本报告聚焦金刚石散热在AI芯片中的应用潜力,阐述其超高热导率(室温2000-2200 W/m·K)和低热膨胀系数带来的热管理优势,并概述核心技术路径如金刚石热沉、金刚石/金属复合材料及微通道液冷的集成。报告指出2026年可能成为金刚石散热的规模化元年,英伟达 Vera Rubin 架构GPU将采用“金刚石铜复合散热盖+45摄氏度温水直液冷”方案,2030年市场空间有望达到53.5亿美元。当前全球企业已进行技术研发与验证,中国企业具成本优势,但高端CVD技术、品牌认知与认证周期等方面仍需突破,存在成本、工艺成熟度及需求不及预期等风险。[page::1]
This paper studies how workers’ beliefs about pay shape the tradeoffs between pay and workplace amenities. Using a four-stage incentivized survey that combines hypothetical choice experiments with elicited pay beliefs and randomized pay-information treatments, the authors document four core findings: (i) baseline pay beliefs are systematically biased downward (about 18% below actual starting pay) and are highly dispersed across individuals and jobs; (ii) respondents over-attribute pay to advertised amenities, yielding a steeper belief slope on amenity value than the actual pay slope, i.e., an augmenting-differentials pattern; (iii) a short-term pay-information intervention raises mean pay beliefs by about 4% and reduces belief dispersion by ~15% but does not alter the amenity-pay slope or stated tradeoffs; (iv) full information (persistent pay transparency) moves beliefs and choices toward higher-paying jobs, aligning implied willingness-to-pay for amenities with full-information benchmarks. These results help explain why hypothetical-choice studies show strong amenity valuations while observational data often fail to show compensating differentials, and they illustrate how pay-disclosure policies may reshape beliefs and, with sustained salience, tradeoffs [page::1]]
This paper develops a dynamic competitive limit-order-book model in which uninformed liquidity demand exhibits heavy tails (Student-t with ν > 2). The key insight is that the tail risk of liquidity demand acts as a state variable that modulates how informative large trades are about fundamental value. When liquidity shocks are light-tailed, very large orders strongly signal private information, but with heavy-tailed liquidity demand, the same large imbalances can reflect rare liquidity shocks, leading to flatter and more concave price impact and slower learning. The authors establish a fixed-point characterization of the equilibrium on a tail-controlled compact class, prove existence of such fixed points, analyze Bayesian learning of the fundamental value, and derive tail asymptotics for price impact that depend on ν, the number of informed traders, and the posterior beliefs. They also provide numerical solutions showing nonlinear, concave margins, persistence of adverse-selection premia, and power-law scaling in large-order regions, consistent with the analytical results. This work identifies liquidity tail risk as a structural variable shaping price discovery, spreads, and market impact [page::0][page::2].
We develop a two-sector, three-labor-type model where AI is produced by labor and used in final goods production, yielding endogenous labor reallocation as AI productivity rises [page::2][page::3]. AI adoption creates a transition in which substitutable labor incomes fall in absolute and relative terms while complementary labor wages rise faster than GDP, and final goods labor tracks GDP one-for-one [page::2][page::3]. The model identifies threshold productivity levels that separate pre-AI, transition, and post-AI regimes; the wage ratio between complementary and substitutable labor depends only on AI productivity and AI-production intensity, not on relative labor supplies [page::3][page::10]. We analyze Pareto-improving tax and regulatory policies that share AI gains across workers, and show how outcomes differ under competitive AI production versus monopoly AI production [page::4][page::21]. The paper also extends to AI-in-AI production and to CES-type AI technologies, confirming robustness of the main qualitative results and highlighting how policy design must adapt as AI productivity evolves [page::30][page::31].
This paper estimates the effect of cross-border transmission constraints on suspected market power abuse in the German wholesale electricity market. Using a 2SRI instrumental variables approach, we study suspected strategic behavior by German gas- and coal-fired power plants in 2022-2024. Cross-border transmission constraints are measured using the maximum and minimum bounds of zonal net position, while suspected market power abuse is measured as the upward or downward deviation of observed dispatch from a modeled competitive benchmark. We find that transmission constraints significantly elevate the likelihood of suspected market power abuse. When headroom for further imports is already scarce, reducing import headroom by one Gigawatt (GW) increases the odds of suspected capacity withholding by 15%. Similarly, reducing export headroom by one GW when it is scarce increases the odds of suspected capacity push-in, a strategy to depress prices, by 16%. These results provide empirical support for interconnection expansion as an instrument to mitigate market power. [page::1]
本研究提出一个连续时间CVaR框架,将两类核心尾部保护并入同一目标:看跌期权(凸性保险)与趋势跟踪(系统性趋势对冲),通过标记市值的期权叠加以解决保费拖累与跳跃定价等不一致问题,并在财富、现货、波动率及趋势信号组成的马尔可夫状态下进行最优控制。理论贡献包括对冲机制的时序分工(跳跃冲击时看跌期权即时再定价,趋势在持续回撤中逐步变防御且无需额外保费)、希尔顿-尤拉塞夫CVaR表述下的HJB方程推导,以及一个四轴诊断层次用于区分条件凸性、尾部可靠性、非压力携带与回撤持续性。通过 stylized Monte Carlo 实验表明,等权混合策略与网格优化混合策略均可在不同情景下降低终端CVaR,相比单一 sleeve,混合权重的最佳位置具有 Calibration 依赖性。研究框架为尾部风险管理提供一个透明的多机制分解与实施路径,帮助决策者在“应对崩盘的凸性保险”和“应对持续回撤的信号驱动保护”之间进行结构化配置 [page::0]