A photograph posted on X on 19 October, captioned “got the HRT gang back together”, connected three of the most consequential young companies in AI and crypto finance. The three men in it had spent three weeks competing against each other as undergraduates.
Ten interns, three weeks, one network
In January 2016, Alexandr Wang was a first-year MIT student on a winter internship at Hudson River Trading, the quantitative trading firm. Around ten to twelve young people were in the room, most of them from Harvard and MIT. Among them were Jeff Yan, who later founded Hyperliquid, Scott Wu, who founded Cognition, and Jesse Zhang, who founded Decagon. The programme ran for three weeks in competition format, and Wu and Yan finished first and second throughout.
The connections predate the internship. Wu has said he met Wang and Zhang on the American maths and programming competition circuit in secondary school. Wang was the first of the group to actually leave and build something, dropping out of MIT in 2016 to co-found Scale AI.
The current numbers are not small. Cognition closed more than USD 2 billion on 8 September, led by a16z and Accel, at a USD 48 billion valuation. Its annualised revenue run rate went from USD 492 million in May to close to USD 900 million. Hyperliquid generated about USD 105 million of revenue in September, according to Grayscale research published on 30 October. Decagon raised USD 250 million on 27 January, led by Coatue and Index Ventures, at a USD 4.5 billion valuation, and has added more than 100 enterprise customers including Avis Budget Group, Block and Deutsche Telekom.
How Wang got to Meta
In June 2025 Meta invested in Scale AI at a valuation above USD 29 billion. Wang, then 28, stepped down as chief executive, joined Meta to run its AI business and kept a board seat at Scale.
Zuckerberg has described the recruitment. The first conversation lasted about 44 minutes. Two days later Wang sent a long memorandum on how he would restructure the AI organisation, arguing for higher talent density in the core team and a rebuilt research stack. Zuckerberg then personally ran the hiring. Wang has said the core research group in Meta Superintelligence Labs is under 200 people and closer to 100, with much larger infrastructure and product teams around it.
By September Meta had shipped Muse, a personal AI agent. On 12 October the lab released Muse Wonder and Muse Wonder Full. After more than a year of reorganisation, Meta could put frontier models in front of competitors again. That is what turned an operational role into speculation about succession after the Financial Times profile.
The succession question has an answer in the filings
Four days after the succession report, Meta disclosed a new equity structure. SEC filings show Zuckerberg holds about 13.4 per cent of the economic interest and about 60.6 per cent of the voting power. The disclosed negotiations state plainly that he would not support financing or other arrangements that cost him voting control.
That does not settle who runs Meta in a decade. It does settle that the decision will not be made by the market, and it puts a ceiling on how much of the succession story is actually about governance rather than about narrative.
Is this the next PayPal mafia
The comparison is unavoidable and partly wrong. PayPal was sold to eBay in 2002 for USD 1.5 billion, and its alumni went on to found LinkedIn, YouTube, Yelp and Palantir, while Peter Thiel co-founded Founders Fund and became Facebook’s first outside investor. That group built its network after a shared operating experience, and then converted it into capital.
The HRT group built its network before founding anything, through competitions and a three-week internship. What it has now is early shared history, some talent flow (Cognition co-founder Steven Hao appeared on a 2018 Scale AI engineering list) and one shared investor: Founders Fund backed both Scale AI’s USD 1 billion round in May 2024 and Cognition’s E round this year.
The difference matters for anyone reading the network as a signal. The PayPal group could move capital and people across dozens of companies because it had already done it. This group has proven it can build three valuable companies. It has not yet proven it can seed the next twenty, and that is the part of the mafia model that compound returns actually come from.
Editor’s note: This is an adapted translation of the original Zhidx report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.zhidx.com/p/607691.html.