco. This is the opening line of a 165- page PDF that Leopold published in June 2024. This essay was called " Situational Awareness: The Next Decade ". His main argument was that AGI, or artificial general intelligence, would arrive by 2027, and the world was not ready for it. Almost immediately, it went viral in Silicon Valley. Investors read it and then started looking for the author. So who was this person in the PDF prophecy really? This is a summary of my post or my perspective on the project.
Leopold was born in Germany to a family of doctors, skipped several grades in school, and graduated from Columbia University in 2021 as the top graduate at the age of just 19. Although his first job out of college was doing charity work within Sam Benkman-Fried's FTX empire itself. Leopold held this position until the collapse of FTX in late 2022. So it probably was n't the best company, but of course I've already covered FTX in detail on this channel. So I won't retell everything that happened, but essentially it was a multi-billion dollar fraudulent crypto exchange that became insolvent. Somehow, this turned out to be not the last company on Leopold's resume to make headlines for negative reasons. After FTX, he ended up at OpenAI. Leopold
worked on the superalignment team. This was a team created to figure out how humans can maintain control over AI systems when they become smarter than us. However, he did not last long, and in April 2024 he was fired. The official reason for his dismissal from OpenAI is that he shared a working paper with outside researchers, but Leopold tells the story differently. He believes the real problems started with a memo he sent to the board of directors a few months earlier. He warned that the company's security was not strong enough to stop foreign agents from stealing its secrets. They say the HR department
told him that this note was, and I quote, " unconstructive." He tells his version of events in a podcast. Last year, I wrote a memo, an internal memo, about security at OpenAI. I found it, you know, extremely inadequate, I thought it didn't protect against the theft of model weights or key algorithmic secrets by foreign entities. So, I wrote this note. I shared it with a few colleagues, a couple of management representatives, who generally said it was helpful. But then, you know, a few weeks later, there was a major security incident. And that prompted me to share a note with a couple of board members. And so, after I did that, a few days later, I was made very clear that management was very unhappy that I had shared this
note with the board. Um, and then I got an official warning from HR because of that note. Two months after his release, he published an essay that made him famous almost instantly. Tim Ferriss introduced it to his audience on "The Tim Ferriss Show," one of the most popular interview podcasts in the world. He called Leopold, quote, " the Nostradamus of AI," end of quote. Ferriss said he didn't consider himself an AI expert, instead directing his entire audience to Leopold, calling his string of predictions astounding. So Leopold gave investors somewhere to put their money, and later that year he launched a fund named after his own essay-"Situational Awareness LP." This was, in essence, a continuation of the essay itself. If general- purpose AI is truly
approaching quickly, why would anyone invest as if it weren't? Naively, one might say: well, I looked at the scaling curves, and they imply that we will be buying a lot more computing power than analysts realize. Yes. Shouldn't these analysts have gone bankrupt by now? What is happening? Yes. I mean, I used to be a proponent of the efficient market hypothesis. I was an economist, you know, yes. But I think there may be groups of people, smart people, say in San Francisco, who have an advantage over the rest of society in terms of their vision of the future. Um, and for example, COVID, right? I think there was the same group of people who just saw it and predicted it.
They played for relegation and made good money. Why is general purpose AI not included in the prices? You know, it's like, why has n't the government nationalized the labs yet, right? It seems like society hasn't priced it in yet, it hasn't fully spread yet, and, again, maybe it's a mistake, right? But I just think that not many people take these ideas seriously. Yes. The investment proposition for general- purpose AI was compelling enough to attract serious money, including from Stripe's Collison brothers, former GitHub CEO Nat Friedman, and investor Daniel Gross. So what exactly did they support? In October 2025,
Fortune magazine published a detailed profile of Leopold, speaking with more than a dozen of his friends, former colleagues, and investors. All this painted the image of a rather controversial figure. People constantly used words like " intense," " scaryly intelligent," and "daring." Some of them said that he had the aura of a child prodigy. This reputation was built over years. He entered Columbia University at the age of 15. At 17, economist Tyler Cowen called him an economic prodigy. Leopold's astronomical profits later confirmed this. But not everyone was impressed.
