As with all Michael Lewis books, this was a very entertaining read. He’s a great writer.
He met SBF ~2 years before the collapse of FTX, and so was uniquely positioned to write a book about him. There are details in this book that you won’t find anywhere else, and he tells a great story.
Unfortunately, he does seem to be taken in by SBF, and obviously hadn’t come to terms with his guilt by the time he published the book (it was published before the trial had started).
Read the book for the entertainment and then read this detailed book review for the missing pieces (including the guilty conclusion).
Most Sundays, I’d learn, Joe and Barbara hosted a dinner that guests remember fondly to this day. “The conversation was intoxicating,” recalls Tino Cuéllar, a Stanford law professor who would go on to become a judge on California’s supreme court and then head of the Carnegie Endowment for International Peace. “Fifteen percent of it was what was going on in your life, fifteen percent was politics, and the rest was ideas. How we thought about what we thought about—aesthetics, music, whatever.”
The Bankman-Frieds weren’t big on the usual holidays. They celebrated Hanukkah but with so little enthusiasm that one year they simply forgot it, and, realizing that none of them cared, stopped celebrating anything. “It was like, ‘Alright, who was bothered by this fact? The fact that we forgot Hanukkah.’ No one raised their hand,” Sam said. They didn’t do birthdays, either. Sam didn’t feel the slightest bit deprived. “My parents were like, I dunno, ‘Is there something you want? Alright, bring it up. And you can have it. Even in February. Doesn’t have to be in December. If you want it, let’s have an open and honest conversation about it instead of us trying to guess.’ ” Sam, like his parents, didn’t see the point in anyone trying to imagine what someone else might want. The family’s indifference to convention came naturally and unselfconsciously
From the widespread belief in God, and Santa, Sam drew a conclusion: it was possible for almost everyone to be self-evidently wrong about something. “Mass delusions are a property of the world, as it turns out,” he said.
Sam saw some merits in a certain kind of libertarianism. But he listened to actual libertarians argue why, for instance, they shouldn’t need to pay taxes. And he thought, Yeah, of course no one likes to pay taxes, but that’s not exactly a philosophy. “They blurred the line between libertarianism as a philosophy and selfishness as a philosophy,” he said. His internal wiring didn’t carry this particular signal. “The notion that other people don’t matter as much as I do felt like a stretch,” he said. “I thought it would be bizarre even to think about.” It was one thing to feel isolated; it was another to believe that one’s place of isolation was the center of the universe. Or that you and what happened to you were the only things that mattered. “It felt unambitious to not care about what happened to the rest of the world,” said Sam. “It was shooting too low to only think about what was going to impact me.”
As Sam would later explain:
When I was about 12 years old I was first becoming politically aware and started to think through social issues. Gay marriage was a no brainer—you don’t have to be a hardcore utilitarian to see that making people’s lives miserable because they’re completely harmlessly a little bit different than you is stupid. But abortion was nagging me a bit. I was pretty conflicted for a while: having unwanted kids is bad, but so was murder.
Then Sam framed abortion as a utilitarian might. Not by dwelling on the rights of the mother or the rights of the unborn child but by evaluating the utility of either course of action.
There are lots of good reasons why murder is usually a really bad thing: you cause distress to the friends and family of the murdered, you cause society to lose a potentially valuable member in which it has already invested a lot of food and education and resources, and you take away the life of a person who had already invested a lot into it. But none of those apply to abortion. In fact, if you think about the actual consequences of an abortion, except for the distress caused to the parents (which they’re in the best position to evaluate), there are few differences from if the fetus had never been conceived in the first place. In other words, to a utilitarian abortion looks a lot like birth control. In the end murder is just a word and what’s important isn’t whether you try to apply the word to a situation but the facts of the situation that caused you to describe it as murder in the first place. And in the case of abortion few of the things that make murder so bad apply.§
This was how Sam figured out who he was: by thinking about things for himself, without a whole lot of concern for the thoughts of others.
Approximately zero of MIT’s physics majors became physicists anymore. Most went to work for Google, or for high-frequency trading firms. Jump Trading, Tower Research Capital, Hudson River Trading, Susquehanna International Group, Wolverine Trading, Jane Street Capital: all these Wall Street companies Sam had never heard of came to the job fair that year inside the MIT gym. And he became just a little curious about them.
The puzzles that the Jane Street traders gave Sam to solve were designed, like the betting games, to expose blind spots in his mind. The one about baseball was the simplest example. What are the odds that I have a relative who is a professional baseball player? one of the Jane Street traders had asked him.
Sam’s first thought was to define the problem. If you didn’t define the problem, you couldn’t solve it. “That was one thing he was testing for with the question,” said Sam. “Did I realize that the question was ambiguous?” What counted as a “relative”? he asked the trader. What did he mean by “professional baseball players”? Every human being is related to every other human being, in some sense. And lots of people who are not in the major leagues are paid to play baseball. “Relative,” the Jane Street trader said, was any second cousin or closer, and “pro baseball player” included both major and minor leagues but nothing else. Sam guessed there were roughly one hundred baseball teams that fit that definition, and that each had roughly thirty players. So: three thousand active pro baseball players, plus maybe another seven thousand retired ones. Ten thousand players in a population of three hundred million Americans. So: one in thirty thousand Americans had played or were playing pro baseball. Sam didn’t know off the top of his head how many relatives the average American had, but he thought thirty was a reasonable guess. Thus the odds this guy had a relative who played pro baseball were roughly one in a thousand.
The numbers were obviously not exactly right, merely a good enough start. But it was here that Sam paused his mental math and said, I think there’s a decent chance you are asking me this question because it is salient for you—because you have a relative who plays professional baseball.
Here things became tricky. The trader might have anticipated that Sam would have this thought. The trader might have intentionally asked a question that he had no special reason to ask, just to trick Sam. Here was just another aspect of the puzzle: you had to figure out how many levels down you should go before you should stop thinking. Sam decided, as he nearly always did, that more than one level down was too clever by half. It was far more likely that the guy had some reason to ask the question than that he did not. He didn’t know by exactly how much, but the mere fact the trader had asked the question shifted the odds that he had a relative who played pro baseball to something better than one in a thousand. “That was the other thing he was testing for,” said Sam. “Did I realize that there was information in the question I was asked?”
In the end Sam put the odds at one in fifty. And it turned out that the Jane Street trader did indeed have a second cousin who had played professional baseball. But none of that was the point of the problem. The point was how Sam framed it, or failed to. “There were no right answers,” said Sam. “There were only wrong answers.”
By the end of the day of interviews, Sam felt he’d discovered something about himself. “I thought, This is correctly testing for something that matters quite a bit, even if I find it hard to articulate what it is,” he said. Nothing in normal life—not even the games and puzzles that had sustained Sam through childhood—could serve as a proxy for whatever “traders” did at Jane Street. “Childhood doesn’t give you a version of this that would tell you that you are good at it,” said Sam. Childhood had given him math, at which he’d been very good but not great. Childhood had given him various strategic board and card games, at which he’d also been very good but not great. The Jane Street traders had tested his mind for qualities it had never been precisely tested for. And it appeared to Sam that God had tweaked trading in various ways, or at least games intended to simulate trading, to make it different from math and board games. Each of those tweaks had made the games more congruent with his mind. “By the end of the day it was clear that it was by far the best I’d ever done at anything,” he said.
