The caption was seven words: “got the HRT gang back together!” Alexandr Wang, Meta’s chief artificial intelligence officer, posted the selfie on X on October 19. Two faces filled the frame with him, both familiar to anyone who follows either artificial intelligence or crypto. The photograph was benign. The replies were not, because the three men are, by the reckoning of the industry’s more excitable corners, the most profitable intern class ever assembled.
The connection is a three-week program in the winter of 2015 at Hudson River Trading, the New York quantitative firm that makes markets in equities and derivatives with machines. HRT had never run an internship before. It took ten people. The format was built the way a trading firm builds everything, as a competition with rankings published after every round, and by Yan’s later recollection the top two places in every round were his and Wu’s.
Those two were Jeff Yan, then a junior at Harvard, and Scott Wu, a freshman. Wang was a freshman at MIT. Yan and Wu had both won gold at the International Olympiad in Informatics, and Wu’s record included a perfect score. Wang’s competition background ran through the American computing and mathematics olympiads. All three had grown up in the same small tournament circuit, which meant they had been reading each other’s code since they were teenagers.
A decade later the three occupy positions that did not exist when they met. Wang founded Scale AI, the data-labeling company that became the default supplier of training data to the largest model developers, and sold a 49 percent stake to Meta in June 2025 for 14.3 billion dollars, a transaction that valued Scale at about 29 billion and installed him as the head of Meta’s superintelligence lab. Wu runs Cognition, whose coding agent Devin was among the first products sold as an autonomous software engineer rather than an autocomplete. Yan built Hyperliquid, a perpetual futures exchange that runs on its own blockchain, raised no venture money, and still operates with a team small enough to seat at a few tables.
The numbers attached to those companies have moved faster than any of their job titles. A tally published this week by the Chinese technology publication TechFlow put the combined value of the businesses founded or led by the ten members of that 2015 class above 100 billion dollars. The three in the photograph account for most of it on paper. Paper is the operative word: none of the three has sold a controlling stake in the thing that made him rich, and two of the three run companies whose valuations rest on revenue that is either private or denominated in a token.
The temptation is to call them a mafia, in the way that the PayPal alumni of 1999 became a shorthand for a network that funded and staffed a generation of companies. The comparison is imprecise. PayPal was one company with a hundred-odd employees, and its alumni dispersed into ventures that shared a product philosophy. HRT was a trading firm that hired ten students for three weeks and then sent them back to school. What the class of 2015 shares is a screening mechanism, not a doctrine.
The firms that hired them have changed as well. Quantitative trading was, until roughly 2020, the default destination for the best twenty-two-year-old mathematicians in the country, offering starting pay that beat every other employer and a flat hierarchy that rewarded output. Then the model labs began paying the same money for work that carried equity in a company whose value could multiply, and the flow reversed. HRT now recruits machine learning researchers and publishes papers on model inference. Its internship program is a pipeline into an industry it no longer dominates.
For the three in the photograph, the shared history is a better story than a business plan. Wang’s position at Meta depends on delivering a frontier model against competitors with more compute and a longer record. Wu’s depends on whether enterprises will pay for agents that write code and take responsibility for the result. Yan’s depends on a market structure argument, that perpetual futures can clear on a public chain without the intermediaries that make the traditional version expensive, which is either obviously right or obviously wrong depending on the quarter.
There is a version in which the intern class is a coincidence: ten draws from a distribution of talented teenagers, three of whom happened to found companies in the same boom. HRT was not trying to build founders. It was trying to find people who could price a derivative faster than the person on the other side.
What the leaderboard could not measure is the thing all three did next, which was to leave trading for businesses with no established market. Wang left Scale’s data operation to run a lab inside a company that had been trying to catch up for three years. Wu built a product whose value customers could not verify until they had already bought it. Yan left market making to write a blockchain. The screening found them. It did not predict them.
The photograph will be reposted for years, and it will be read as evidence of something. The accurate reading is narrower. Three people who were good at the same thing at seventeen ended up in the same three industries at twenty-nine, and the industries happen to be the ones setting prices for everything else. Their classmate who went into medicine is not in the frame. The picture is a sample, not a thesis.
