Hyperliquid: In Singapore With Jeffrey Yan

On a Friday in January, before dawn, a 43-year-old was taken from his home in Saint-Léger-sous-Cholet, in western France. He was driven 30 miles to the small commune of Basse-Goulaine, where he was beaten, bound, and abandoned. Twelve hours later, once the sun had set in the suburbs of Paris, three men armed with a single handgun kicked in the door of a family home in Verneuil-sur-Seine. They beat a husband and his wife in front of their children, cable-tied all four of them, turned the house over, and left for the train station. It was the 70th such attack worldwide in under a year.

Two days later, I boarded a flight to Singapore.

I’d flown to visit a team of 11, but the first person I met at their office was not one of them. He was a solid American with close-cropped hair and stubble, sitting behind an Apple laptop at a small desk in the corner of the lounge, with a frame that suggested he was not there to write code. He was a bodyguard.

One of the firm’s co-founders, who goes by iliensinc, short for Aliens Incorporated, had walked me from my hotel to their office, and as we made our way through streets canopied by rain trees, she told me they had not always been in this part of Singapore. The business had started in a co-working space in the financial district, but her co-founder—the only person on the team who does not work under a pseudonym—had begun to attract attention. First it was just glances, people straining to place his face. Then strangers started approaching him. Then someone followed him into the elevator of his apartment. So the firm moved somewhere quieter, to a building where nobody would think to look for them.

Even their cleaner does not know what they do. In her mind, she dusts for a merchandising company that makes stuffed cats. There are 34 cuddly toys in the office, so the confusion is understandable. The firm’s mascot is a cat named Hypurr, and 12 of them are perched on a cabinet, but there are also sharks, lizards, koalas, penguins, and dragons, several of which are draped over Dell monitors like furry gargoyles. One engineer is responsible for most of the creatures. His wife won’t let him bring any more home, so he takes them to work. The team has not corrected the cleaner’s assumption. 

That is because Hyperliquid, a blockchain and cryptocurrency trading exchange, is one of the most profitable enterprises per employee on earth. Last year, its 11 employees generated over $900 million in profit. It is three years old, has a market capitalization of $10 billion, and has never taken a dollar of venture capital. The main figure behind it, Jeffrey Yan, is 31 years old and has become, not entirely by choice, one of the more recognizable faces in an industry where success increasingly gets you kidnapped.

Before Hyperliquid, Yan lived in Puerto Rico and ran, more or less alone, one of the largest anonymous trading operations in crypto. It was called Chameleon Trading—Chameleon had been his handle for video games in middle school. He started it with $10,000 of his own savings, and for two and a half years, it grew at thousands of percent a year. When he told me his returns, he immediately tried to talk me out of finding them impressive. I noted his objection. I also noted that Chameleon had made him very rich. He was 27 and he was free. To every surfer and bartender and waitress in San Juan, he was just another kid in board shorts.

Now he sat cross-legged on a gray armchair in a guarded office in Singapore, barefoot in black shorts and a dark blue T-shirt, explaining to me why all of finance needed to be rebuilt from scratch. What I wanted to know was why he had traded the first life for the second.

It was not money, he said. Yan did not grow up wealthy, and nothing about his life suggests he is interested in living as a wealthy adult. He wears the same Lululemon shorts and T-shirt every day. He owns 15 pairs of the shorts and 10 of the shirts, in three colors each. The office around him offered no evidence of wealth. The furniture belonged to the previous tenant. The team’s only additions were two board games in the lounge, NFTs on the walls, and the stuffed cats. I confirmed this when I found four books on a shelf and recognized one as Frank Slootman’s Amp It Up, a management book whose thesis is that most people do not work hard enough. I mentioned it to iliensinc. She shrugged. The mantra was theirs; the book was not. Nor were the three bottles of Grey Goose and Macallan I found in the kitchen, untouched since a community event two years ago that failed to meet the minimum spend. This team drinks tea.

Nor was it a love of crypto. Bitcoin, still the industry’s bellwether, had fallen roughly 30% from its peak in early October 2025. Gold, whose job Bitcoin was supposed to take, was up 7% over the same three months. Most tokens had fared worse. When I asked Yan about the negativity surrounding the industry, he did not defend it. “There is a lot of sketchy behavior in the space,” he said. “It may be healthy that people are realizing these things are not what they’re advertised to be.” He does not consider Hyperliquid a crypto company. “People don’t say we’re internet companies these days,” he told me. “We use crypto, but that doesn’t define us.”

Only two of the team, including Yan, worked in crypto previous to Hyperliquid. This is partly by design. The early crypto crowd, as he described them, were primarily interested in making money quickly. He is building for the long run, he said, and that aligns with people who think more like technologists than like traders. But it is also a supply problem. Hyperliquid recruits from international math and science olympiad podiums. Yan won gold in physics at 18. One of his engineers holds a silver in informatics. Another trained with the U.S. national team. Yan would like to hire more, and since my visit earlier this year he has added two, but the pool of people at this level who are willing to build in crypto has been thinned by years of scams and broken promises and, lately, by artificial intelligence.

What, then, was Yan, who had already made enough money to do anything, doing here?

The answer, at least to the outside world, is becoming clearer.

Hyperliquid is a blockchain with its own trading exchange built on top. On a traditional exchange, a company holds your money and controls the infrastructure. On Hyperliquid, you keep custody of your own money and the platform is public. Yan’s vision for it, stated without irony, is to house all of finance. That is a mark of ambition or absurdity, depending on whether you are looking at the cats or the platform’s numbers. Because in the months since my visit, markets that have traded the same way for over 100 years have started, in small and measurable ways, to bend.

Hyperliquid began in 2023 with perpetual futures, a type of derivative and the single largest market in crypto. A perp is a bet on the price of an asset you never own, and unlike a traditional future, it never expires. The market for these bets is six to eight times larger than the one for buying and selling the assets themselves, roughly $7 trillion a month, and until recently, virtually all of it ran through centralized exchanges. The biggest, by far, was Binance. No decentralized platform had made a dent. Hyperliquid was the first, growing to roughly 14% of Binance’s market share.

