全球对「主权人工智慧」（由 Nvidia 执行长 Jensen Huang 定义为五层基础设施架构的概念）的竞争，已从寻求数位自主的诉求，转变为美国和中国的核心治理政策。两国政府体认到人工智慧是关键的国家基础设施，而非仅是软体，因而正从监管转向直接的股权参与。这一战略转折的标志是 Trump 政府提议入股 OpenAI 和 xAI 等前沿人工智慧公司，这与其先前为保护国内半导体能力而对 Intel 进行的 10% 股权投资相呼应。

与此同时，中国正在执行类似的「晶片到模型」战略，并在国家引导的资金支持方面拥有十年的领先优势。中国国家人工智慧产业投资基金最近主导了 DeepSeek 的一轮重大融资，使其以超过 500 亿美元（$50bn）的估值成为该国最具价值且仅专注于人工智慧的新创公司。值得注意的是，该国家投资工具是唯一获得直接公司所有权和投票权的投资者。这项投资标志著从支持硬体制造商转向保障模型层的过渡，其中 DeepSeek 是专门设计运行于中国国产的 Huawei Ascend 晶片集群上。

全球两大数位经济体向国家所有权的转变，带来了前所未有的治理挑战。借由将公共资本直接投入具竞争力的人工智慧实验室，两国政府正将国家安全目标置于传统商业增长和股东回报之上。随著地缘政治竞争越过保障实体晶片的第一阶段，以及支持软体模型的第二阶段，它正进入一个关键的第三阶段，其焦点在于智慧本身的直接所有权和主权治理。

The global race for "sovereign AI"—a concept defined by Nvidia CEO Jensen Huang as a five-layer infrastructure stack—has transitioned from a pitch for digital self-determination to core governing policy for the United States and China. Recognizing AI as critical national infrastructure rather than mere software, both governments are shifting from regulation to direct equity participation. This strategic pivot is highlighted by the Trump administration's proposal to acquire equity stakes in frontier AI firms like OpenAI and xAI, mirroring its previous 10% investment in Intel to protect domestic semiconductor capabilities.

Simultaneously, China is executing a similar chip-to-model strategy with a decade-long head start in state-led funding. China’s National Artificial Intelligence Industry Investment Fund recently led a major funding round for DeepSeek, making it the country's most valuable AI-only startup with a valuation exceeding $50 billion. Notably, the state vehicle was the sole investor to obtain direct corporate ownership and voting rights. This investment marks a transition from backing hardware manufacturers to securing the model layer, with DeepSeek specifically designed to run on domestic Huawei Ascend chip clusters. (Key numbers: 500, 50bn)

This transition to state ownership in the world's two largest digital economies introduces unprecedented governance challenges. By investing public capital directly into competitive AI laboratories, governments are prioritizing national security objectives over traditional commercial growth and shareholder returns. As the geopolitical race moves past the first phase of securing physical chips and the second phase of backing software models, it enters a critical third phase focused on the direct ownership and sovereign governance of intelligence itself.