Dr. Nianzu Xiong

In this paper, we study how a platform should raise capital when the entrepreneur privately knows the platform's prospects and the buyers of its tokens become its users. The platform can issue debt, outside equity and tokens. We first show that, under any investor belief, a contract affects the entrepreneur only through two numbers: what investors receive if the platform fails, which determines how much of the contract is mispriced, and how many investors become users. Debt is the best security on the first margin and tokens on the second, while outside equity is dominated on both. Under pooled beliefs, the high type therefore issues safe debt first, then tokens, and risky debt only once token capacity is exhausted, and airdrops are dominated by sales unless sales already cover the investment. Then we study the signaling game. In the baseline, the low type sells more tokens than the high type in every separating equilibrium, so a larger token sale is bad news. Refinement selects least-cost separation for small projects, pooling for large projects, and semi-separation in between. Furthermore, we show that token sales become good news when users are costly to serve and good platforms retain users better. Finally, issuance costs cap token sales at an interior level, and a retention role for tokens can put them ahead of debt.

In this paper, we model the race for the AI frontier as an all-pay contest in which two labs buy compute from a supplier that earns a margin on every dollar they spend. The prize grows with both labs' spending through knowledge spillovers. Ownership links the layers: the supplier holds shares in the labs, and each lab's major investor holds shares in the supplier and possibly in the rival lab. We derive the symmetric equilibrium in closed form and show that circular holdings raise compute if and only if the lab's stake in the supplier, weighted by the margin, exceeds its stake in the rival weighted by one minus the spillover. The supplier's stake in a lab is neutral on its own, but it amplifies the lab's stake in the supplier. Unlike in R&D contests across industries, a lab's major investor can gain from a compute boom, because the supplier's margin is not competed away. The impossibility theorem reappears at the level of the ownership network: if the spillover plus the in-network margin is below one, no circular arrangement can raise compute and benefit its participants. We then show that exclusive deals discourage the rival lab and create value exactly when inclusive deals destroy it, that the labs' outside investors bear the losses from value-destroying booms, and that circularity makes compute more sensitive to margin compression. Finally, we add a racing layer of data-center and power developers.

In this paper, we study how elite institutions select people once generative AI can do much of the technical work that has historically been the main route of entry for outsiders. Technical skill can be checked by anyone, whatever the candidate's background. The skill that matters after AI is judgment, which is harder to observe, and information about it comes from referees who are not disinterested. We first show that letters from referees at other institutions carry no information in any equilibrium, because those referees want their students placed regardless of quality, while sponsors inside the institution share in its reputation and transmit information credibly, but with a bias. We then show that, as AI makes checkable skills less valuable, outsiders are under-selected even by unbiased evaluators, any given favoritism matters more, the apprenticeships that once trained and revealed talented outsiders disappear, and winning a patron gets easier. AI can push the other way if it is used to verify ideas: a simple pass–fail test to which the outsider's home institution commits in advance places outsiders more often than full disclosure would. Finally, because reputation is shared among the members of an institution, these forces erode quality slowly and invite entry by institutions that evaluate outsiders through verifiable channels.