He then showed the students a slide listing the sorts of careers they might pursue, were they to use their careers to save lives. He’d grouped these into four broad categories and offered examples of each: Direct Benefiter (doctor, NGO worker), Money-Maker (banker, management consultant), Researcher (medical research, ethicist), and Influencer (politician, teacher). Eventually, he told the students, you were going to have to choose which sort of career you would pursue. Each career type came with the opportunity to save lives, but the math was different for each, a bit like the math for which hero to play in Storybook Brawl. A Researcher or an Influencer stood a chance of saving some massive number of lives. The agronomist Norman Borlaug (Researcher), for instance, had invented disease-resistant wheat, which had saved roughly two hundred fifty million people from starvation. Researcher and Influencer, however, were tricky career choices: it was difficult to predict who would be good at them, and even harder to forecast their effects. The odds of any given Researcher or Influencer saving vast numbers of lives were vanishingly small.
The clearer choice—the choice MacAskill dwelled on in his talk—was between Direct Benefiter and Money-Maker. Put bluntly: Should you do good, or make money and pay other people to do good? Was it better to become a doctor or a banker? MacAskill made a rough calculation of the number of lives saved by a doctor working in a poor country, where lives were cheapest to save. Then he posed a question: “What if I became an altruistic banker, pursuing a lucrative career in order to donate my earnings?” Even a mediocre investment banker could expect sufficient lifetime earnings to pay for several doctors in Africa—and thus would save several times more lives than any one doctor.
Then he pushed his point a step further, in the direction of the investment banker. “Making a difference requires doing something that wouldn’t have happened anyway,” he said. If you didn’t become a doctor, someone else would take your place and the doctoring would still get done. Of course, if you didn’t become a banker, someone else would also take your place—but that person would spend his money on houses and cars and private schools for his kids and perhaps some non-life-saving donations to Yale. Very little of the replacement banker’s earnings would find its way to doctors in Africa. All those people you might have saved if you had become a banker and given away your money would die. Thus anyone with the ability to go to Wall Street and make vast sums of money had something like a moral obligation to do so—even if they found Wall Street faintly distasteful. “Many lucrative careers are really pretty innocuous,” said MacAskill, helpfully.
Adverse selection was a favorite topic at Jane Street. In this context it meant that the person most eager to make a bet with you is the person you should be most worried about betting against. When people wanted to bet—or trade—with you, there was usually a reason: they knew something you did not. (That they had a second cousin who had played in the minor leagues, for instance.) The first thing you did when someone offered you a bet was to make sure you weren’t missing what they might know. Some piece of information. Some non-obvious angle to the problem. Lots of bets looked stupid after the fact because the person on the receiving end hadn’t thought about why a bet had been proposed in the first place. Jane Street hammered this bitter fact into you every day, and these gambling games were the tool.
The new financial markets had some peculiar properties. For a start, they were increasingly automated. People didn’t trade directly with people. People programmed computers to trade with other computers. Removing humans enabled financial trades to happen faster and more frequently than ever before. Speed became maybe the single most valuable attribute in a trading system. The markets were engaged in a kind of information deforestation—an attempt to reduce to zero the amount of time it took for any piece of information to be registered in the prices of financial assets. “It’s the most complex and efficient game in the world,” said Sam. “More effort has gone into optimizing the game than has gone into anything else.” From the sums of money being extracted from the game—and the sums of money being handed by high-frequency trading firms to US stock exchanges for faster access to their data—you could see that an advantage of a few milliseconds was clearly worth billions a year to whoever possessed it. Whether the speed added anything of value to the economy was another question: Did it really matter if asset prices adjusted to new information in two milliseconds rather than a second? Probably not, but the new technology definitely made it possible for the financial sector to raise the rents that it charged the real economy.
It also changed the kind of person who was extracting those rents. As late as the summer of 2014 you could still see, in the shapes of the bodies on the Jane Street trading floor, the changes that had occurred in financial markets. The older traders, anyone over the age of thirty, were built differently than the younger ones. They were bigger and taller. Their voices were louder. The people who had founded Jane Street back in 1999 were a motley crew of white guys from all over the place. They’d come of age at a time when trading was still done human-to-human, either on trading floors or in trading pits. In a crowd their bodies needed to be seen and their voices needed to be heard. They were also less obviously intellectually gifted. They tended to be quick at mental math but less good at higher-order analytical thinking. On one of those “March of Progress” charts that dramatized the evolution from ape to man they represented perhaps the penultimate stage of Financial Man: hair mostly gone, nearly upright, but still carrying a club on their shoulder, which they used to impose a greater taste for hierarchy on the more egalitarian younger traders.
The younger traders were full Homo sapiens. They’d been harvested from the tiny slice of the population identified early in life as having a gift for higher-order thinking. Many had gone to math camp in high school. Almost all had excelled in computer science or math at MIT, Harvard, Princeton, or Stanford. They were less socially adept than the older traders, because they could afford to be. Now that trading was done machine-to-machine, it mattered less how well traders negotiated with other people. What mattered was their ability to help the machine replace humans in financial markets—either directly, by writing code, or indirectly, by giving instructions that might be codified. To their minds it was silly not to just let the computer do all the mental math.
A trader on the international ETF desk, for instance, might start with the following question: When the price of oil moves during US trading hours, what happens to the ETFs filled with stocks of companies in big oil-producing countries whose markets are closed? If oil prices pop higher during lunchtime in New York City, the shares of, say, Nigerian companies will probably follow them higher—but the Nigerian stock market is closed. ETFs filled with Nigerian stocks, however, are live and trading on US exchanges. Perhaps these US-listed ETFs did not respond as quickly as they should have to movements in oil prices? Perhaps there is a chance to anticipate the rise that will inevitably occur in Nigerian stocks when the Nigerian stock market opens tomorrow? Perhaps others had not thought of this? There was no way to answer those questions without running a study of the historical price movements. Jane Street’s traders spent a lot of their time engaged in these financial research projects.
It wasn’t enough for the trader to make money. You needed to be able to explain why you were making money. A great trader at Jane Street was not a great trader unless he could explain why he was a great trader—and why some great trade existed. As one former trader put it, “It was, Why are you great, and how do we replicate you? And if you could not answer the question, they doubted you.” But these little research projects didn’t need to begin in a dignified way, with some theory about why some market might be inefficient. Often they’d be triggered by some weird event the trader had observed while trading. For example, you might notice, as Sam once did, that exactly twelve hours after the price of certain South Korean stocks rose on the Seoul stock exchange, the price of certain other Japanese stocks rose on the Tokyo stock exchange. Your first thought might be that this is merely a coincidence. But then it keeps happening. You dig into some old data and find that the same thing has been happening in these stocks for several months. You might trade on it—and buy the Japanese stocks the instant the South Korean stocks rise. You might even make money.
You wouldn’t have satisfied the Jane Street system, however, because you didn’t know why the Japanese stocks were rising in price twelve hours after the South Korean stocks. And so you looked even further into it—as Sam had. And he found that the prices of both the South Korean and the Japanese ETFs were being driven by a single trader at a German bank. Every few days, the German bank trader had a bunch of buy orders to execute, in both South Korea and Japan. He’d make his South Korean purchases before calling it a day, passing the Japan orders off to his Asian colleagues to handle when they awakened in Tokyo. The Jane Street trader could now happily see the pop in the South Korean ETF and buy the Japanese ETF until the German died, retired, or figured out how much his laziness was costing him.
Sam found lots of trades whose success turned on the idiocy of some other trader or trading algorithm. Asher trades. For a two-week stretch, Canada’s main stock market index behaved weirdly at the opening every morning. At 9:30 it would pop higher or drop lower with unusual violence and then, at 9:31, revert to its previous levels. That wasn’t how a market normally behaved in response to news. Something else was going on. Sam made a study and discovered that a month earlier, someone had done a massive, multibillion-dollar-contract-sized trade in options on the Canadian stock market index. The trader who had done it needed to hedge his position whenever the price of the Canadian index moved. To do this, the trader had created a bot, which mindlessly bought the Canadian index when it went up, and sold the Canadian index when it went down, thus causing it to go either up or down more than it otherwise would have done. On days the Canadian stock market opened at a higher price than the day before, the bot would buy the index, driving its price even higher, requiring the bot to buy even more. It did the same thing in reverse when the index opened at a lower price than the day before. For two weeks Sam’s trading desk made a small fortune simply by selling the Canadian index after the bot had bought it, and buying the Canadian index after the bot had sold it, until the trader who had created the bot wised up and turned it off. “It was essentially reverse engineering someone else’s dumb algo,” said Sam.
