Subscribe to Bloomberg.com for unlimited access to all our coverage.

Being bullish on artificial intelligence has been, for the most part, a great trade. Nvidia Corp.’s stock is up 83% over the past two years. SK Hynix is up 220%. Anthropic is up more than 400% in less than a year. Over the past few years, a consensus has developed that AI will be big, that companies will spend tons of money building it, and that companies that sell chips and power into the AI boom will make a lot of money. There have been related fears that traditional software companies will lose out. Adobe Inc. is down 51% over the last two years.

Let’s say you were really smart two years ago. What should you have done? I mean:

1. Buy stock in AI model and infrastructure companies;
2. Ideally pick the right ones, the ones that will be winners in the AI boom; and
3. Short traditional software companies that will be losers in the boom.

Great trade! You would have made a zillion dollars. Do it all day long. Put all of your money into this trade. Put more than all of your money into this trade. Raise money from outside investors, and put their money into the trade. Borrow money from banks, and put that money into the trade. To be clear, this is not investing advice. It is hindsight. Buying stocks that go up and shorting stocks that go down is a good investing strategy and you should, um, have done it, a lot.

Leopold Aschenbrenner did. Here’s a Wall Street Journal profile of him from June:

Aschenbrenner had no professional investing experience when he launched his AI-focused firm, Situational Awareness, less than two years ago, with a few hundred million dollars. 

Prescient stock picks and whooshes of inflows have vaulted its assets under management to more than $20 billion, according to people familiar with the matter, approaching the size of Bill Ackman’s Pershing Square and Dan Loeb’s Third Point.

Situational Awareness has gained about 270% after fees this year through May and is up more than 1,000% after fees since inception, one of the people said. One of the fund’s most successful bets is a stake in Anthropic that today accounts for about one-fifth of its assets, the person said.

His core thesis, as CNBC notes today, was that the AI boom would be big and real:

Aschenbrenner became prominent in technology and investing circles after publishing a series of essays in 2024 arguing that rapid advances in artificial intelligence would require an enormous expansion of computing power, advanced semiconductors, memory and energy infrastructure. Those ideas became the intellectual foundation for Situational Awareness after he left OpenAI.

That is the central modern AI thesis, and Aschenbrenner was early and smart and articulate about it. This allowed him to raise money, and the fact that it has been basically correct allowed him to return 270% through May.

The only note, really, is that if you are all-in on this thesis, you might be more than all-in on this thesis. You won’t put 100% of your money (and your investors’ money) into the AI boom. You’ll put, like, 300% of your money into the AI boom. You’ll borrow money to lever up your bets on the AI boom. As your AI stocks go up, you’ll borrow more to buy more. Getting a 200% return on your money by buying SK Hynix stock is great, but getting a 1,000% return on your money requires borrowing more money to buy more stock.

This is a naturally long-term trade. Aschenbrenner’s famous June 2024 essay series is titled “Situational Awareness: The Decade Ahead.” The point is not, like, “SK Hynix will beat earnings expectations next quarter”; the point is stuff like “by the end of the decade, we are headed to $1T+ individual training clusters, requiring power equivalent to >20% of US electricity production.” A lot of people believe this, and a lot of money has been mobilized to bet on it. Tech companies and private credit firms are raising tens of billions of dollars of bonds to build data centers. Frontier AI labs and public AI companies like Alphabet and SpaceX are selling tens of billions of dollars of stock. The way those bonds work is that investors give the data-center builders their money and get paid back over decades as the data centers make money; the money doesn’t have to be repaid for years. The way those stocks work is that investors give the AI companies their money and hope that their stocks become more valuable; they never have to be paid back.

If you are a hedge fund borrowing money from banks to buy AI stocks, the way that borrowing works is, uh, if the AI stocks go down you get margin calls? Like, that day? Your money is not locked up for the long term, and you might have to repay it at any time. There is a mismatch between your thesis, which is measured in decades, and your funding, which is kind of overnight. The AI thesis is up a ton over the past two years, but it is down quite a bit over the past two weeks, and CNBC reported today:

The battered hedge fund founded by former OpenAI researcher Leopold Aschenbrenner is unwinding many of its trades after big losses on artificial intelligence stocks and a bad bet against  software stocks left it scrambling to raise cash, according to people familiar with the matter. ...

