In September 2021, four friends decided that one month was enough time to build an NFT marketplace. Then they heard a competitor was preparing to launch on Solana. The month became about eight days. Zhuojie Zhou and Sidney Zhang took the coding half of the assignment; Jack Lu and Zhuoxun Yin took the plan and the creators. By the scheduled launch, Zhang remembered, the product was still shaky. An outage in the wider ecosystem handed the team two more days of repairs. Magic Eden went live roughly ten days into the sprint.
This is the sort of tale startup founders tell with a smile once the servers have survived. For Zhou, better known as Rex, the sprint was an unusually public test of skills accumulated in less theatrical places: academic papers about graph sampling, feature flags at Checkr, developer infrastructure for PyTorch, and the stubborn work of making software reliable enough for strangers to use. The drama of a launch is immediate. Its preparation can take a decade.
A dissertation about finding your way
Before there was a marketplace, Zhou studied networks. At George Washington University, his research dealt with a deceptively awkward question: how can you sample a huge online social network when you cannot simply inspect the whole thing? His doctoral dissertation, defended in May 2015, was titled “Faster Sampling Over Online Social Networks.” His published work examined ways to make those walks through a network faster and more useful.
It is tempting to turn every detail of a founder’s education into destiny. That would be too neat. Graph sampling did not prescribe an NFT business. It did give Zhou a practiced way to think about a system whose shape is only partly visible, whose participants are connected, and whose useful paths must be discovered under constraints. Much later, a marketplace would pose a human version of the same puzzle. Creators need collectors; collectors need things worth finding; the software has to bring both sides together before either loses patience.
There was an entrepreneurial detour even during graduate school. A faculty adviser’s CV lists Zhou on the WiseAgg team that won George Washington University’s 2012 business plan competition, first among 144 submissions. The record gives a glimpse of a researcher willing to test an idea outside the lab. It does not tell us what he thought the next ten years would look like. By 2014, according to his professional biography, he had also begun exploring cryptocurrency through mining, years before the token market would become his full-time work.
The release switch
Zhou later worked as an early engineer at Uber and Checkr. At Checkr, he published detailed accounts of two open-source projects that are less glamorous than a digital collectibles market and perhaps more revealing of his habits. Flagr gave developers a way to switch features on gradually, test variations and change configuration without redeploying an entire application. OpenMock helped teams imitate the services their software depended on, so they could test their own code without waiting on the outside world.
Feature flags sound like a small convenience until a product has users who depend on it. Then a release is a negotiation with risk. Should everybody see the new behavior at once? Should a few users see it first? How will the team know whether the change helped? Zhou’s 2017 explanation of Flagr spends its energy on exactly these questions. It describes evaluators, metrics and practical rollout decisions, then shows how Checkr used the service in production. The point was not a clever toggle. It was room to learn without making every experiment irreversible.
“We were small and paranoid.”Zhou on Magic Eden’s early fundraising decision
There is a professional temperament in that choice. OpenMock asked a related question: how do you move quickly when the thing you need to test is entangled with everything else? His published description laid out mocks for HTTP, Kafka and AMQP. Again, the visible result was a tool; the underlying obsession was control over an unruly system. Zhou’s public biography also credits him with contributions to Kong and with later work on Facebook AI’s PyTorch developer infrastructure. His own LinkedIn posts speak warmly about maintaining that stack. To a user, infrastructure appears only when it fails. To an engineer, that invisibility can be the achievement.
A marketplace in a hurry
The opening Magic Eden found was specific. Solana offered a growing home for NFTs, but a polished marketplace for creators and collectors was still up for grabs. Lu saw the opportunity while working in crypto in Hong Kong. Zhang wanted to return to the field. Yin brought exchange and product experience. Zhou, who had worked with Zhang at both Uber and Facebook, became the fourth founder and the chief engineer.