Several people noted that his ideas weren't all that original. He simply took what people in the AI industry were already saying and believing and presented it better than anyone else. The story of the path to AGI is actually not new. Here's David Chalmers talking about it in 1996. So you have a machine that's as smart as we are, then maybe in 10 years we learn to program a little better, and the machine becomes a little, just a little smarter than us. Then this machine, in turn, will probably become a better programmer than we are. She will be able to create a better car than we can. So she will create an even smarter machine. This becomes an endless spiral. So I think that once we get to the point where a machine is as smart
as we are, the point where machines are a thousand times smarter than us won't be that far in the future. One former OpenAI researcher said that the ability to sense the zeitgeist in AI was Leopold's main superpower. Others simply found it difficult to work with him. Another OpenAI researcher said that Leopold took pleasure in annoying his superiors. Another called him harsh. Whatever you think of him, he clearly believed in his ideas and presented them with such conviction that he forced investors to invest hundreds of millions. And he did all this at the age of 23 in a firm that reportedly had only eight employees. So,
eight employees, a portfolio that would eventually reach tens of billions of dollars, and a young founder with almost no professional trading experience. What could have gone wrong? Silicon Valley didn't stop to think about it, and they didn't particularly care. But Wall Street does. As the fund grew, Leopold reportedly headed to New York in search of a different class of investors, larger institutional capital that could expand situational awareness. But this was no longer Silicon Valley. The reception there was noticeably colder. Blackstone reportedly agreed to a meeting but declined. According to Rob Copeland's report, another wealthy New York investor asked Leopold a fairly simple
question: "What happens if you're wrong?" Not necessarily wrong forever, but what if AI adoption slows down, or the emergence of general- purpose AI is delayed, or is your thesis simply ahead of its time? End of quote. Leopold reportedly did not have a detailed answer, he simply believed his thesis was correct. This part of the episode is sponsored by WhisperFlow. While preparing these videos, I spend a lot of time researching, writing, and scripting, and sometimes it gets pretty exhausting. With a hundred tabs open and deadlines looming, I often think much faster than I can type. That's why I started using WhisperFlow. WhisperFlow is a voice-to- text tool that works in almost every
app: Gmail, WhatsApp, Slack, pretty much everywhere I write. Instead of typing, I speak naturally and it turns my thoughts into clean, formatted text, ready to send. Talking has always been faster than typing. The difference is that most dictation tools capture every pause, every " um", every correction, every change of thought. Hey, Gotham, um, actually, let's move tomorrow's meeting, sorry, better to Thursday, around 2 o'clock. No, let it be 3. And can you send me the final draft in advance? Thank you. WhisperFlow actually cleans everything up while you speak, so the final result looks much closer to what you would have written yourself. You could say that many apps now have voice dictation, and that's true, but each
app does it in its own way. Having one system that works the same everywhere has proven to be extremely useful Whether I'm writing emails, replying on WhatsApp, or taking notes, it also helps a lot when brainstorming with the team. Instead of stopping every few seconds to type, I can just keep talking and stay focused on the idea. Another thing I noticed is how well it handles technical names and industry jargon. Hey Tamsin, can we change the montage for the sections dealing with Nvidia and ASML lithography? Can you also download the LUT files I sent you for Premiere Pro? And over time, it learns the words you use most often thanks to its own dictionary. The result is simple. I spend less time typing and more time
thinking. If you want to try WhisperFlow yourself, you can get started for free. And if you use my link below, you'll get 1 month of WhisperFlow Pro with unlimited words for free. Thanks to WhisperFlow for supporting ColdFusion. Now let's return to the story of Leopold. The strategy itself was simple on paper. He bought shares of everything related to artificial intelligence. Leopold could have followed the traditional Silicon Valley path and created a venture fund that would invest in private AI startups before they went public, but he didn't. The question is why? asked Leopold this very question, and received the answer that AGI would change the world and the economy on a scale no one had ever seen before. So, according to Leopold, if you want to
profit from something this large, you have to be in markets where you can move serious money quickly. In other words, publicly traded companies, those that anyone can buy or sell in a matter of seconds. Remember this thought, because it will play a cruel joke with him. Leopold bought the physical foundation of the AI boom: the chip manufacturers, energy companies, and cloud service providers that build it all. Stocks like SK Hynix, CoreWeave, and Nebius. He then opened short positions in software companies like Adobe and Salesforce- enterprise tools that he believed AI would make redundant.