The constant hunt for statistical patterns in markets led to all sorts of strange insights. Every time Brazil won a World Cup match, the Brazilian stock market tanked, for instance, because the win was thought to increase the shot at reelection of Brazilian president Dilma Rousseff, perceived to be corrupt. A faster and better sense of the Brazilian soccer team’s odds in the next match gave you a weighted coin to flip in the Brazilian stock market. In late October of 2016—to take another example—global stock markets were moving around noticeably in response to any news that seemed to alter Donald Trump’s chances of becoming president. At that moment, the upcoming election seemed as if it might be the most consequential election for global financial markets in modern times. The traders on Jane Street’s international ETF desk kicked around ideas about how to trade it. And someone pointed out just how slowly, by high-frequency trading standards, election results found their way into financial markets.
In his annual reviews, his bosses let him know that they had him ranked at the top of his Jane Street class. He wasn’t the firm’s most profitable trader, but he was still young, and doing very well. They’d paid him $300,000 after his first year, $600,000 after his second year, and, after his third year, when he was twenty-five years old, were about to hand him a bonus of $1 million. In his reviews, Sam pressed his bosses to paint a picture of his financial future at Jane Street. It would depend, of course, on Jane Street’s overall performance, they said, but ten years in, if he kept on doing as well as he had been doing, he’d be making somewhere between $15 million and $75 million a year. “Jane Street’s idea was to make people so happy they wouldn’t leave,” said one former trader.
Adam had listened to Sam go on about his beliefs until he finally said, If you really believed all that, you wouldn’t eat meat. At little cost to yourself you reduce a lot of suffering. Sam was serious about minimizing suffering. Sam also liked his fried chicken—but that wasn’t really an argument. “Whatever he said had been rattling around in my head, but I had been avoiding it because of a thought I did not want to have,” said Sam. “The thought was: I spend thirty minutes enjoying chicken and the chicken endures five weeks of torture.” There was nothing to do but overhaul his diet, and he did. “There are easy vegetarians and there are hard ones, and he was a hard one,” said Adam. “It’s unusual to change something like that when it’s difficult.”
Which told you something about Jane Street. In 2104, the year Sam joined Jane Street, Virtu Financial applied to the US Securities and Exchange Commission to sell shares to the public. Its prospectus revealed that in 1,238 days of trading, it had had exactly one losing day. It had just ended a year in which it had made money trading in the stock market every day. How does any firm do that? the reader might intelligently wonder. The answer is outside the scope of this story but is partly addressed in a book I wrote in 2014, called Flash Boys. The point here is that while the high-frequency trading firms appeared interchangeable from a certain distance—their trading was automated, they all acted as intermediaries in financial markets—they differed in how they made their money. Firms like Virtu and Citadel paid US stock exchanges for speed advantages that enabled them to trade against others in the market with very little risk, which explained why they never lost money. They ended each trading day without any positions in the market. Their skill, such as it was, was to create a faster picture of the stock market for themselves than others were able to—which was why, when they went looking for young talent, they wanted computer programmers who could speed their machines more than traders who could make risk decisions. Jane Street had never gotten seriously into the US stock market speed games, and perhaps regretted it. Its relative strength had always been in arguably fairer markets, where they couldn’t simply buy the advantages offered to high-frequency traders by, say, the New York Stock Exchange. If firms like Virtu and Citadel were playing a speed game, firms like Jane Street were playing a brain game.
He’d been on a starting salary of $300,000 a year at Facebook when, after only five months, he’d lost his stomach for the work.
They didn’t blow up, not at first. Those first few weeks, they made no real money, but then they had only a few people and Sam’s bonus money. By the end of December, they’d hired a bunch of people and raised $25 million in capital. Gary, basically all by himself, had written the code for an entire quantitative system. That month they generated several million dollars in profits. In January 2018 their profits rose to half a million dollars each day, on a capital base of $40 million—whereupon an effective altruist named Jaan Tallinn, who’d made his fortune in Skype, handed them $130 million more to play with.
The trading from the start was chaotic. Much of the money they made in their first two months came from just two trades. The frenzied demand for bitcoin created weird distortions in global crypto markets. By December 2017, retail speculators in South Korea were driving bitcoin to prices 20 percent higher than they were on US exchanges, sometimes more. Anyone who could find a way to sell crypto in South Korea and at the same time buy it outside of South Korea could lock in vast profits. It wasn’t trivial to do, however. To open a crypto account on a South Korean exchange, just for starters, you needed to be South Korean. “We found a graduate student friend in South Korea and traded in his name,” recalled Nishad, who now saw why maybe it would take Jane Street a while to export radical efficiency to crypto markets. Jane Street would smell legal trouble; Jane Street would at the very least be embarrassed if it wound up as news in the New York Times that they’d hired a South Korean grad student to front their business. “It was borderline illegal, but in practice, who goes after you when you do this?” said Nishad later. “No one.” This was the very beginning of Nishad’s financial education: there were laws that, in theory, governed money; and then there was what people actually did with money. “That’s where I learned what the law is,” said Nishad. “The law is what happens, not what is written.”
Ex post
Here was one example of the games that were played: Several of the Asian exchanges offered a Bitcoin contract with one hundred times leverage. Every now and then, some trader figured out that he could buy $100 million worth of bitcoin at the same time he sold short another $100 million worth of bitcoin—and put up only a million dollars for each trade. Whatever happened to the price of bitcoin, one of his trades would win and the other would lose. If bitcoin popped by 10 percent, the rogue trader collected $10 million on his long position and vanished—leaving the exchange to cover the $10 million he’d lost on his short. But it wasn’t the exchange that covered the loss: the exchange did not have the capital to cover the loss. The losses were socialized. The customers—usually those on the winning end of the trade—paid for them. And the losses could be huge. The Chinese-founded exchange Huobi had one loss so big that it docked all traders on the profitable end of trades nearly half their gains.
CZ was Changpeng Zhao, CEO of the crypto exchange Binance. Born in Jiangsu province, he was raised, from adolescence, in Canada, and educated there, returning eventually to China, with Canadian citizenship.
A futures exchange was different in important ways from a spot exchange. On a futures exchange, traders put up as collateral only a fraction of the bets they made. On a trade that was losing money, the exchange typically asked for more collateral at the end of the day. If a trade went bad fast, it could wipe out the collateral and leave the exchange on the hook for losses—whereupon the exchange turned to its customers to cover the losses, as crypto exchanges had, historically. The design that FTX (Gary) had come up with solved the problem, in an elegant way. It monitored customers’ positions not by the day but by the second. The instant any customer’s trade went into the red, it was liquidated. This of course was unpleasant for the customers whose positions went into the red. But it promised to do away with socialized losses, which had plagued crypto exchanges from the beginning. The new exchange’s losses would never need to be socialized, because the exchange would never have losses.
Somewhere between Goldman Sachs and Facebook, Ramnik had given up looking for passion in his work. If he seemed older than he was, it was because he was letting go of one of the things that defines youth: hope. “The smartest minds of our generation are either buying or selling stocks or predicting if you’ll click on an ad,” he said. “This is the tragedy of our generation.” The effect of the tragedy had been to shrink his ambition. He was thinking less and less about changing the world and more and more about making himself and his wife comfortable within it. “I’d read a study that there was a fifteen percent happiness gain if you could walk to work,” he said.