They divided the labor plainly. Lu and Yin wrote down what the site should become and looked for creators. Zhang and Zhou wrote what users could actually click. The first version did not have the luxury of time or a settled market to imitate. The founders took a week off their regular jobs; the product arrived after the unexpected two-day reprieve. By launch, fewer than 20 creators were aboard. Within two days, Lu recalled, that count had reached 50. Creators brought their own communities with them, and the network began to act like a network.
Revenue followed quickly. Within weeks, Magic Eden was bringing in about $100,000 a day, according to the founders’ later account. It was a welcome figure and a disquieting one. A crypto marketplace can have a splendid Tuesday and a miserable Friday. Zhou remembered the team as “small and paranoid,” uncertain how long the income would last. They raised seed money about a month after launching as a backstop. There is a subtle contrast between the engineering instinct to control a rollout and a market that will not accept a feature flag. The former helped them ship; the latter taught them to keep reserves.
The company stayed distributed. Its first employees were scattered across countries; the founders kept hiring beyond San Francisco. In June 2022, less than a year after launch, Magic Eden announced a $130 million round at a $1.6 billion valuation. The funding figure became an easy headline. Zhou’s remark about paranoia is a better description of the job. A company can be valued as though the future is obvious while its engineers are still studying tomorrow’s traffic.
When the chain changes, the work changes
Magic Eden began with Solana, then expanded to other blockchains and products. Each addition changed the engineering problem. Ethereum and Polygon came with their own standards and communities. Bitcoin Ordinals brought a different method of making digital collectibles and a different set of technical habits. The marketplace added a wallet and other tools. What had been a race to assemble one coherent place to trade became a continuing exercise in joining systems that were built apart.
A colleague’s public account credits a weekend side project by Zhou and Kevin Faveri with the start of Magic Eden’s Bitcoin Ordinals product. That is a striking echo of the company’s origin: a small experiment, made quickly, meeting an emerging community at the right moment. Zhou later posted about the details of Bitcoin transactions, including partially signed Bitcoin transactions and atomic swaps, with the enthusiasm of somebody who finds the plumbing at least as interesting as the shop above it.
At Solana Breakpoint in 2023, he returned to a theme from the Checkr years: tools should be open enough for other builders to use and inspect. He described how difficult the Solana ecosystem had felt to a newcomer writing a marketplace contract, and how an openly shared example helped him learn. In his own LinkedIn writing, he connected Flagr and PyTorch to his enthusiasm for building in public. The argument was practical as well as philosophical. A public protocol gives the next team a foothold that his team had once needed.
“Creators are always the central of what we’re doing.”Rex Zhou, speaking about the marketplace’s creator community
That sentence points beyond code. A marketplace is a piece of software, but the software has no reason to exist without people making things and people caring enough to collect them. Zhou told a Breakpoint audience that NFTs can hold a community’s culture. The observation is almost comically far from the precision of a graph-sampling dissertation, yet the path between them makes sense. A network is not merely its nodes and edges. A successful product must understand why people move through it.
The engineer’s version of magic
Zhou’s public record does not read like a founder’s mythology. It is full of talks about API gateways, posts about feature deployment, repositories, research papers and notes on protocol design. Even the Magic Eden launch story, with its ten-day clock and sudden rise, contains a quiet engineering admission: the product needed those two extra days. The outage was inconvenient for the ecosystem and useful for a team still fixing its first release.
By the time the company had a valuation large enough to attract easy adjectives, Zhou was still talking about the work beneath the market: how to build across chains, how to give creators better tools, how to share infrastructure. The $1.6 billion figure describes what investors believed Magic Eden might become in 2022. It cannot describe the years of research and engineering that made its first ten days possible, or the changes required after that first market shifted.
There is a pleasant irony in a company named Magic Eden being partly built by somebody so interested in removing mystery. Zhou’s tools expose the switches, mocks and protocols that make a product behave. They turn apparent magic into a sequence of decisions that other engineers can see, test and improve. The crowd arrived quickly. The engineer had been working on the doors for years.