Shorting means you bet on a stock falling, and at that moment you make money. On paper, this looks like hedging: long positions on one side and short positions on the other. But in practice, both sides depended on the same belief: that AI would turn everything upside down, both at once and on Leopold's timetable. Leopold didn't stop at just investors' money. He used 400% leverage. But the thing is, for a year it worked just amazingly. His AI stocks soared, and he made billions. And then came July 2026.
Rumors and panic spread. Chinese models with open scales could perform the vast majority of tasks as well as leading models, but at a much lower price. American markets were scared. Public reports show that the fund's largest long positions were concentrated in these six names. By the end of July, each had fallen 50-78% from their recent peaks. At the same time, the software stocks that, not down. So both sides of his "hedge position" started losing money simultaneously. This is exactly what hedging is supposed to prevent. The fund's value was falling rapidly, and its three prime brokers-Bank of America, Goldman Sachs, and JP Morgan Chase- began knocking on the
door. They demanded more supplies than Leopold had available. When the fund cannot meet this requirement, the bank sells its positions to recover its money. He was in big trouble. The calls were coming in non-stop, and something was bound to break. And do you remember what Leopold liked about public actions? The thing is, you could quickly deposit and withdraw serious funds from them. It turned out that it worked both ways. After all, what had cost 45 billion was now worth 10 billion. In the end, a man in his 20s lost $35 billion.
By July 30, Leopold's stock portfolio, worth about $16 billion, had been sold overnight at a discount of more than 10% to market value. This was the largest part of a The buyer was Citadel, one of the most successful hedge funds in the history of Wall Street. Citadel won the deal in an open auction. The Leopold Foundation survived only because of its share, worth about $5 billion, which was private and illiquid. Anthropic is also the company where his future wife, Avatar Bowden, works as the CEO's chief of staff. Instead, Citadel offered to take his entire portfolio of publicly traded stocks. These are companies like SK Hynix, Coreweave, and Nebius. Anything that has a current stock price. The sale of these shares raised the cash that
creditors were demanding, so Anthropic remained. Leopold sent a letter to his investors acknowledging that the fund was at the epicenter of what he called market " turbulence." It's very British to describe a crisis without calling it that. The letter ended with a postscript inviting clients to invest new funds starting August 1. Then, as planned, he got married. But here's where things get a little weird, because if this had happened on Wall Street, rather than Silicon Valley, losing $35 billion in a few weeks would have led to an exodus. But Leopold did not stop. In fact, a few days after the Citadel deal closed, Leopold transferred $400 million to an unnamed private company. He had already invested $100 million in the same
company a month earlier. Leopold never explained why he invested half a billion dollars in it. But that's the whole point. He's a man who saw borrowed money nearly destroy everything he built, and in a matter of days he's investing half a billion dollars in a single private company. The same confidence created a $45 billion fund in less than two years, but led to a $35 billion hole in two weeks. He used leverage, but it's not so much about that as it is about how big it was. He really got into leverage. We reported a leverage of 4:1. For every dollar he borrowed, he borrowed a good three or four dollars on top. He also
used options that added leverage, but it wasn't just about them. There are many leveraged hedge funds, but they usually trade less volatile assets, bonds, etc. So you can leverage if the underlying investment is not volatile, but if you're going to bet on the rise or fall of AI-related stocks, it's inevitable that at some point you're going to have a major setback that leads to a margin call, a panic, and what happened here. This is a mechanism in simple words. Borrowed money piled on top of some of the most volatile stocks on the market. Leopold used approximately four-fold leverage, meaning every dollar of gain or loss in the underlying stocks was a four-fold hit to the
fund's equity. SanDisk, one of the fund's largest holdings, fell 57% in less than a month. With four times leverage, such a drop could completely wipe out the capital backing the position. But there is another problem with leverage that is less obvious. The expected return roughly scales with the amount of leverage you take. You double your exposure -and in theory, you double your profit. But volatility is not behaving so pleasantly at all. Damage caused by sudden fluctuations can grow faster. With four times leverage, you greatly increase the chance that one of these unfortunate swings will wipe out so much capital that you will never be able to participate in the recovery. Imagine that his entire thesis turns out to be correct and AI appears exactly