Then, in October 2025, Hyperliquid did something a centralized exchange cannot: It let anyone launch new perp markets on the platform, for any asset with a price feed. An independent group called Trade[XYZ] has been the most prolific. It started with a silver market. By January, its 24-hour volume had reached roughly 2% of the CME’s; the Chicago Mercantile Exchange is the largest derivatives exchange on Earth, founded in 1898. Then Trade[XYZ] listed crude oil. Oil has always traded on markets that shut for the weekend. On a Saturday in late February, the United States and Israel began bombing Iran. The CME was closed. Hyperliquid was not. Daily volume for crude went from $21 million to $3.7 billion. A month later, Trade[XYZ] launched a perp on the S&P 500, officially licensed by S&P Dow Jones Indices. It trades around the clock, weekends included.

The most consequential products on Hyperliquid are now being built by people who do not work for Yan and never will.

The founder of Trade[XYZ], who asked not to be named, had bought his first Bitcoin in 2013 for $66 and spent the years since as an investor, not a builder. He had not planned to start a company. He told me that if it were not for Yan, he would no longer be in crypto. “Hyperliquid has the chance to save crypto,” he said.

Still, none of it explains why Hyperliquid might become what Yan says it will, in an industry where things have a habit of looking as if they will right up until they don’t, or why he had given up the life he had in Puerto Rico to find out. These questions sat with me on my first afternoon at the office as iliensinc and I talked in the lounge, a stuffed cat on the table between us, the smell of ginger and sesame from lunch still in the air. She told me the team had asked Yan the same thing three years ago, when he announced that Chameleon was over. Her answer began not with crypto, but with the kind of person Yan is. I should ask him about his mother, she said.

Yan prefers to take his meetings outside. We sat on the covered terrace, furnished with four gray lounge chairs and a coffee table. Cars passed on the street below. Every few minutes, a gardener fired up a strimmer. The steady beeping of a pedestrian crossing came and went.

Yan sat with his feet tucked under him, and when I asked about his mother, he thought for a moment. She had a saying, he told me. A Chinese idiom. Rén wài yǒu rén tiān wài yǒu tiān, which roughly translates to “Beyond the person, there are greater people; beyond the sky, there is more sky.” She was not the kind of mother who pushed, but she wanted him to know that however good he thought he was, he was seeing only a small piece of what was out there.

She raised him and his younger sister alone in the center of the most lucrative stretch of geography in the history of American business: Redwood Shores, between San Francisco and Palo Alto. Oracle’s mirrored headquarters towered over the neighborhood. The neighbors were engineers and product managers, and their children were being groomed, already, for exactly the kind of life Yan would later build. His parents, both Chinese immigrants, had divorced when he was in third grade. His father left. His mother, an accountant, worked overtime through every tax season and between them, and he could see it. “I could tell that people were more well-off than us,” he said, “but it’s never something I resented. It’s not very expensive to go play outside.”

His school had no culture of academic competition. Despite her sayings, his mother did not push Yan. Until he became a teenager, nobody pushed anything. He played outside, went to school, came home, played some more. He was, by the standards of his zip code, the rarest thing: a child left alone.

In eighth grade, a friend who had just transferred from a private school took him along to a math competition. The friend wanted company. Yan had never seen anything like it. School math was nothing like this. There were no formulas to memorize, no calculations to grind through. You were given a question, sometimes only a sentence long, and left to find your way into it. The answer was not a number. It was a proof, a complete argument showing why something must be true. At the end, they ranked you, the way they rank sprinters. To Yan, it was the best part of sport fused with the best part of understanding the world.

That summer he woke at five every morning, downloaded old competition papers from the internet, and worked through them alone in his room. He had no tutor. He could not afford a summer program. Nobody was telling him to do it. “It turned out, I was super competitive,” he said. “There was this race I wasn’t aware of, that other kids had been doing their whole lives, and I was behind.”

A year after he started, by which time he was in ninth grade, he had qualified for the U.S. Math Olympiad training camp, which comprised the top 50 high school students in the country. He was among the youngest in the room. He did not make the national team. He did not care, he said. For three weeks he sat among teenagers who could stare at three sentences for five hours and find, buried somewhere inside them, truths invisible to most other minds. There is no Roger Federer of mathematics, Yan told me, but at the very highest level, there is something like what Federer has. There is a style to the work, an elegance to the way a proof is constructed, and at the training camp he saw it up close for the first time. “It’s like being able to play football with Tom Brady,” he said. “But a nerd version of that. Most people don’t get that feeling.”

The next year, he failed to advance past an intermediate qualifying round in mathematics. He was 16, and he would have to wait another full year before he could try again. I asked whether it was the first time he had experienced failure. “Losing is a pretty common experience,” he said. “Most people are losers. There’s usually only one winner.”

The problem was not the losing; it was the emptiness. “It felt like there was this void,” he said. “I should be studying something.” So he found some physics textbooks that the upperclassmen used. His school did not teach the subject until junior year, but he had just learned calculus, and for the first time he understood what it was for. He discovered Feynman’s lectures. “I consumed them like a TV show,” he said. Within a year, again self-taught, he became one of the top five young physicists in the country.

He made the U.S. Physics Olympiad team and traveled to Estonia—his first time in Europe—and won a silver medal. The following summer, in Copenhagen, he won gold, ranking 24th in the world. He was 18, and he came home to the Bay Area understanding that his mother had been right about the sky. There were exactly 23 people beyond him.

Harvard University covered nearly all of his tuition. During the spring semester of his freshman year, Yan enrolled in computer science 124, data structures and algorithms. It is a course taken mainly by sophomores and juniors, and it has a reputation for misery. Students in Harvard’s course guide have described it as “a necessary evil.” One review warned, “No social life. You will be maiden-less.” There were 150 students. Yan, the freshman, finished first, and it was not close.

At Harvard, students are assigned to upperclassman houses after freshman year. Yan drew Pforzheimer, where he became close with Scott Wu, two years younger, whom he had first encountered at a summer program for olympiad kids. Wu had won three consecutive gold medals for the United States at the International Olympiad in Informatics, the last with a perfect score, and would later co-found Cognition AI. When Wu was assigned to Pforzheimer as a sophomore, he texted Yan, “Yo, I’m in Pfoho.” Yan wrote back, “Let’s go!” 

Wu would find Yan at the grand piano in the common room, teaching himself jazz, playing and replaying phrases until they locked in. They played chess and Go and poker together, and they spent hours talking about what it meant to be the best at something. Yan would talk about Faker, the greatest League of Legends player of all time, and about the great Go players and the best HFT traders. “He would always think about what makes a person special,” Wu told me. “What is the essence of this field? And what does it mean to get really good at it?”

Wu remembered Yan as unusually contrarian. Most students at Harvard, absorbing the same information from the same environment, arrived at roughly the same conclusions. Yan never did. He was also, Wu said, very funny. “Very deadpan. He’d just say something totally different from what you expect, but deliver it in the most dry manner possible.”