“How do you determine something is credible?” he said. “It’s by association. Trust comes from preexisting relationships.”
The question of who regulates any given crypto product in the United States turns on whether the product is defined as a security or a commodity. Bitcoin was defined early on, in 2015, as a commodity, and so is regulated by the CFTC. FTT—or for that matter a leveraged Bitcoin token—would likely be defined as a security and so fall under the jurisdiction of the SEC.
Selling a new business to a VC was apparently less like selling a sofa than it was like pitching a movie idea. The VCs’ eagerness to buy turned less on your hard numbers than on how excited they became about the story you told. It was as if they spent their day listening to stories and picking the ones they liked best. There was no rhyme or reason to their evaluations: English class all over again. Sam quickly figured out that most of the stories these people heard were just not very persuasive. “Most people tell stories that are trivial to falsify,” he said. The story he and Ramnik told was not that way. It went roughly as follows:
Global stocks traded $600 billion a day, crypto was now trading $200 billion each day, and the gap was closing. Inside of eighteen months, FTX had gone from nothing to the world’s fifth-biggest crypto exchange, and every day, it was seizing market share from its competitors. They were now the only crypto exchange making a priority of obtaining licenses and going legit. They were also the only crypto exchange that hadn’t in one way or another offended US financial regulators. Once licensed in the United States, a crypto exchange like FTX could also trade stocks or anything else people wanted to trade and challenge, for example, the New York Stock Exchange. To that end, Sam had already incorporated a business called FTX US—but was being careful about allowing customers to trade stuff on it of which the SEC might disapprove. “The argument was, ‘Look how fast we’re growing, the market is huge—and we’re going to be the credible party,’ ” said Ramnik.
Even if the VCs didn’t all realize that Sam was playing a video game at the same time that he was talking to them, most sensed that he didn’t care what they had to say. Ramnik came to think that Sam’s indifference to their feelings actually heightened their interest in him. That FTX was profitable and didn’t really need the money probably also helped. In the end, between the summer of 2020 and the spring of 2021, in four rounds of fundraising, they sold roughly 6 percent of the company for $2.3 billion. Roughly one hundred fifty different venture capital firms invested. All of them caved to Sam’s refusal to give them a seat on the board (he had no board) or any other form of control over the business
And yet FTX was just a piece of a much larger puzzle of Sam’s design. He owned 90 percent of Alameda Research. And the nature of Alameda Research was changing. It was still a quant trading firm, with its good months and bad months, but its traders were playing, in new ways, with bigger and bigger sums of money. Crypto world had created what were, in effect, unregulated new banks. People would deposit their crypto with, for instance, Genesis Global Capital or Celsius Network, and receive some rate of interest, and these pseudobanks would re-lend the crypto to traders like Alameda Research. In early 2018, rich effective altruists had charged Sam interest rates of 50 percent a year. Three years later, Genesis and Celsius were willing to lend billions to Alameda Research at interest rates that ranged from 6 to 20 percent. And there were other, even more mysterious billions inside of Alameda that no one knew about. “FTX is smaller than people think, and Alameda is bigger,” said Ramnik. “Way bigger.”
It was never clear where Alameda Research stopped and FTX started. Legally separate companies, they were both owned by the same person. They occupied the same big room on the twenty-sixth floor of an office building. They shared the same vista of the forest of high-rises surrounding Victoria Harbor and, twenty miles beyond that, China. Sam’s desk was positioned at one end of the identical long trading desks used by both Alameda and FTX, with a clear view of both. It didn’t occur to anyone that there was anything weird about Alameda covering the start-up costs of somewhere between $5 million and $10 million for FTX. Ditto FTX selling FTT and using the capital not to expand FTX but to trade inside Alameda. It seemed perfectly natural for Alameda to control all the remaining FTT, and use it as collateral in its trading activity. Sam didn’t even try to hide what he was doing. FTT “has single-handedly fixed [Alameda’s] equity problem,” he wrote, in a memo to employees. He retained 90 percent of Alameda Research, with Gary owning the remaining 10 percent. Even after he’d sold stakes in FTX to the one hundred fifty venture capitalists, Sam still owned more than half the company. The third-largest shareholder, Nishad, owned a mere 5 percent of the company.
Caroline didn’t like it and let him know, in a series of long, businesslike memos. “There are things I want out of our relationship that I feel like I’m not getting to the extent I want,” she wrote, in early July 2021. The usual bullet points followed:
Matt Levine’s excellent forty-thousand-word article in Bloomberg Businessweek, “The Crypto Story.”
For example, they all professed to care about “humanity,” while at the same time often being a bit slow to love actual people. “It doesn’t really start with people,” said George. “It starts with suffering. It’s about preventing suffering. They care about animals in the same way. They also care about not having the earth blown up by an asteroid. But it’s not a longing for a connection.”
They also cared about the logic underpinning their behavior; consistency was, for them, not the hobgoblin of a little mind but the mark of a big one. They brought logic and rigor to their most emotionally fraught decisions—for instance, the decision whether to have children. “A lot of EAs chose not to have kids,” said George. “It’s because of the impact on their own lives. They believe that having kids takes away from their ability to have impact on the world.” After all, in the time it took to raise a child to become an effective altruist, you could persuade some unknowably large number of people who were not your children to become effective altruists. “It feels selfish to have a kid. The EA argument for having a kid is that kid equals happiness and happiness equals increased productivity. If they can get there in their head, then maybe they have a kid.”
The rate of return inside Alameda was steadily declining, but with access to vast amounts of cheap capital, its raw trading profits kept rising: from $50 million in 2018, to $100 million in 2019, to $1 billion in 2020 and again in 2021. And those were just the trading profits; the numbers didn’t include the seemingly vast unrealized gains on Sam’s token stashes.
In March 2020, a Silicon Valley engineer named Anatoly Yakovenko launched a new and better blockchain that offered a solution to maybe Bitcoin’s biggest weakness as a means of exchange: it was way too slow. Bitcoin could only validate seven transactions a second. The new Solana blockchain promised to process up to sixty-five thousand transactions a second. Sam had no independent ability to evaluate the thing, but he asked people who did and soon decided that Solana might be the crypto of the future. Even if it wasn’t, Solana’s story was good enough that other people might see it that way and drive up the price of its token. Eighteen months later, Alameda owned roughly 15 percent of all Solana tokens, most purchased at twenty-five cents apiece. The market price of Solana had gone as high as $249, a thousand-times increase on what Sam had paid for the tokens, and the face value of Sam’s entire stash was roughly $12 billion. It was hard to know the resale value of such a huge stake. But there was a real market for Solana tokens. Two billion dollars’ worth of the things traded each day. “I watched it in wonder,” said Sam.
There was more like this in Sam’s dragon’s lair. Alameda had also hoovered up about half the existing FTT tokens, for instance, creating, in effect, a second stake in FTX for Sam, with a claim on about one-sixth of all FTX’s revenues. In the past eighteen months, FTT’s price had risen from roughly $3 to roughly $80. Again, it was hard to say how much Sam could unload FTT for, had he tried to sell his stake all at once. But the crypto lenders were happy to lend him billions of dollars against his FTT, so, to get his hands on cash, he didn’t need to sell it.
Then there was Sam’s equity stake in FTX, which was very real indeed. A large number of venture capitalists had paid $2.3 billion for a mere 6 percent of it. Sam had good reason to believe that he might now sell an even smaller piece for several billion more to an even bigger group. FTX underpinned his growing empire: a real business with booming revenues and profits. It hadn’t even really needed venture capital. (As if to make the point, Sam, having taken $200 million from Sequoia Capital in exchange for a piece of FTX, turned around and invested $200 million from Alameda Research in one of Sequoia’s funds.) FTX was now the world’s fastest-growing crypto exchange, and the casino of choice for big professional traders. In less than three years, it had gone from 0 to 10 percent of all crypto trading. In 2021, it would generate $1 billion in revenues.