The fund grew to as big as $45 billion at the start of July before big losses took hold, according to a person familiar.

Several of the firm’s prime brokers — including Bank of America, Goldman Sachs and JPMorgan Chase — have been working with the fund as it seeks to meet margin requirements or reduce positions in an orderly fashion, according to people familiar with the discussions. The brokers have been marketing a group of the firm’s holding on both the long and short side for sale prior to Thursday’s start of trading, according to people familiar with the situation. …

And the Wall Street Journal reports:

Situational Awareness, the highflying artificial-intelligence-focused hedge-fund firm, sold the bulk of its stock portfolio to Ken Griffin’s investment firm Citadel after suffering deep losses, according to people familiar with the matter.

The firm had been seeking buyers for its holdings and trying to raise new capital in recent days, some of the people said. Its public investments include South Korean chip maker SK Hynix and others that had been stung by an investor backlash to AI.

A crude but useful characterization is that Situational Awareness is really really good at thinking about the long-term implications of AI, and Citadel is really really good at thinking about funding risk. [1] So now Citadel owns Situational Awareness’s long-term AI bets. And those bets seem to be up since Citadel bought them.

On prediction markets, you can bet on some stupid stuff. Some stupid markets are reasonably liquid, and you can bet large amounts of money on the stupid stuff without moving prices much. And then maybe you can make the stupid stuff happen, so your bets pay off. I wrote a few months ago:

One, uh, accomplishment of modern prediction markets is that they can, in theory, allow for an arbitrarily large amount of betting on arbitrarily small quantities of reality. You can bet on what words someone will say at a press conference, and make tens of thousands of dollars if you’re right. ...

And so one meta-strategy for trading on prediction markets is to try to find the largest gaps between (1) the amount of money to be made and (2) the amount of underlying reality. And then, you know. Make a big bet, and cause a small change to reality so your bet pays off.

Also, though, the opposite. Prediction markets are meant to be truth machines, to produce socially useful information about the probability of future events. Some of those events are stupid, and most of them are sports, but some of them are important. The classic use case for prediction markets is election betting: You can bet on who will win an election, and the betting prices reflect the wisdom of the crowd about who will actually win. Kalshi and Polymarket have signed up journalistic organizations as partners to use their market prices in their election coverage.

And so another meta-strategy for prediction markets is to find the largest gaps between (1) the amount of money required to move prices and (2) the importance of the underlying reality. And then make a relatively small bet that causes a relatively large move in the market-implied probability of your favorite candidate winning an important election.

And then what? The classic form of market manipulation is that you manipulate a small illiquid market (the election contract) to make money in a bigger market. I am not sure what the bigger market is. You could imagine some story like “I will short bonds and then bid up the probability of a socialist candidate winning an important election, which will cause bond yields to spike and make me more money on my short than I spent on the election bets.” Sure. 

Another possibility is: “I will run for office, I will bid up my probability of winning, and then I will go to donors and say ‘hey I’m in the lead’ and they will give me money.” I guess you use the money to, like, pay your buddies consulting fees, and also to pay for the prediction-market bets.

But most people assume that you would do this manipulation, not to make money, but to increase the probability of your preferred candidate actually winning. Raising money is useful in politics! Having good news stories written about you is useful! If the prediction markets say you are likely to win, that can get you money and coverage and support, which will make you more likely to win in real life. And then you can be president or whatever.

That is the theory. I am never sure how seriously to take this. It relies on election markets being inefficient: The theory is “you can spend a little money to move the election odds in your favor, because nobody will trade against you to make an easy profit from your manipulation.” The basic premise of prediction markets is that that’s not true. The premise of prediction markets is that there are enough economically motivated sharp bettors that, if you spend some money trying to puff up a no-hope candidate’s probability of winning an election, a few sharp bettors will say “lol that’s free money,” bet against you, and quickly push the probability back to its correct level.