In the summers, Yan worked. He interned at Google X, building tools for the self-driving car project before it became Waymo. He interned at Tower Research Capital, a trading firm. During his senior year, he worked part-time at Nuro, another self-driving car company, largely because he thought four years of college was at least one year too many. 

In the winter of his junior year, he and Wu were among 10 interns in the first program that Hudson River Trading had ever run. HRT is one of the most successful quantitative trading firms in the world. Also among the 10 were Alexandr Wang and Jesse Zhang, who would go on to co-found Scale AI and Decagon, respectively. The internship was structured as a three-week competition. In every round, Wu and Yan finished first and second.

Yan graduated with an undergraduate degree in mathematics and a master’s in computer science and joined HRT full-time in late 2017. He was assigned to the algorithms team in U.S. equities. Once a week, he sat down with his manager. The manager had supervised a number of new hires. Usually these meetings had a rhythm. The hire would hit a wall in the code, and they would work through it together, and the hire would go back and hit the next wall. Yan did not hit walls, his manager recalled. He came in with ideas. The meetings were efficient, and something about them nagged at his manager. It took him a while to place it. Yan was doing everything right, but none of it seemed to matter to him. When Yan came in after eight months to say he was leaving, his manager understood. His email announcing Yan’s departure was, by the firm’s standards, notably warm.

Yan liked HRT. He thought trading was the purest real-life game you could play. You were right or you were wrong and the market told you which. A lot of the smartest people in the world were competing against you, and in the process of playing this brutal game against each other, you produced an extremely valuable product for the world: liquid, efficient markets. But he had spent eight months improving a system that was already very good, inside a firm that would be very good without him, which meant he did not have a good answer to the question he couldn’t stop thinking about: What value are you adding to the world?

In December 2017, the answer found him. Bitcoin was near $20,000. Coinbase was the most downloaded app in the country. Billions of dollars were pouring into initial coin offerings like Jesus Coin. It was crypto Christmas. Yan had first heard about Bitcoin during the HRT internship, when two former partners had come to pitch the interns on it. It hadn’t clicked for anyone. But, while still at HRT, he found Ethereum’s yellow paper, and it described a computer that ran computations the whole world agreed on and no one person could shut down. He was touching finance every day. He could see what it ran on. The paper described a way to replace trust with code. “I felt like I could go and build a thing that would revolutionize finance.”

He left HRT around April 2018 to build a prediction market where users could bet on the weather or elections or sport. Anything with an outcome. It would run on the blockchain, where no single entity controlled the money. The architecture rested on an idea Yan believes he and his co-founder were the first to arrive at: off-chain matching, on-chain settlement, because Ethereum was far too slow to run a real exchange on. The money would sit in a smart contract, governed by code, but the user would see something fast and clean. The decentralized promise of crypto with none of the friction. He built it with his college roommate Brian Wong, who had also left HRT, out of Binance Labs’ first startup incubation in San Francisco. They called it Deaux. 

Kalshi was founded in 2019 with the same thesis. Polymarket followed in 2020. Between them, Kalshi and Polymarket are today worth north of $40 billion.

Deaux got 100 users.

As Yan recounted this, the Singapore sky opened. Fat, heavy drops, the kind of rain that fills gutters in minutes. From the terrace we could hear it hammering the street below, the cars growing louder as tires hissed through the water. 

“There was no shot it was going to work,” he continued. By the time Deaux launched, Bitcoin had fallen more than 80%. Jesus Coin had died and not been resurrected. Nobody wanted to bet on what the weather would be tomorrow. More than that, though, Yan and Wong had scarcely thought about regulation. Kalshi would spend three years fighting for regulatory approval before it launched a product. 

When Deaux shut down, Scott Wu was one of the only people on Earth sorry to see it go. He had been one of five regular users.

Yan returned more than half the $450,000 investment. He was still under a non-compete from HRT, so he went to Tahoe, California, with a friend who had an overlapping one, and they snowboarded until the snow melted. Then he traveled to China, Japan, and Peru on a low budget. There was, he tried to convince me, a surprising amount of skill to being a tourist. He did not have it.

In late 2019, when his non-compete expired, Yan moved to Puerto Rico, where people could legally bring their capital-gains tax rate to nearly zero. He had $10,000 and a sense that something big was coming.

His partner came with him to Puerto Rico. They shared a one-bedroom apartment near the beach that cost less than $2,000 a month, but sharing implies a degree of togetherness that Yan did not make time for. Nor did he have a monitor, so he commandeered the television and set himself up in the living room. For the first year or so, she was allotted roughly 30 minutes of his attention per day. The rest belonged to the trading algorithms scrolling across the television.

Yan worked 14 hours a day, at a minimum, easily clocking 100 hours a week. He started with Python scripts, writing code that connected to crypto exchanges and traded on his behalf around the clock. He monitored them, refining the logic, following the data, and tearing the system down when it wasn’t working as he wished.

He could do this because crypto was open in a way that traditional finance had never been. In equities, like those he’d traded at HRT, placing a single order on a single exchange required connections to 13 lit exchanges across three co-location sites in New Jersey, compliance with a thicket of SEC regulations known as Reg NMS, microwave links to Chicago for CME futures data, and tens of millions of dollars in setup costs. In crypto, everyone—whether an HRT employee or a man working off a television—connected to the same janky HTTP infrastructure meant for building web pages. All you needed was a server on Amazon Web Services.

For almost two years, his partner had no idea what was happening on the other side of the television. Their lives had not changed. They paid the same rent. They ate the same food. She knew he was passionate and driven, and assumed he was doing reasonably well, but there was no material evidence of his success. Then, one Friday evening in the summer of 2021, she tried to get him out the door for a dinner reservation she had made a week in advance. He would not leave.

“You don’t understand,” he told her. “If I don’t fix this bug right now, I’ll lose $100,000.”

After that evening, Yan decided to turn it into a real business. He needed someone who could do everything but code. At Harvard, there had been a person in Pforzheimer who seemed on top of everything in her life, all at once, which was as foreign to him as a skill could be. But last he’d heard, iliensinc was in Asia, working as chief of staff at a venture capital firm, traveling between Tokyo, Seoul, and Hong Kong.