as he predicted, but only six months later than the market expected. This will not help if there is a 50% subsidence first. Your creditors will demand billions in collateral, and your positions will be forcibly sold. You will find yourself out of the market and in a difficult position before your thesis even gets a chance to materialize. With that much leverage, you have to be damn sure you're right and your timing is right, otherwise. The kind of investment firm I would use for hedging. And it's like, you know, or AI comes along this decade, and yeah, your human capital is depreciating, but you've converted it into financial capital. Or, you know, no AI this decade, in which
case the firm might not perform well, but you're still young, in your 20s, and you're still smart. Perfectly. Leopold joked that his own life was effectively insured: human capital depreciates while financial capital grows to compensate. Or it will take longer and he will be young, sharp-minded and have enough time to start over. This is a witty joke. This is also a true hedge: two outcomes, and in both cases he wins. However, his life plans turned out to be a very different story than his hedge fund, which he ironically did not hedge properly. Both the long and short positions were based on the same assumption. Buying AI infrastructure and opening short positions on
software driven by the same AI narrative doesn't level anything. This actually just doubles your exposure, because it's one idea disguised as two deals. So when things took a turn for the worse in July, there was nothing on the other side of the balance sheet to soften the blow. Cliff Asness, a hedge fund manager with thirty years of experience, was asked what he thought about the whole saga. His answer is borrowed from what he called the Kelly rule. This is a formula that tells investors how much to bet, depending on their confidence. His answer: never use "full Kelly." "Full Kelly " is essentially a mathematically aggressive betting option based on your estimated edge. The problem is that a calculation is only as good as your
assumptions, and in the real world, those assumptions may be wrong. That's why investors bet only a fraction of Kelly, sacrificing some of the theoretical profit for a much higher chance of survival when reality isn't in their favor. It's still early days for AI, and no one can know for sure how things will turn out. Large Language Models (LLMs) may never lead us to AGI, but a change in the underlying architecture or another invention in computer science combined with LLMs may well. We just don't know. For example, I've been playing around with GPT-6 Astra, and I'm amazed by some of the things it can do.
The introduction of the circular token process seems to have improved the model architecture, but despite this, GPT-6 can still fail at other simple tasks. For the average user, this may not be a problem, but for applications in mission-critical corporations, it complicates the situation and could become an obstacle to receiving those trillions of dollars that AI companies have borrowed. Even if Leopold is right and AGI comes out next year, its unbalanced deals and excessive leverage made it quite likely that it would go bankrupt when volatility increased. Confidence and risk management are different disciplines, and he learned this the hard way. But fortunately for him, it was just a hard lesson, not insolvency.
In the end, he kept his foundation, kept his wife, and probably his followers. He lost $35 billion on this, but, strangely enough, within a few days he wrote another check for a nine-figure sum. Perhaps it's conviction, or perhaps it's the same " blind spot" that led him to this disaster in the first place. But in any case, Silicon Valley is not worried. But bond markets, on the contrary, are worried. This is a huge risk, which we will talk about another time. What impact do you think the issuance of technology stocks has on global returns?
This is just one of many factors driving up long-term bond yields, and investors should be very flexible here. Yes, 30-year US bonds are at their highest level since 2007, and that says a lot. In the latest news, JP Morgan has reportedly severed ties with Leopold. Well, it makes sense since he lost so much money, but he's back to his old ways, investing in the same companies that caused him losses: SK Hynix and SanDisk. So, what do you think about this? Is Leopold right about AGI, he just rushed it? Does the collapse of his fund indicate how confident we should be about the future of the AI industry? Let us know in the comments.
Anyway, I hope you enjoyed it. My name is Dagogo. You watched Cold Fusion, see you soon in the next episode. If you want to watch something different about technology or business, check out the channel. There is a lot of interesting content here. Okay, thank you, friends. All the best. Cold Fusion. This is new thinking.