Still there must be at least some cases where the sharps are asleep or underfunded. There are limits to arbitrage in real markets; they must be more binding in prediction markets. Bloomberg’s Liam Vaughan reports:

After a couple of weeks teasing out his candidacy, Matt Mahan, San Jose’s Democratic mayor, officially entered California’s gubernatorial race on Jan. 29. It was a crowded field, but the unassumingly handsome, 43-year-old centrist had a lot going for him. …

His campaign received a helping hand from Polymarket, one of the world’s largest prediction markets and, increasingly, a driver of political discourse in the US and beyond. That morning a user named Fine-Tributary spent $44,000 buying “yes” contracts for Mahan — that is, contracts that would pay out if the candidate won. It was a massive bet relative to the size of the market so early in the race, pushing Mahan’s price from the low teens to 96¢ out of a maximum of $1. In prediction markets, price is understood to equate to odds, meaning Polymarket, for the moment, was indicating a 96% chance Mahan would win the race. The odds fell back as switched-on traders hoovered up “no” contracts in response, but the following day Polymarket still gave Mahan a 36% chance of replacing Gavin Newsom. Traditional polls had him around 4%. …

That afternoon the New York Post and its new sister title, the California Post, ran a story about the Polymarket spike under the headline “Matt Mahan records huge surge in support of California governor run.” It stated “betting markets” were “predicting tough-on-crime Dem with Silicon Valley ties has what it takes” to defeat an “underwhelming field of lefty Democrats.” The article didn’t mention that the move was driven almost entirely by a solitary, anonymous wallet that had placed only one other trade on Polymarket, for $500, five years earlier (and it hasn’t traded since the Mahan bet). …

Mahan didn’t respond to multiple emails and messages left with his office. But people in his orbit clearly understood the platform’s potential. Tech entrepreneur Adam Kalamchi, Mahan’s college roommate at Harvard University, wrote on LinkedIn: “Whatever you can do to help his candidacy — donate, post, push up the Polymarket bids — please do!” ...

Mahan’s campaign ultimately lost momentum, and after a few weeks his Polymarket numbers fell back in line with the polls. 

Is posting on LinkedIn “push up the Polymarket bids” … commodities market manipulation? 

The basic rule is that, if the directors of a company decide to sell the company for cash, they have to maximize value for shareholders. If one bidder will pay $50 per share, and another will pay $60, the $60 bidder wins. [2] This is called the board’s “Revlon duty,” after the Delaware case about it.

The rule is not quite that, if the directors of a company think it’s only worth $50 per share on its own, and some bidder comes in offering $60, the directors need to sell. But it tends to push in that direction. “The fiduciary duties owed by directors of a Delaware corporation require the directors to seek to maximize the value of the corporation over the long-term for the benefit of the stockholders,” and if the directors are pretty sure that they can’t create more than $60 per share of long-term value, they should sell for $60 in cash.

This can be unsatisfying. When Twitter Inc.’s directors sold out to Elon Musk for $54.20 per share, they seemed sad about it. Twitter’s chief executive officer expressed some doubts that selling to Musk was in the best interests of Twitter, that it would be good for the product or the employees or the users or the advertisers. But he was pretty sure that “this offer at the price it ended up at was in the best long-term interest of our shareholders,” and the shareholders were all that mattered.

There is, however, another kind of company, a “public benefit corporation,” or PBC. The basic point of a PBC is that its directors have different fiduciary duties: They can “manage or direct the business and affairs of the public benefit corporation in a manner that balances the pecuniary interests of the stockholders, the best interests of those materially affected by the corporation's conduct, and the specific public benefit or public benefits identified in its certificate of incorporation.” If Twitter had been a PBC, it could have rejected Musk’s bid by saying “well, this bid is in the best interests of our shareholders, but not of our users or employees or our public purpose as a town square, so, no.” 

I mean, that’s what I always assumed, but the PBC is a somewhat new technology and hasn’t been tested much. For instance: If a PBC’s board puts the company up for sale, does it have to sell to the highest bidder? Surely not? Surely it can sell to the nicest and most public-benefit-y bidder? But Revlon is pretty deeply ingrained in Delaware law and the brains of mergers-and-acquisitions lawyers, so maybe? In any case, yesterday the Delaware Court of Chancery ruled, no, a PBC doesn’t have to sell to the highest bidder:

This case presents an issue of first impression: how Revlon and its progeny apply, if at all, to the board of a public benefit corporation navigating a change-of-control transaction. …

To say that directors of a public benefit corporation “must perform [their] fiduciary duties in the service of [the] specific objective” of “maximizing the sale price of the enterprise,” or that they “must focus on” that “primary objective[,]” would be inconsistent with the public benefit corporation statute’s requirement that directors consider and balance other interests against stockholder pecuniary interests.