When he reached out, he found her in San Francisco. COVID had grounded travel, and the job that had taken her across Asia had become a series of midnight phone calls from her apartment. Yan explained what he needed. He offered no job description, no title, and almost no detail about what she would be doing. But she had spent three years evaluating founders for a living, and whatever Yan was describing, she thought he was not the kind of person to bet against.

The firm officially got a name, Chameleon Trading, and iliensinc began joining Zoom calls with business development teams at the exchanges, lending a veneer of professionalism to what was, in physical reality, one man above the beach in San Juan. Beneath the behemoth market makers—firms like Jump Trading, Tower, HRT, and Jane Street—there existed a tier of anonymous firms whose scale nobody could quite verify. Chameleon was one of the most substantial.

By 2022, Yan was getting restless. He had been in crypto for four years, plugged into various markets, centralized and decentralized, and had grown to care about the space beyond his own profit and loss. Bitcoin had given the world a way to hold and move money without trusting an intermediary. Ethereum had given it a computer that no one person could shut down. Between them, they had laid out almost everything you would need to rebuild the financial system. But the industry had done almost nothing with either. The two largest exchanges, Binance and Coinbase, were centralized. Crypto kept reintroducing the thing it was supposed to eliminate.

That summer, iliensinc arranged a team off-site at a hotel in the English countryside. She had grown Chameleon into a team of six by then. Yan gave her a budget of one bitcoin. The team flew to London, visited the British Museum, and spent a few days at the country estate. Their leader, away from his screen for the first time anyone could remember, was not entirely at ease.

When they returned to Puerto Rico, the trading continued. But Yan told his team they were going to start building something new. He wasn’t sure what. He had ideas, none of which he found convincing. He just knew that Satoshi Nakamoto’s original vision for Bitcoin was being quietly buried by the industry Satoshi had created, and it bothered him more than it should have bothered someone making millions off everything it had failed to build.

Yan, it seemed to his team, had gotten a little too much fresh air.

In November 2022, FTX, the third-largest cryptocurrency exchange in the world, valued at $32 billion, collapsed in nine days. It had been lending its customers’ deposits to Alameda Research, a trading firm run by the founder’s girlfriend. When users asked for their savings back, the cash was not there. Less than six months earlier, Terra, a cryptocurrency ecosystem worth $50 billion, had gone to zero in three days. It had tried to build a dollar-pegged currency backed by nothing but the system’s own logic. The algorithm that was supposed to hold the peg accelerated its collapse. Two of the largest projects the industry had ever produced, dead within half a swing around the sun.

Yan had seen enough. He told his team of six they were done trading. They may disagree, he said, but Chameleon was over. If he was wrong, they could always go back to trading. Several of them did disagree, and several would leave. But that didn’t change Yan’s mind. There were no investors to consult, no board to convince; it was his money and his call, and there was a new mission.

“I was overly confident that FTX would be the downfall of centralized exchanges,” Yan told me. “But it was helpful because it gave me the conviction to go after this massive market.”

The market he meant was perpetual futures. They were born out of an insight Robert Shiller, the economist, had in the ’90s. A traditional futures contract has an expiration date. When it arrives, a trader either takes delivery of the underlying asset—oil, wheat, pork bellies—or closes their position and opens a new one, paying fees each time. Shiller asked the obvious question: If almost nobody who trades a pork belly future wants pork bellies, why force the contract to expire? 

Traditional markets, which already had workable solutions, saw no reason to change. In 2016, a crypto exchange called BitMEX did, and since then perps have become the dominant way crypto trades. The contracts never expire. Traders can take heavily leveraged positions, often 10 or 20 times their capital. The fees and liquidations they generate have turned centralized crypto exchanges into some of the most profitable companies in the industry. 

By late 2022, no one had built a decentralized version worth using. The reason was the underlying technology. In most modern markets, trading runs through an order book. Buyers say what they will pay. Sellers say what they will accept. When the two line up, a trade happens. The more people in the market, the smaller the gap between those prices. That is broadly how markets from the New York Stock Exchange to Binance work. But an order book does not just process trades. It also has to keep up with a constant flood of updates as traders move their prices again and again, often many times before a deal is struck. Existing blockchains were bad at this. They were too slow, too expensive, and too awkward. Every update cost money and took time to clear. Running an order book on them was like trying to run the New York Stock Exchange over dial-up.

Yan and his team looked at every blockchain that other projects were building on, and none came close to what they needed. So they built their own. In three months, Hyperliquid had enough of a custom blockchain to run an exchange on top of it. Yan then spent much of that year on Twitter, making the case for what Hyperliquid offered and why it was better than what the industry had settled for. 

The trouble with an exchange is that it’s useless until it isn’t. A buyer who shows up to an empty market has no one to buy from. The conventional answer is to pay market makers so that anyone who arrives has someone to trade against. You pay them in cash or equity or a cut of the token. Several approached Hyperliquid. One told iliensinc, flatly, that his firm was a kingmaker. “You’re never gonna take off if you don’t pay us.”

They did not pay him. They did not pay anyone. Hyperliquid launched at the end of February 2023, and through March and April the user base consisted largely of NFT collectors who had never traded a perp in their lives, placing $10 trades and learning leverage through paper trading competitions. There were no serious users.

He just knew that Satoshi Nakamoto’s original vision for Bitcoin was being quietly buried by the industry Satoshi had created, and it bothered him more than it should have bothered someone making millions off everything it had failed to build.

Then, in May, Yan took the strategies that had made Chameleon one of the most successful anonymous trading operations in crypto and put them into an on-chain vault called HLP, the Hyperliquidity Provider. You could deposit $10 or $10 million. There were no fees and no carry. The vault ran automated strategies, and every dollar of profit flowed to whoever had put money in. The accounting was entirely on the blockchain. If you deposited $10, you could watch, in real time, as your $10 grew. If FTX had been built this way, Alameda’s hole would have been visible to the world.

HLP solved two problems at once. The exchange got liquidity. And users who provided it got access to something traditional finance had never offered. One early Hyperliquid user described it as the first time in history that an ordinary person could invest in a high-frequency trading strategy with no fees.

“I would have paid Jeff a 2% management fee and 50% carry to be in this thing,” they told me. “Instead, a nobody with no network, sitting anywhere in the world, was able to access one of the great market-making strategies in crypto. People still don’t understand how special this was.”

Few understood it at the time. By fall, crypto prices were rising every day, and depositors watched HLP’s balance drop while Bitcoin climbed. The algorithm was doing its job and making money on its trades, but because everything ran on-chain, it couldn’t hedge its exposure to the broader market. A traditional market maker would have offset that risk on a different venue. HLP, by design, could not. So while it won trade after trade, it was effectively short a market that kept going up. People were furious. Other projects attacked Hyperliquid on Twitter and in Discord, and Yan attacked back. It was early enough that he still took things personally.