The US was now, in Sam’s mind, the holy grail. It had an incumbent crypto exchange, Coinbase. But Coinbase’s CEO had already written insulting tweets about the SEC. And Coinbase, compared to FTX, was a boring and bloated casino. It had fifteen times the number of employees FTX did, and only about a fifth of FTX’s volume. Charging retail investors fees between five and fifty times what FTX charged, it was still running big losses. Even so, it was a public company, with a market capitalization of more than $75 billion. If FTX was granted a license to offer crypto futures in the United States and was given full access to US investors, it might steal Coinbase’s customers, along with its market cap. Or so Sam thought—which is why he also thought that the license might double or even triple FTX’s value overnight.
Wash trading, as it was called, would have been illegal on a regulated US exchange, though the sight of it did not bother Sam all that much. He thought it was sort of funny just how brazenly many of the Asian exchanges did it. In the summer of 2019, FTX created and published a daily analysis of the activity on other exchanges. It estimated that 80 percent or more of the volume on the second- and third-tier exchanges, and 30 percent of the volume on the top few exchanges, was fake. Soon after FTX published its first analysis of crypto trading activity, one exchange called and said, We’re firing our wash trading team. Give us a week and the volumes will be real. The top exchanges expressed relief, and gratitude for the analysis, as, until then, lots of people assumed that far more than 30 percent of their volume was fake.
“We’re currently ahead in tech and favorability ratings, and behind in name recognition,” he wrote. “We need to get 50m low engagement users to decide to switch from Coinbase to FTX. This will take a fairly forceful nudge!”
He began by noting how few marketing campaigns had had the effect that he hoped to achieve with FTX’s. He counted only three:
Once their name was on an American stadium, no one turned down their money.¶ They showered money across US pro sports: Shohei Ohtani and Shaquille O’Neal and LeBron James became spokespeople. They paid Major League Baseball $162.5 million to put the company name on every umpire’s uniform. Having the FTX logo on the umpires’ uniforms, Sam thought, was more useful than having it on the players’ uniforms. In basically every TV shot of every Major League Baseball game, the viewer saw the FTX patch. “The NBA put us through a vetting process,” said FTX lawyer Dan Friedberg. “Major League Baseball just said okay!”
You would think—Sam had initially thought—that if you were going to pay some NFL quarterback to stand up and say he used FTX, it would make little difference whether it was Tom Brady or Aaron Rodgers or Dak Prescott. You might see that Brady would be a bit better but think Rodgers’s endorsement must be worth, say, half of Brady’s. But everywhere Sam went, people mentioned that they had heard of FTX because of Brady. Hardly anyone mentioned any of the other endorsers. “It was very clear which things had an effect and which did not,” said Sam. “For the life of me, I can’t figure out why this is. I still don’t know how to verbalize it.” The Martian had discovered another weird but true fact of modern human life: at any given moment there were only a few people inside the collective imagination.
In the end, George drew up the only internal org chart ever made of Sam’s sprawling creation. By the time he was done, he’d discovered many interesting things. Twenty-four different people thought that they were reporting directly to Sam, for example. The group included Sam’s father, Joe, and Sam’s childhood friend Matt Nass, whose game, Storybook Brawl, Sam had for some reason just bought. This group did not include the chief financial officer, because FTX did not have a chief financial officer. There was no chief risk officer or head of human resources, because they had none of that, either. “It seems like a clubhouse more than a corporation,” said George.
It did have a chief technology officer, Gary Wang, but Gary was socially isolated, with no one reporting to him. “Gary was off in his own little box,” said George. In an ordinary tech company, a bunch of programmers would be reporting to the chief technology officer. In FTX they apparently all reported to Nishad Singh. Ryan Salame, who had come and gone from the Bahamas in a flash and now seemed hardly involved with the company, was somehow the CEO of the entire international business, with twenty-seven people reporting to him. Ramnik Arora, whose official title was still Head of Product, clearly had nothing to do with product but sat on top of a small pile of people in charge of both raising and investing vast sums of money. George just put him in a little box marked “Ventures.” About half the entire company reported to the first two young women Sam had hired upon his arrival in Hong Kong, Constance Wang and Jen Chan. Most of those people, George noted, were East Asian women.
Then there was Caroline Ellison. Caroline was apparently alone in charge of the twenty-two traders and developers working inside Alameda Research, about half of whom had followed Sam from Hong Kong to the Bahamas. This surprised George a bit. “She never said anything about Alameda,” said George. “Neither did Sam. This was clearly wanting not to think about it.”
Inside of three years, Sam would deploy roughly $5 billion on a portfolio of three hundred separate investments—which worked out to a new investment decision roughly every three days.
He’d invest in new crypto tokens, like Solana, and old companies, like Anthony Scaramucci’s investment firm, SkyBridge. He’d acquire companies obviously relevant to FTX—a Japanese crypto exchange called Liquid, for instance—and companies that had no obvious connection to crypto, like the studio that had developed Storybook Brawl. The money nearly always came not from FTX but from Alameda Research, which Ramnik and everyone else thought of as Sam’s private fund.
Often Ramnik was intimately involved with a purchase, but nearly as often he only learned what Sam had done after the fact. Sam had invested $500 million in an artificial intelligence start-up called Anthropic, apparently without bouncing the idea off anyone else. “I said to Sam after he did it, ‘We don’t know a fucking thing about this company,’ ” said Ramnik. About the same time that he was trying to decide whether to sink more money into Twitter, Sam was handing $450 million to a former Jane Street trader named Lily Zhang, to create a second crypto quant trading fund based in the Bahamas, called Modulo Capital. So far as Ramnik could see, Sam had told no one about that until he’d done it. Back in March, Sam had promised to invest $5 billion with a Hollywood agent turned investment manager named Michael Kives, without consulting Ramnik or anyone else. Sam had only met Kives a few weeks before he’d made that commitment. He’d known nothing about him, not even how to pronounce his name.
His political spending was distributed sloppily into three buckets. The first, and smallest, bucket contained his narrow business interests: a few million dollars donated to politicians and interest groups willing to push for legislation that would allow Americans to trade the crypto contracts on FTX inside the United States that foreigners did on FTX outside of it.
In Sam’s mind, his money wasn’t crypto money. It was effective altruist money that he happened to have obtained through crypto. Along with his brother, Sam had looked at the world and decided that two EA-related causes made more sense to address with his money than any of the others. And that a lot of the money needed to be sneaky.
Their first, less sneaky initiative was pandemic prevention. On the list of existential risks to humanity, pandemics occupied a special place. Unlike, say, an asteroid strike, the threat felt real, and politicians could be persuaded to take it seriously. Unlike, say, climate change, hardly anyone was talking or thinking seriously about how to address the problem—even after a million Americans had just died from a new pathogen. Unlike, say, preventing a war on humanity by artificial intelligence, there were some obvious though expensive things to do to mitigate the risk. For instance, someone really did need to take the lead in the creation of a global system of disease prediction, one that resembled the global system of weather prediction. Sam guessed that it would take $100 billion, which put it beyond his reach.