I guess that’s pretty obvious, but there you go. 

Here is a history of crypto bankruptcies in three parts. In the first phase, in the early days of crypto, (1) crypto prices mostly went up over time, but (2) crypto exchanges were constantly getting hacked and losing investors’ money. A crypto exchange would have $100 million worth of customer crypto assets, it would get hacked and lose half of its crypto, and it would declare bankruptcy. The bankruptcy process would take a while to sift through all of the claims, and by the time it was ready to pay back its customers, it would have, like, $300 million worth of crypto: Crypto prices had gone up so much that the non-stolen assets were worth considerably more than the dollar value of the customers’ claims. The customers could be made whole (in dollar terms), with money left over. This is approximately the story of Mt. Gox, an early crypto exchange that went bankrupt in 2014.

In the second phase, crypto prices were volatile and people built leveraged financial institutions on top of them. A crypto exchange would have $10 billion worth of customer assets, and then all of a sudden it would have $5 billion of customer assets because of some combination of (1) maybe stealing the money, (2) crypto prices crashing and (3) a self-reinforcing cycle in which crypto prices depended on confidence in the exchange but confidence in the exchange depended on crypto prices. The exchange would declare bankruptcy, the bankruptcy process would take a while, and by the time it was ready to pay back customers, crypto prices would have recovered cyclically and it would have, like, $11 billion worth of crypto. The customers could be made whole (in dollar terms), with money left over. This is approximately the story of FTX, which went bankrupt in 2022, inaugurating a long crypto winter.

In both of these cases, the customers could still reasonably feel aggrieved: If the exchange hadn’t lost their crypto and gone bankrupt, they’d have participated in all of the gains on crypto prices; as it was, they only got their money back in dollar terms at the lows. But that’s life in crypto bankruptcy.

We might now be in a third phase, which goes like this:

• In the good times (pre-2022), crypto was hot, and people built leveraged financial institutions that held customer money and also did some crypto mining because why not.
• Then, for the usual combination of reasons (fraud, volatility, reflexivity), the leveraged financial institutions went bankrupt without enough money to pay back their customers.
• The bankruptcy process took a while.
• These days, eh, crypto, whatever; you can’t rely on a rally in crypto prices to bail out the customers.
• But if you did Bitcoin mining, you’re in luck: Owning a big data center full of graphics processing units, with good access to electric power, is super valuable right now. You can do AI with that! 
• So the customers could be made whole (in dollar terms), out of AI revenue.

CoinDesk reports:

Ionic Digital (IOND), the bitcoin miner formed out of Celsius Network’s bankruptcy, rose 26% on its Nasdaq debut, valuing the company at $2.8 billion following the exchange’s largest direct listing since 2021.

The Washington D.C.-based company, which is pivoting to power AI calculations, was created in January 2024 to acquire Celsius’ mining assets under the bankrupt lender’s court-approved reorganization. … Ionic issued “37 million shares of Class A common stock to eligible holders of certain claims against Celsius Network and its affiliates,” according to its registration statement. ...

Ionic decommissioned bitcoin mining at its Ward County, Texas, site in December and committed its 234 MW of capacity to Nscale under a 126-month lease carrying $1.95 billion in contracted revenue, according to the registration statement.

We talked about Celsius a lot back in the day; I once called it “perhaps the most ridiculous of the big crypto companies.” Mostly the ridiculousness was in its core business, which involved borrowing from customers at 18% interest and yeeting their money directly into the sun. But a small sideline of the ridiculousness was that it got into Bitcoin mining. Which, over the long arc of bankruptcy, worked out for its customers!

World Liberty Financial is a crypto thing whose fundraising pitch had two main elements:

1. It was going to build a decentralized financial blah blah blah, and
2. Most of the money it raised would go directly to Donald Trump.

Which element of this pitch was more important? Man, come on. At New York Magazine, Jen Wieczner writes:

Making a large investment in the project was widely seen as a play to curry favor with Trump. The project was upfront about the fact that 75 percent of net revenues would go to the president’s company DT Marks DEFI LLC (of which Trump owns 70 percent and family members own the rest). It was as though World Liberty had opened up a way to give money directly to Trump. Even in the profiteering world of crypto, an arrangement so lopsided, so nakedly self-dealing, was remarkable.