But HLP was never meant to be the answer. Yan had built it to bootstrap liquidity until independent market makers arrived, and he could see that the opportunity for them was obvious. Demand was outstripping supply, and wide spreads meant easy money for anyone willing to quote. He wrote documentation. He wrote long posts on Twitter explaining how market making worked. He walked firms through onboarding. Most were reluctant. Every other exchange paid them. Yan refused, and HLP could not scale to fill the gap either. “Alameda was essential to how FTX worked,” he said. “We did not want HLP to be essential to how Hyperliquid works.”

The metrics were climbing, but so were the complaints. Market makers, in theory, should have been arriving any moment. But if they didn’t, and the users left first, it was over.

You can always count on one group to show up, though. The Venture Capitalists.

Their analysts had been using the exchange themselves, quietly, on their own time, and one by one they had gone to their partners and said: This is actually good. The partners picked up the phone. Yan and iliensinc had done no outreach. They didn’t have a pitch deck. The protocol was generating fees, but Yan had insisted from the start that none of it would flow to the team. When VCs got on a call and asked if there was a deck, Yan and iliensinc would just talk, and eventually they would understand that no, there was not.

By January 2024 the funds were visiting in person. iliensinc knew this process. She had worked as an investor. She started walking Yan through the terms he should know, the rights to watch out for. For about two weeks, he went along with it. “It almost felt like second nature,” he told me. “Oh, the VCs are reaching out. I guess it’s time to raise a round.”

His one condition was that he would consider only a term sheet at a billion-dollar valuation. It had been less than a year since Hyperliquid launched. The team was burning hundreds of thousands of dollars a month out of Yan’s personal savings. When an investor met his number, Yan took a weekend and thought about it. 

He asked people who had run startups and the VCs themselves to explain to him what exactly the point of raising was. But they could not convince him their money was worth more than their money. At some point, he told me, it felt right to say no. And once it felt right, that was the end of it.

On Monday morning, he told iliensinc, “We’re not going to take it.”

“What the fuck?” 

She couldn’t believe it. She was the one managing the money, watching it burn. Now a fund was offering around $100 million, and he was turning it down after she had spent two weeks preparing for exactly the opposite. The rest of the team took it no better. 

He called the fund and declined. They did not believe him either. Surely he was accepting someone else’s term sheet. He was not. Hyperliquid was not a company. It was a protocol, and its neutrality from day one was the point. “If Bitcoin had raised VC rounds,” he said, “I really don’t think it would be Bitcoin. Its entire value proposition would have been destroyed.” Plus, he didn’t need the money. To this day, Yan still pays many of the team’s costs himself.

On January 28, 2024, he tweeted four lines:

No investors. 

No paid market makers. 

No fees to the dev team. 

No insiders.

Hyperliquid has one meeting a day, a morning stand-up, and I watched it on my second day in Singapore. The team crowded around one of the engineer’s screens. A dragon plushie sat on top. They were testing a new feature called portfolio margin, and the conversation was mostly about what could go wrong. It was also, for long stretches, not a conversation at all. Yan would cross his arms, bow his head, and contemplate his bare feet. The engineer beside him did the same. These silences were not awkward and they were not brief, and nobody in the room seemed to find them unusual.

Part of the reason for this dynamic is temperament. The team is young, between 24 and 31 years old, and nearly all of them are supremely intelligent introverts. But Yan, when I asked him afterward whether he reads much, suggested there was more to it than shyness.

“I’ve read way fewer books than conventional wisdom would say is optimal,” he told me, smiling behind dark-framed glasses. “It’s very time-consuming to read a book in a way that permanently shapes who you are. The return on time is not super good.”

He stretched his jaw for a moment—a habit I would come to recognize—like someone popping their ears on a plane. It is a particular hazard of writing about young technologists that they will, sooner or later, tell you they don’t read. I was grateful, then, when Yan clarified that he reads a book every other month and that he looked forward one day to sitting down and reading all the books he hadn’t read. Then he continued to explain why reading more would have to wait.

“If you’re not the first person doing something,” he said, “it’s probably not worth your time to be doing it at all. I genuinely believe that. And if you operate under those assumptions, then reading isn’t very helpful. If there’s already helpful context on what you’re doing, then it’s probably been done. And if it’s been done, why are you doing it?”

In late 2023, Hyperliquid faced another problem for which crypto had a well-worn playbook. Yan, as with the others, had no interest in following it. A crypto project’s token gives its holders a stake in the project’s success. Deciding who gets tokens first, and on what terms, is typically done through a points program. A project announces that using its platform will earn the user points. Users assume the points will later convert into tokens. Then they flood in, hoping to accumulate as large a stake as they can before the conversion. 

The problem is that the users who flood in are not, for the most part, users at all. They are professional operations that reverse-engineer the formula, run automated strategies to harvest the maximum reward, and leave. The actual users, the ones the program was designed for, get what’s left.

Hyperliquid’s version launched on November 1, 2023. Users who traded on the platform accumulated points each week, but the program had no published formula. Nobody knew how it worked. Every Friday, iliensinc announced the week’s points, and a ritual formed around it. Users watched for her handle to appear as typing in Discord, then gathered to compare what they had received, share screenshots, and build theories about how the system worked. “It’s critical to reward real users,” Yan said. “It’s very hard to define, but Hyperliquid’s points program probably took the percentage of farmers from 99% to 20%.”

Around this time, the market makers, the ones Yan had refused to pay directly, began to show up. One of them, one of the biggest market makers on Binance, had been wary of new venues after FTX. But he had mutual contacts who spoke highly of Yan, and in September 2023, at a conference in Singapore, he met Yan and iliensinc for the first time. “Jeff was ambitious but he wasn’t arrogant,” the market maker told me. “He was very measured in how he described what he was trying to do, and he just checked all the boxes.” He walked out and texted his team: We should integrate. Two weeks later, they were live.

What the market maker found when he plugged in confirmed what the users were discovering on their own. The infrastructure was thoughtful in ways that only a trader would notice. Hyperliquid had built in a kind of speed bump that made it harder for the most aggressive quantitative firms to pick off other market makers. The feature has since been copied across the industry. The effect was that market makers could display deeper liquidity without needing to be at the bleeding edge of latency to survive. Yan had effectively chosen to sacrifice some exchange volume—the kind generated by firms sniping each other—in favor of better prices for ordinary users. It was a trade-off that reduced Hyperliquid’s own revenue. 