Badgering elected officials into taking an interest in pandemics was one part of Sam’s strategy. The second part was getting some new pandemic warriors elected to Congress. Sam’s political operation had figured out, or thought they had, that it made a lot more sense to spend money in primary than in general elections. Voters could be swayed in primaries as they couldn’t in generals. Much of persuasion in primaries was just name recognition, which you could buy with ads. They’d also already figured out, or thought they had, that a million dollars dropped into a close congressional primary gave them a one-in-five chance of swinging it to their candidate. The problem was that they had no way to determine in advance which of the five they’d be able to influence. And so they’d adopted a strategy of finding as many congressional candidates as they could who would support spending on pandemic prevention and buying their elections in bulk, while at the same time doing their best to disguise that the money involved had anything to do with crypto.
Of course, winning one out of five congressional races meant you lost the other four. Sam’s political portfolio resembled his venture capital portfolio: in pursuit of crazy rewards, it took what, after the fact, looked like insane risks. In a very short time, Sam’s money had bankrolled some of the most spectacular failures in the history of political manipulation.
But the bucket was very much the subtext of the dinner Sam was headed to, because in McConnell, Sam had found someone as interested as he was in another existential threat to humanity: Donald Trump. Trump’s assault on the government, and on the integrity of US elections, belonged, to Sam’s way of thinking, on the same list as pandemics and artificial intelligence and climate change. Across the land, Republican primaries were littered with candidates who were willing to behave as if the presidential election had been stolen from Trump. They faced candidates who were forced to pay lip service to the idea. McConnell’s people had already figured out which was which, and McConnell was intent on defeating the former. “He’s already done the work,” said Sam. The work, he added, was to distinguish “people who would actually try to govern versus people who would undermine the government.”
Once you’re in the argument, however, you’ll find it difficult to escape a certain logic: the expected value of reducing even the minuscule likelihood of an existential threat to all future human beings is far greater than the expected value of anything you might do to save the lives of the people who currently happen to be alive. “The core argument was like, look, the future is vast,” said Sam. “You can try to put a number on it but obviously anything that flows through to that is going to have a vast multiplier.”
One day some historian of effective altruism will marvel at how easily it transformed itself. It turned its back on living people without bloodshed or even, really, much shouting. You might think that people who had sacrificed fame and fortune to save poor children in Africa would rebel at the idea of moving on from poor children in Africa to future children in another galaxy. They didn’t, not really—which tells you something about the role of ordinary human feeling in the movement. It didn’t matter. What mattered was the math. Effective altruism never got its emotional charge from the places that charged ordinary philanthropy. It was always fueled by a cool lust for the most logical way to lead a good life.
They’d been doing this for only a year and already had been pitched nearly two thousand such projects. They’d handed out some money but in the process they’d concluded that conventional philanthropy was kind of dumb. Just to deal with the incoming requests—most of which they had no ability to evaluate—would require a big staff and lots of expense. Much of their money would end up being used on a vast bureaucracy. And so they had just recently adopted a new approach: instead of giving money away themselves, they scoured the world for subject matter experts who might have their own, better ideas for how to give away money. Over the previous six months, one hundred people with deep knowledge of pandemic prevention and artificial intelligence had received an email from FTX that said, in effect: Hey, you don’t know us, but here’s a million dollars, no strings attached. Your job is to give it away as effectively as you can. The FTX Foundation, started in early 2021, would track what these people did with their million dollars, but only to determine if they should be given even more. “We try not to be very judgy once they have the money,” said Sam. “But maybe we won’t be reupping them.” The hope was, first, that these people on the ground would know better than anyone what to do with the money and, second, that some people might actually have a genius for giving away money. “It’s trying to blast through the hesitation,” said Sam. “The default to inaction.”
It did the opposite. A risk analysis company called Gauntlet, which studied the price movements of various crypto tokens, had maybe the best picture of what actually happened next. Within twenty seconds of Caroline’s tweet came a rush to sell FTT by speculators who had borrowed money to buy it. The panic was driven by an assumption: if Alameda Research, the single biggest owner of FTT, was making a big show of being willing to buy a huge pile of it for $22, they must need for some reason to maintain the market price at $22. The most plausible explanation was that Alameda Research was using FTT as collateral to borrow dollars or bitcoin from others. “You don’t tell someone a price level like $22 unless you have a lot of confidence that you need that price,” the CEO of Gauntlet, Tarun Chitra, told Bloomberg News. By Monday night, the price of FTT had fallen from $22 to $7. The half a billion dollars of his own money that CZ had elected to incinerate was, in the grand scheme of things, such a trivial sum that hardly anyone paid it any more attention.
By Tuesday, the relevant math was fourth-grade level. Before the crisis, FTX was meant to be holding about $15 billion worth of customer deposits.* Five billion of that had already been paid out to customers, and so, still inside FTX, there should have been roughly $10 billion. There wasn’t. The only remaining assets were whatever was left of the dragon’s hoard inside of Alameda: a big pile of FTT, another big pile of Solana tokens, an assortment of crypto tokens that would be even harder to sell, $300 million worth of Bahamas real estate, and a truly massive heap of Sam’s venture capital investments—including the stake in Twitter, which Sam had never bothered to sell. There was still perhaps as much as $3 billion worth of hard currency and bitcoin that they had yet to return to customers—but the vast majority of the secret stash had no immediate market. Much of what Caroline and Sam went back and forth about the first two or three days of the run on FTX was just this. Caroline, who by then was beaming in from the Hong Kong office, would appear on a video call. Sam would go down the list of the many things either he or she had bought and ask: How long will it take you to sell this? For most, the answer was too long.
Though Caroline was in charge of Alameda Research, she seemed totally clueless about where its money was. She’d come onto the screen and announce that she had found $200 million here, or $400 million there, as if she’d just made an original scientific discovery. Some guy at Deltec, their bank in the Bahamas, messaged Ramnik to say, Oh, by the way, you have $300 million with us. And it came as a total surprise to all of them!
Eventually Ramnik gathered that they needed to raise $7 billion, fast, to fill what they thought might be a $7 billion hole. (The exact number shifted around a lot those first few days.) To his obvious question—Why was there a hole in the first place?—Sam and Nishad and Caroline offered fuzzy answers. Gary sat quietly off to one side.
Caroline thought in periods but spoke in question marks and exclamation points. She made the sounds of uptalk and uncertainty while delivering a message that was brutally simple: they were bankrupt.
On Friday, Nishad was gone, which was just as well, as by then the police in the Bahamas were preparing to arrest any remaining leaders. That afternoon, roughly $450 million in crypto vanished from the wallets inside FTX. No one knew who the hacker was; everyone just assumed it was an inside job; lots of people suspected Sam and Gary.
Still, Zane figured there was no way that FTX was in real trouble. It made no sense. The price of FTT shouldn’t have any effect on the value of the exchange, any more than the price of Apple stock should have on Apple’s iPhone sales. Just the reverse: the exchange’s revenues drove the value of FTT. “If FTT goes to zero, so what?” said Zane. The other reason it made no sense was that FTX had been so wildly profitable. “I know how much real revenue we were making: two bips [0.02 percent] on two hundred fifty billion dollars a month,” said Zane. “I’m like, Dude, you were sitting on a fucking printing press: why did you need to do this?”
To Zane it didn’t matter. There was one question he dwelled on: Why had neither he nor anyone else he knew seen this coming? He had the beginning of an answer. “Sam’s oddness,” he said. “His oddness mixed with just how smart he was allowed you to wave away a lot of the concerns. The question of why just goes away.”
And there was a third argument, made by Sam, that anything that happened would need to happen wherever Gary was, because Gary was the only one who could explain the code that had governed the business. “At the end of the day, the deciding factor in the jurisdictional dispute is Gary,” said Sam, the night Zane left, “because he’s the only one who knows how to use a computer.”