That said, though, the decentralized blah blah blah part of the pitch was also important. A pitch like “hey give some money to Donald Trump for no reason” raises some pretty obvious issues, while a pitch like “hey give us money and we’ll build a decentralized blah blah blah and Donald Trump will get a cut” raises … uh, really the same issues, but maybe you can look past them? It was very useful to everyone involved in World Liberty to believe that it was building a real business, the decentralized financial system of the future, and not just transferring money to Donald Trump.

Wieczner’s article is about Justin Sun’s feud with World Liberty over his investment. Sun, the crypto billionaire behind the Tron blockchain, was being investigated by the US Securities and Exchange Commission at the time he invested millions of dollars in World Liberty, and now he is not. “Sun’s team has emphatically rejected the notion that his World Liberty investment greased the gears in Washington for the dismissal of his case, suggesting the timing was coincidental.”

But Sun tells Wieczner that he totally believed in the decentralized blah blah blah and was shocked, shocked to discover that World Liberty has not in fact revolutionized the financial system:

Sun would come to believe he’d made a colossal mistake. World Liberty would disappoint his expectations and those of many other investors: The decentralized crypto products it promised would largely never materialize. ... “Today, the thing I regret the most is not only that my money got stuck but that my name is on the project,” Sun says. “I believe many other people also got harmed.” … He couldn’t hide his outrage at World Liberty — for making a mockery of the principles it was purportedly founded on and for, in his eyes, essentially stealing his money in full public view. “I never expected this to happen to me,” says Sun. “This is the worst-case scenario.” …

The “financial revolution” World Liberty promised never materialized. Sun says he gave the World Liberty team ideas for products it could build, but no one followed through on any of them. “I started to notice what I believed to be red flags,” Sun says now. “They seemed to be interested only in raising money, not in developing actual products.”

Uh huh. Wieczner also talks to other people who were less surprised:

“Anyone I know that got involved with the project just did it because they knew Trump was enough of a crook to not let this fail,” says the longtime crypto investor who did not invest. “That anybody could be surprised by what has happened is the only shocking part.” …

“It’s scammers scamming scammers,” more than one crypto investor told me. The former World Liberty adviser said the scenario reminded him of the meme in which two Spider-Mans point at each other. 

Sure. Elsewhere, Bloomberg reports on the fate of a World Liberty Financial treasury company, which is exactly what you’d expect:

Last summer … ALT5 Sigma Corp., a blockchain firm with fewer than two dozen employees, suddenly pivoted to buying virtual tokens issued by World Liberty Financial. … To buy World Liberty tokens, the company raised $750 million from heavyweights including hedge funds Point72 Asset Management and ExodusPoint Capital Management. …

Retail investors snatched up the stock, looking to ride the wave of the new powerful connections. The company, which was worth roughly $100 million in the month before the deal, suddenly cracked $1 billion in market value. Donald Trump Jr. and Eric Trump, the president’s two oldest sons, heralded the transaction by ringing the Nasdaq opening bell.

Nearly a year later, the buzz is gone. ALT5 has changed its name, its CEO and cycled through three auditors. The company now known as AI Financial Corp. is worth less than $60 million and faces getting kicked off the Nasdaq stock exchange. The tokens it bought at 20 cents each have slumped to about 6 cents apiece.

A World Liberty Financial treasury company! Come on. 

NYSE Parent ICE to Purchase MarketAxess for About $6 Billion. A Backlash Against Anthropic Is Brewing in Silicon Valley. Meta shares tumble as Mark Zuckerberg tries to sell his vision for AI ‘agents.’ Thinking Machines Cofounder to Return to OpenAI. ‘My life’s screwed’: Korean investors stress out after AI bubble bursts. Millennium’s New Capital Raise Targets a Record $20 Billion. Secret Talks With Kushner Led to FIFA’s Controversial Sale Plans. The Rise of Million-Dollar Companies With Just One Employee. Aston Martin investors kept in dark over details of asset shift. Kevin Warsh’s stripped-back Fed communication ‘already backfiring,’ say investors. Taco Bell’s Plan to Win Back Customers After Cyclospora Illness Outbreak. $18K-a-summer Maine camp sues private jet firm over problem flights. LeBron sunburn.

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Or you can subscribe to Money Stuff and other great Bloomberg newsletters here. Thanks!