It was at that same conference, Token2049, that Yan and iliensinc decided to move the team. The regulatory picture for crypto derivatives in the United States was uncertain, Yan told me, and building there felt like a risk they didn’t need to take. One lawyer I spoke with described it as a period in which American regulators used “every means available to ban the technology from the country.” iliensinc looked at Hong Kong, Switzerland, and Singapore and settled on Singapore. It was modern and safe, and there were no distractions.

By the spring of 2024, the team had moved. It suited Yan because the city-state was boring. He has two modes: working and working out. He swims, he runs, he’ll do anything that will exhaust him without risking injury, a principle he arrived at after a moped accident in Puerto Rico left a scar on his face and cost him a week away from his keyboard. Exercise exists to clear his head so he can go back to building. His one concession to leisure is Sunday mornings. The rest of the week belongs to Hyperliquid. He even cuts his own hair because, well, going to a barber takes time.

He does not consider this unusual, or rather, he considers most people’s relationship with work unusually slack. “I think people are just a bit too soft in general,” he said. “The brain is an organ. If you need to work more hours, you can train for it.”

He has learned not to impose this on the team. They eat lunch together every day, family-style, around a black wooden table. On Thursdays they eat Chipotle. There is no Chipotle in Singapore, so they gave the recipes to their chef, who now makes it for them. The conversation over lunch typically drifts toward whatever the team has been watching or listening to. Yan, when this happens, tends to go quiet and look like he is thinking about something else, which he probably is.

By that same spring, Hyperliquid was handling over $1 billion in perps volume a day, and the infrastructure was creaking under the weight. One afternoon, the alerting systems went off and kept going. The platform could not handle the number of users who had arrived. It was Hyperliquid’s first downtime. But all anyone outside the office cared about was the impending Hyperliquid token.

In May, Yan tweeted a roadmap for the next six months. It was full of technical ambitions. It did not mention a token.

In the months before, Hyperliquid had expanded from derivatives into spot trading. The first token it listed was Purr, named after the cat. Spot trading was a necessary step. To launch Hyperliquid’s token, the team needed a spot market to trade it on. But it introduced a problem the derivatives exchange had never faced. When you trade a perp, no one needs to hold the underlying asset. You’re betting on a price. When you trade spot, someone has to take custody. That was something Yan did not want to do. The entire point was that users controlled their own assets.

To solve this without becoming a custodian, he realized he had to stop thinking of Hyperliquid as an exchange that sat on a blockchain and start thinking of it as a blockchain with an exchange built into it. The blockchain the team had built to run the exchange, already processing hundreds of thousands of orders per second, could be made programmable. It would be an open system on which anyone could write code and build financial applications, the way thousands of developers already did on Ethereum. The difference was that Ethereum was too slow to run a proper exchange, which is why Yan had built his own chain in the first place. 

If he opened up that chain, assets could be brought onto Hyperliquid through decentralized bridges secured by the protocol itself, without any single party taking custody. And anyone who built on the programmable layer would have access to the exchange’s order book and all the liquidity sitting in it. A developer could build a lending platform or a stablecoin or a mobile trading app and plug it directly into the same markets where professional firms were quoting billions of dollars a day.

Yan dislikes analogies. He will tell you that Hyperliquid has no parallel in traditional finance, that people prefer to fit new things into old categories rather than understand them on their own terms, and that this is a mistake. But for those of us who are not Yan, it was like Amazon building cloud services to power its marketplace and then realizing the cloud services were bigger than the marketplace. The phrase Yan used, for the first time in that Twitter post, was that Hyperliquid would house all of finance.

He had been reluctant to make this change. He told me he had subconsciously not wanted to sign up for it. Building a virtual machine into Hyperliquid was an enormous project, and the team didn’t know if it was possible. They didn’t know how much of the work they would have to do from scratch. But at some point, he said, it became obvious. If they didn’t do this, they would spend years stitching together components that were sort of like Binance and sort of like Ethereum, but neither one properly, and they were going to regret it.

The community was furious. They had expected an airdrop. Instead they got a tweet about infrastructure. Comments with a thousand likes quoted a meme from Breaking Bad: “We had a good thing.” “I hate this. You betrayed us.” Users didn’t want a blockchain. They wanted their money. Xulian, who had joined the team after a 15-minute user interview that lasted an hour and a half and never quite ended, absorbed the anger. “Jeff thinks about what is best for the long term,” he told me. “We just really don’t care if something looks good right away.”

The loud people, as iliensinc described them, eventually got tired of complaining. The team spent the next six months working through spot, building out the programmable layer, testing it on a separate network, preparing for staking. Then, on November 29, a Friday, HYPE arrived.

Hyperliquid airdropped 31% of its total token supply to roughly 94,000 early users. There were no conditions or vesting schedules. If you had used the platform and earned points, you woke up that morning with tokens in your wallet, richer than you had been when you had gone to sleep. At the opening price, the airdrop was worth over $1 billion. At all-time highs, it would reach $16 billion. It was the largest wealth transfer in the history of cryptocurrency, and every dollar of it went to users.

The team’s own allocation, 23.8%, was smaller than the community’s share and vests over years. On the day itself, they received nothing. VCs received nothing either. If they wanted the token, they had to buy it on the open market, at the same price as everyone else, on Hyperliquid, because it was not listed anywhere else. That was another thing you had to pay for.

Yan didn’t have to explain anything on Twitter that morning. “Woke up to a mid 6 figure airdrop,” one user wrote. Another replied, “Today, HYPE changed my life. Enough money to sort myself comfortably for years, help my family and invest heavily into the bull run.” Someone else: “7 figure airdrop godbless jeff.”

“I felt very good,” Yan told me. “It’s rare that people who are early to something can all participate in the upside and gain meaningful ownership over a network.”

I asked him how it had been since, having such a public price on everything they were building.

“It sucks,” he said.

It was a Wednesday night in late March 2025 when iliensinc’s computer began to beep. She was on a call. She ended it. On her screen, the balance of HLP, Hyperliquid’s community vault, was falling.

A trader had spent the preceding days probing Hyperliquid’s defences with small, coordinated positions. Now the probe was over. They opened three positions on Jelly Jelly, an obscure token worth roughly $15 million, with daily volume of $72,000. One was a large short. Two were longs. The short was designed to fail. The trader was betting against a token they were about to pump, and when the position collapsed, someone else would be left holding it. It was like pulling the pin on a grenade and handing it to someone else.