The list of assets included the details of hundreds of private investments Sam had made over the previous two years, apparently totaling $4,717,030,200. The liabilities now had a line item more important than everything else combined: $10,152,068,800 of customer deposits. More than $10 billion that was meant to be custodied by FTX somehow had ended up inside Sam’s private trading fund. The document listed only $3 billion in liquid assets—that is, US dollars or crypto that could be sold immediately for dollars. “I was like, Holy shit,” she said. “The question is: Why?” It was the same question Zane had asked. “We had so profitable a business,” said Constance. “Our profit margin was forty to fifty percent. We made four hundred million dollars last year.”
At the moment of its collapse, FTX had had more than ten million account holders, to whom it owed $8.7 billion. Nearly half of those losses, or $4 billion, were concentrated in these fifty accounts. The biggest losers not employed by either FTX or Alameda were high-frequency trading firms. Near the top was Jump Trading ($206,160,600.00), and at the bottom was Virtu Financial Singapore ($10,095,336.83). The real names of about half the list were concealed. The entity listed as Tai Mo Shan Limited—and out more than $75 million—was actually another affiliate of Jump Trading. Many of the disguised accounts belonged to FTX employees. Constance herself had lost around $25 million. She still had $80,000 in an ordinary bank account she’d kept from her previous life, but otherwise she’d lost everything.
As she had also overseen the sales team, she knew most of the names on the list, especially the high-frequency traders. She knew that every one of them had been intensely suspicious about the relationship between FTX and Alameda Research. “Everyone cared about it,” said Constance. “It was literally the first thing I was asked every day. Is Alameda Research front-running us? Does Alameda Research get to see other people’s trades? Does Alameda get less latency?” In other words: Did Alameda enjoy the same unfair trading edge on FTX that the high-frequency traders enjoyed on Nasdaq and the New York Stock Exchange? Oddly enough, it had not. Instead, FTX had simply loaned Alameda all of the high-frequency traders’ deposits…for free!
FTX had also done other things to jeopardize the high-frequency traders’ money, along with everyone else’s. It had exempted Alameda from the risk rules that governed all the other traders. The trades made by every other trader on FTX were liquidated the moment their losses exceeded the collateral they had posted. That’s why FTX felt so much safer than the other crypto exchanges. No single trader was allowed to lose so much money that it put the exchange, and everyone who traded on it, at risk. For Alameda Research, however, an exception had been made. Sam’s private trading firm was allowed to lose, in effect, infinity dollars before its trades were liquidated. “No one ever asked about liquidation,” said Constance. “And no one ever asked, ‘Is our money actually inside Alameda?’ ” Sam was right: People don’t see what they aren’t looking for.§
The story Sam told Constance ran as follows. There had been two different ways that money that should have been in cold storage inside FTX instead wound up in Alameda’s hot little hands. The first was through Alameda’s normal trading activity. Like every other trader, Alameda had been allowed to borrow from the FTX exchange by posting collateral. As collateral, Alameda had used, among other things, FTT—the token that was, in effect, equity in FTX. The price of FTT had collapsed with FTX. The collateral was now worthless, and some of the loans remained unpaid. In Sam’s story, there was a reason that Alameda had been exempted from the rules that governed every other trader on FTX, and had liquidated their trades when the losses exceeded the value of their collateral. Back in 2019, when FTX was created, Alameda was by far its biggest trader. At the start, Alameda was on the other side of most trades that occurred on FTX. It helped the market on the exchange to work better if Alameda could occasionally run losses—for example, if they needed to step in and acquire another trader’s losing positions after FTX liquidated them.
In Sam’s telling, FTX had switched off Alameda’s risk limits to make itself more appealing. The losses caused by this unsettling policy were in any case trivial. Ordinary trading loans made by FTX to Alameda constituted a small fraction of the losses to customers; on their own, they wouldn’t have posed a problem. The bulk of the customers’ money inside of Alameda that should have been inside FTX—$8.8 billion of it, to be exact—resided in an account that Alameda had labeled fiat@.
The fiat@ account had been set up in 2019 to receive the dollars and other fiat currencies sent by FTX’s new customers. Alameda Research had created the account only after FTX had been unable to get its own bank accounts. Back in 2019, no real bank in the United States had been willing to offer its services to a new international crypto exchange. The crypto entities that they did bank, like Alameda Research, usually disguised their association with crypto. The biggest US crypto exchange, Coinbase, had by some miracle persuaded Silicon Valley Bank to give it an account—and thus a mechanism for Coinbase to receive US dollars from, and send US dollars to, its crypto trading customers. A US bank account had thereby given Coinbase a big advantage, but how exactly they’d obtained the account was a story for another day; the story for this day is how FTX failed to find its own US bank for sending and receiving dollars. From its founding in the spring of 2019 until July 2021, when it finally persuaded a bank in San Diego called Silvergate Capital¶ to open an account in its name, FTX had no straightforward way to accept dollar deposits.
In Sam’s telling, the dollars sent in by customers that had accumulated inside of Alameda Research had simply never been moved. Until July 2021, there was no other place to put them, as FTX had no US dollar bank accounts. They’d been listed on a dashboard of FTX’s customer deposits but remained inside Alameda’s bank accounts. Sam also claimed that, right up until at least June 2022, this fact, which others now found so shocking, hadn’t attracted his attention. He wasn’t managing Alameda Research; Caroline was. Toward the end of 2021, when the flow of new dollars into the fiat@ account trickled to nothing—as customers could now deposit their dollars directly onto FTX, through a US bank—Alameda Research had a net asset value of $100 billion. That number was of course wildly unreliable, as it was simply the market value of lots of cryptocurrencies for which the market might vanish, if Alameda tried to sell into it. But even if you valued the contents of Alameda more rigorously, as Sam sort of did in his head sometimes, you could still easily get to $30 billion. The $8.8 billion that should not have been inside Alameda Research was not exactly a rounding error. But it was, possibly, not enough to worry about. As Sam put it: “I didn’t ask, like, ‘How many dollars do we have?’ It felt to us that Alameda had infinity dollars.”
That feeling would have changed by late spring of 2022. Between the start of April and the middle of June, the price of a bitcoin fell from just over $45,000 to under $19,000. Entering that summer, the relative importance to Alameda of the $8.8 billion had skyrocketed. But Sam wasn’t managing the risk inside Alameda Research, according to him. Caroline was. Perhaps because he and Caroline were barely speaking by that point, she hadn’t bothered to raise, directly with him, her worries about the risks she’d been running.
Right up until October 2022, in Sam’s telling, he’d had only two brushes with this huge unexplained pool of other people’s money that had accumulated inside Alameda, and that Alameda was increasingly dependent upon. The first was truly bizarre: in mid-June, Caroline had become alarmed to discover that the fiat@ account had swelled from $8.8 billion to $16 billion. She shared her worries not with Sam but with Nishad, who in turn informed Sam and Gary—whereupon Gary discovered that it was just a bug in the software. The real number in the fiat@ account hadn’t changed: it was still $8.8 billion.
Three months later, in September, Caroline pulled Nishad aside and told him that she was growing more and more worried about Alameda’s market exposure. Nishad had taken Sam out onto the balcony of the Orchid penthouse and relayed the message—but without explicitly mentioning the fiat@ account. At that point, in Sam’s telling, Sam thought that Alameda might be in trouble. He decided to dig into its accounts on his own and understand the problem. By October, he had a clearer picture. It was only then that he could see that Alameda had been operating as if the $8.8 billion in customer funds belonged to it. And by then it was too late to do anything about it.
Constance heard Sam out. She listened to his story. But she refused to believe it. She suspected that he was omitting some big, important fact—say, a sudden trading loss inside Alameda Research that had caused him to actively grab customers’ money and move it into Alameda. “It is crazy,” she said. “He made me try to believe it was an accounting error.” She didn’t know how or why he had consciously decided to take customers’ money and use it as his own, but she felt sure he had.
She’d seen how critical Alameda’s willingness to trade anything with anyone at any time had been to the successful launch of FTX. She didn’t even think it fishy that a crypto exchange had its own internal trading team. “Most of the exchanges did this,”* she said. “All the Chinese ones. It’s only a matter of how big the trading team is and what they are doing.”
In the previous three and a half years, nearly $9 billion more had entered Sam’s World than had exited it. When FTX stopped returning funds to customers, on Tuesday, November 8, it still had $3 billion on hand. That dropped the missing sum to $6 billion. (The roughly $450 million stolen in the hack three days later is irrelevant to this calculation.)
Eventually we arrived at the question whose answer might offer clues to other puzzles: Where did the money go? It wasn’t the last time I’d ask it. Like Constance, I’d poke and prod and always come away with the sense that I’d learned less than I needed to know. But on that evening, Sam filled in one piece of this particular puzzle: FTX had lost a lot of money to hackers. To avoid encouraging other hackers, they’d kept their losses quiet. The biggest hacks occurred in March and April 2021. A lone trader had opened an account on FTX and cornered the market in two thinly traded tokens, BitMax and MobileCoin. His purchases drove up the prices of the two tokens wildly: the price of MobileCoin went from $2.50 to $54 in just a few weeks. This trader, who appeared to be operating from Turkey, had done what he had done not out of some special love for MobileCoin. He’d found a flaw in FTX’s risk management software. FTX allowed traders to borrow bitcoin and other easily sellable crypto against the value of their MobileCoin and BitMax holdings. The trader had inflated the value of MobileCoin and BitMax so that he might borrow actually valuable crypto against them from FTX. Once he had it he vanished, leaving FTX with a collapsing pile of tokens and a loss of $600 million worth of crypto.
The size of those hacks was an exception, Sam said. All losses due to theft combined had come to just a bit more than $1 billion. In all cases, Gary had quietly fixed the problem and they’d all moved on and allowed the thieves to keep their loot. “People playing the game,” was Sam’s description of them. (He really was easy to steal from.)
The hacks reduced to $5 billion the number of unexplained missing dollars. Sam was no help in reducing the number further. Either he didn’t know where the money had gone or he didn’t want to say. He dismissed the most obvious explanation: Alameda had suffered some big trading loss in the great crypto crash of 2022. The collapse of FTX felt a bit like the case of the missing Ripple, but on a far grander scale. This time the question of where the money was would take longer to answer, and the person most qualified to figure it out was soon gone.
Nardello firm was a lot of former FBI guys. (Corporate motto: We find out.) Chainalysis, the crypto sleuthing firm
That was in early 2023. By late April, John Ray’s head was on a swivel. “This is live-action,” he said. “There’s always something every hour.” One day, some random crypto exchange got in touch and said, By the way, we have $170 million in an account of yours: do you want it back? Another day, some random FTX employee called them out of the blue to say that he’d borrowed two million bucks from the company and wanted to repay the loan—of which, so far as Ray could see, there was no record.
Several months into the hunt, Ray’s sleuths had discovered that “someone had robbed the exchange of four hundred fifty million.” They’d stumbled upon not the simple hack of November 2022 but the complicated BitMax and MobileCoin hacks of $600 million in the spring of 2021. (The dollar value changed with fluctuations in the price of the stolen crypto.) They’d tracked the hacker not to Turkey but Mauritius. “We have a picture of him going in and out of his house,” said Ray. He was pretty sure he was going to get most of that money back. “We believe there are a lot more of these,” said Ray. Eventually, I figured, he’d find the $1 billion or so lost in hacks that Sam would have just told him about, if he’d been willing to talk to Sam.*
Once Ray’s people had finished doing the math, they concluded that FTX still owed its customers $8.6 billion.
To wring the money out of the people Sam had hurled it at, John Ray needed to prove two things. The first was that FTX hadn’t received equivalent value for its money. You couldn’t claw back money from a plumber who’d been paid some normal sum to unclog an FTX drain. But you could claw back money from the researcher to whom FTX had handed a grant to invent drains that never clogged. Simply not getting value for money was not enough for Ray to get it back, however. He also had to prove that at the moment Sam gave money away, it wasn’t his money to give. And the only way it wasn’t Sam’s money was if FTX, in the moment he gave away the money, was insolvent, or nearly so. Ray’s various attempts to claw back money raised an interesting question, which his team had yet to answer in an intelligible way: At what point was there less money in all of Sam’s World than was supposed to be inside of FTX? Exactly when did FTX go broke?
Instead of answering the question, Ray launched a blitzkrieg of lawsuits against various people to whom Sam had handed money. These were really fun to read. They were still legal texts, but they all had subtexts. Plus, Ray wrote to attract media attention. “You have to tell a story,” explained Ray. “Nobody wants to read X dollars wired to Y, blah, blah, blah. You need the imagination of a child to write this stuff.” In the first eight and a half months, he filed nine of these clawback lawsuits. Ray mainly targeted insiders—Sam, Sam’s parents, Caroline, Nishad, and so forth—or people to whom Sam had forked over huge sums to invest on his behalf.‡‡ His most revealing target, at least to me, was FTX’s lawyer Dan Friedberg.
Outside the US courtroom, Friedberg had one of the better views of what had occurred inside Sam’s World and maybe the finest view of the role Sullivan & Cromwell had played in it. Inside the courtroom, Dan Friedberg’s experience was deemed irrelevant. And that, it seemed, was the end of it. US bankruptcy judges have sensational powers to determine which evidence to admit in a case.
At the end of June 2023, John Ray filed a report on his various collections. “To date, the Debtors have recovered approximately $7 billion in liquid assets,” he wrote, “and they anticipate additional recoveries.” Seven point three billion, to be exact. That haul didn’t include the Serum, or any large clawbacks, or the money stolen by the guy in Mauritius, or the stake in Anthropic, or most of the other private investments. An investor who was hoping to bid for the remaining portfolio told me that, if it was sold intelligently, it should go for at least $2 billion. That would raise the amount collected to $9.3 billion—even before anyone asked CZ for the $2.275 billion he’d taken out of FTX. Ray was inching toward an answer to the question I’d been asking from the day of the collapse: Where did all that money go? The answer was: nowhere. It was still there.
The closer a person was to him, and to the business, the more questions they had about it. Zane Tackett, for instance, could not understand why, toward the end of 2021, Sam hadn’t simply replaced the customers’ deposits inside of Alameda Research with loans from the crypto banks. Back then, Alameda could have borrowed $25–$30 billion without much trouble. Why not take that money and move the $8.8 billion of customer money back into FTX, so that if Alameda blew up it would take the crypto banks, rather than FTX, with it? Ramnik had a different question. He and Sam had invested billions of dollars of Alameda’s money—and yet he’d never seen Sam pay attention to the risks Alameda was running. Sam’s attention always seemed to be somewhere else. The question Ramnik wanted to ask Sam was, “Why the fuck did you spend the last year playing Storybook Brawl?”
I had my own questions, of course. The first one had to do with financial incentives. None of the characters in this financial drama had behaved as financial characters are expected to behave. Gary had owned a piece of Alameda Research, but his stake in FTX was far more valuable. Nishad owned a big chunk of FTX and none of Alameda Research. Ditto Caroline, who ran Alameda Research but owned shares only in FTX. None of these people had any interest in moving money out of FTX into Alameda Research in a way that put FTX in jeopardy. Just the reverse: it might as well have been their money that was being moved. And yet at least up until late spring of 2022, when crypto prices began to plunge, and possibly much later than that, none of them expressed disapproval of the risk being taken with their fortunes. Why not?
Want to get my latest book notes? Subscribe to my newsletter to get one email a week with new book notes, blog posts, and favorite articles.