The someone else was HLP. On Hyperliquid, when the order book cannot absorb a trader’s liquidation, the community vault takes over the position and unwinds it over time. Under normal conditions, this is routine. But Jelly Jelly had almost no order book, and while HLP sat trapped, unable to exit, the trader was buying Jelly Jelly on the open market as fast as they could. The price rose more than 500% in under an hour. With every tick, the vault’s losses grew.

iliensinc stared at her screen as the losses passed $5 million, then $8 million, then $12 million. There was nothing in the system to make it stop. Nobody had designed for a world in which someone would use a $15 million token as a weapon.

Across Asia and Europe, the validators came online. Hyperliquid’s blockchain was secured by roughly two dozen of them, independent operators who verified every transaction and earned voting rights by staking large quantities of HYPE as collateral. Many had been using Hyperliquid since before the token existed. They could see what was happening on the same public ledger that anyone, anywhere in the world, could see, and they did not regard it as a trade. Within minutes, all of them voted to delist Jelly Jelly and settle the position at the price that had existed before the manipulation began. Every user with a legitimate position was made whole. The only person who lost money was the attacker.

He realized he had to stop thinking of Hyperliquid as an exchange that sat on a blockchain and start thinking of it as a blockchain with an exchange built into it.

The episode surfaced a question that Hyperliquid’s critics had been waiting to ask. If two dozen validators could override a market price and settle a contract at a number they chose, how decentralized was the system? Yan did not dodge it. The validator set was small by design. A system that ships upgrades every few weeks cannot coordinate a thousand participants for each one. The set would grow over time, but not at the cost of the speed that had gotten Hyperliquid here.

“The fix took a month. It sucks to have to learn it from an attack rather than someone just telling you,” Yan said. Hyperliquid, which has never paid a market maker and never taken a fee for the team, will pay up to a million dollars for a bug report. “But these people obviously weren’t trying to inform us of an issue. They were trying to exploit it.”

As the attack happened, Binance and OKX, two of the largest centralized exchanges in the world, listed Jelly Jelly perpetual futures on their own platforms. On Twitter, a user had tagged Yi He, one of Binance’s co-CEOs, urging her to list the token. “If you list Jelly Jelly,” they wrote, “Hyperliquid will probably be done for.” Yi He replied, in Chinese, “Okay, got it.”

This, then, was the reward for ambition. You leave a beach in Puerto Rico where nobody knows your name. You build something from scratch using your television and your own savings. You turn down $100 million. You give billions to strangers. And what do you get?

War.

In 2023 and 2024, Hyperliquid had been small enough to be left alone. The airdrop changed that. A market capitalization of $4.2 billion, then $9 billion, then more, meant that every large business in crypto could now see the outline of a future in which Hyperliquid took their lunch. Binance announced its own decentralized exchange. Coinbase and Robinhood began offering futures products. New protocols launched with Hyperliquid as their target. And then someone followed Yan into his elevator at home.

It was the kind of thing that might have been nothing, but in 2025, violent attacks on crypto holders had nearly doubled. In France, a co-founder of a hardware wallet company had a finger sawed off and a photograph of it sent to his business partner as ransom. A family in Canada was waterboarded. Cryptocurrency transfers are instant, irreversible, and require no bank to approve them. A man with a wrench and a wallet address can drain a fortune.

Yan moved to a safer location, hired a bodyguard, and became somewhat confined to the safest island city on Earth. When he travels, two personal security guards accompany him. iliensinc began quizzing the team on what to say if a stranger asked where they work. It is why almost every person who spoke to me for this profile appears under a pseudonym.

When I asked Yan for the hardest moment of 2025, he did not mention Jelly Jelly or the competitors or the bodyguard. He told me about API servers.

Through the summer, as Bitcoin climbed past $100,000 and Hyperliquid processed more than $400 billion in volume a month, the servers that connected market makers to the blockchain began to fail. Too many firms had onboarded, each one sending a torrent of orders, cancellations, and updates, and the infrastructure that relayed all of it to the chain could not keep pace. An order that should have cleared instantly was taking three seconds. 

The chain itself stayed up. User funds were never at risk. But three seconds, in a market where fortunes turn on milliseconds, was a warning. “If we’re hitting congestion when it’s not extraordinarily volatile,” iliensinc said, “it’s unacceptable for when that event happens.” Yan stopped sleeping for weeks. He would go to bed at 1:30am and get woken at 3am by someone pinging him that things were breaking again. The team rewrote the servers from the ground up.

On October 10, the event happened. President Trump threatened 100% tariffs on Chinese imports, and over $19 billion in leveraged crypto positions were liquidated in 24 hours, the largest wipeout the industry had ever seen. More than 1.6 million traders were caught in a cascade that fed on itself, with forced selling driving prices lower, triggering more liquidations, driving prices lower again. 

Hyperliquid had no downtime and halted no withdrawals. The rebuilt servers held. The Jelly Jelly fixes held. HLP backstop-liquidated billions and earned $40 million doing it. But because every transaction on Hyperliquid’s blockchain is public, anyone could count its liquidations. The other exchanges didn’t report theirs with anything like the same accuracy. Binance published only one per second. The data aggregators that major outlets relied on used what they were given, and what they were given was misleading. The media reported that Hyperliquid had processed more liquidations than any other exchange. It looked like the most dangerous place to trade for the simple reason that it was the most transparent. 

Three days later, while the rest of crypto was counting its dead, Yan’s team shipped an upgrade that would define what Hyperliquid was becoming: Hyperliquid Improvement Proposal 3. HIP-3 let anyone who staked 500,000 HYPE tokens deploy new perpetual futures markets on the platform, set their own parameters, choose their own price feeds, and keep half the trading fees.

By the end of the year, its second full year in operation, Hyperliquid had earned about $900 million in profit. Not a dollar of it went to the team. Ninety-nine percent was automatically converted into HYPE and burned, removed from circulation forever, returning nearly all of the platform’s earnings to anyone who held the token.

When I asked iliensinc how she reflected on 2025, she said, “It felt like we grew up.”

On my last afternoon at the office, I sat with Yan at the black dining table by the kitchen, within earshot of the untouched whiskey, where the team eats lunch together every day. I had questions I’d been saving.

Hyperliquid had spent the previous year handing away pieces of itself. Builder codes, launched before HIP-3, let independent developers build trading apps on the platform’s order book and keep a share of every fee their users generated. Matt Huang, the co-founder of Paradigm, one of crypto’s largest investment firms, called it “a brilliant way of franchising out the user experience.” Those teams had earned more than $70 million since October 2024. 

HIP-3 went further. In the six months since it launched, seven independent teams had deployed hundreds of markets, most of them for assets that had nothing to do with crypto: oil, gold, stock indices, foreign exchange. Trade[XYZ], the largest deployer, had been growing 38% a week since October 2025. It had done more than $130 billion in volume across 192,000 traders. Markets created by independent deployers now accounted for half of Hyperliquid’s total volume. In February 2026, HIP-4 was announced. When it arrives, anyone will be able to deploy options or prediction markets on the platform. HIP-3 has opened Hyperliquid to any asset with a price. HIP-4 will open it to any event with an outcome. 

The most consequential products on Hyperliquid are now being built by people who do not work for Yan and never will. I asked him how he thought about that. What his team should build, and what should be left for others.

“It’s a dynamic question, and I don’t think there’s a correct answer,” he said. “The most important angle is philosophical. Are you building a finance super app, like Robinhood, or are you building a financial system?” He conceded that he didn’t know which would win. “But I think an accessible financial system is a better outcome for the world. One that’s on rails that are public and not owned by a single company.

“To build that, we often think about what we need to do for others to come and succeed and own a business on Hyperliquid. When people compete and own their own thing, it leads to a more resilient and scalable system.”

He said the path of least resistance was to build everything yourself, keep it within one company. They had chosen the opposite. “It’s the hard way to do things, but we care about how we get to our goal, because how we get there determines what it actually is at the end that we’ve built.”

Trade[XYZ]’s founder told me he thought a day might come when nobody would even know they were using Hyperliquid. “Maybe in the final state, it’s simply the infrastructure and liquidity for finance,” he said. “And perhaps Interactive Brokers and Phantom, and whoever else, end up interacting with users. That’s kind of wonderful.”

Paradigm, Huang’s firm, invested a significant amount on the open market soon after the HYPE token launched. “It’s all the more amazing,” he told me, “because it’s been built by a team of 11 people.” Eleven people, and barely any AI. In the office, there were separate AI laptops that ran the latest models. They were used to explore ideas only. “We carefully monitor AI’s capabilities,” Yan said. “It’s not good enough yet to write important code.”

At the opening price, the airdrop was worth over $1 billion. At all-time highs, it would reach $16 billion. It was the largest wealth transfer in the history of cryptocurrency, and every dollar of it went to users.

I asked Yan about the biggest cloud hanging over all of it. Hyperliquid has done over $4 trillion in cumulative volume since 2023. It has 37% of the decentralized perpetual futures market. And it has done all of it without users in the largest capital market on Earth being able to touch it. Americans are locked out.

The obstacle is Dodd-Frank, a U.S. law passed after the 2008 financial crisis, which requires every derivatives transaction to flow through a regulated intermediary. The irony is that Hyperliquid’s public ledger already gives regulators what Dodd-Frank was designed to produce: real-time visibility into all the leverage in the system. But until the U.S. Commodity Futures Trading Commission writes new rules, there is no legal path for Americans to trade derivatives through a decentralized protocol. True to his philosophy, Yan did not build a policy team himself. A month after my visit, the Hyperliquid Policy Center launched as an independent nonprofit, led by Jake Chervinsky, a prominent lawyer who had spent a decade in crypto. The Hyper Foundation, an independent body supporting the growth of the Hyperliquid ecosystem, contributed one million HYPE tokens, worth $28 million, to fund its launch.

Yan acknowledged that Hyperliquid had reached a size that meant building and hoping was no longer a strategy. “There are people lobbying in the other direction,” he told me. “I can’t say with super-high confidence which way it’ll end up. But regulation ultimately reflects the will of the people, and I’m optimistic about where it’s going.”

There was one final question I had been saving all week: You don’t really think Hyperliquid will house all of finance, do you?

He smiled, which he does more than you’d expect for a man who cuts his own hair. “I mean, I suppose ‘all’ is a bit of a superlative,” he said. “That is our aspirational goal. But it’s really hard to do, and multi-decade goals are very presumptuous.”

“It’s the difference between Go and chess,” he continued. “In chess, the better you get, the more moves ahead you read. In Go, there are just too many possibilities. The focus is more on building intuition for the next move rather than trying to read the whole tree.”

I must have looked like I needed more, because he put it yet another way. He had always tried to live by this principle: Be very confident you’re going in the right direction, and execute well on the step you’re taking right now, without knowing exactly where you’re going.

The following evening, a Friday, the team went to dinner at a Chinese restaurant inside one of the city’s hotels. The engineer whose stuffed animals had colonized the office couldn’t make it. Everyone else was there, plus me. We were led through a quiet lobby and down a corridor to a private room paneled in dark wood, with carved lattice screens and a round table set for dinner. At the far end, behind a partition, armchairs were arranged around a coffee table. We sat there first and drank tea. 

The room was cold, the air-conditioning set for a warmer evening than this one. Someone handed the youngest engineer a blanket. He pulled it over his shoulders and discovered it was Christian Dior. This prompted a conversation with Yan about luxury brands, a topic on which neither had any evident background. One of them pronounced LVMH as “LHVM.” Neither corrected the other. iliensinc, in her Ralph Lauren cap, sighed.

When we moved to eat, the lazy Susan began to turn and didn’t stop. The dishes were placed around its edge, one after another, until a wide blue-and-white porcelain bowl arrived, and the table went quiet. Inside it, a shallow pool of water covered a bed of pebbles and tiny leaves—a koi pond in miniature. A scalloped white bowl of noodles was placed at the center, and three small orange fish circled it, swimming the moat between the two bowls. The waiter explained the dish to us. The fish, he said, rest for 30 days so they can work for five minutes. We watched them swim their circles. Then they were taken away to begin another month off.

We left around 9:15pm, walking out into a light rain. I said goodbye and got into a taxi to the airport. A few minutes from the hotel, the taxi began to climb a long left-hand bend up to the expressway, and as it came around, the financial district swung into view: HSBC, J.P. Morgan, Standard Chartered, Deutsche Bank, Citi, their logos bright against a black sky. Then the road straightened east, and they fell away behind me, tower by tower, until there was nothing in the mirror but the wet road. Yan had gone in the opposite direction, back to work, where he’d left his bodyguard.

Dom Cooke is the managing editor of Colossus.