自 2025 年中以来，硅谷至少出现了三起重大的 AI“并购式招聘”：Meta 向 Scale AI 投资超过 140 亿美元并吸纳其 CEO；Google 以 24 亿美元许可 Windsurf 技术并将其联合创始人与研究团队并入 DeepMind；Nvidia 则押注 200 亿美元在 Groq 的推理技术上并招募其管理层。这些交易凸显了资本规模的急剧上升，以及人才本身成为核心资产的趋势。与此同时，Waymo、OpenAI、Anthropic 等前沿实验室之间的人才流动频率明显加快，研究人员在不到两年的周期内反复流动，形成高强度“音乐椅”式竞争。

这一变化被投资人称为科技创业公司的“解构化”。过去，创始人和早期员工通常至少待到四年期权归属，甚至直到上市或并购；而在生成式 AI 高速扩张的背景下，初创公司从一开始就被视为可能被拆分的组合资产。金钱是重要驱动因素之一，2025 年 Meta 向顶级 AI 研究员提供的薪酬方案高达数千万甚至上亿美元。同时，文化与机会成本也在变化，近五年内，越来越多计算机科学博士在完成学位前转向产业界，反映出长期停留的相对回报正在下降。

投资人因此加强防御机制，在交易条款中加入需董事会批准的知识产权许可限制。历史对比显示，这种流动性前所未有：1986 年成立的 Thinking Machines Corporation 从约 50 人扩张到 1996 年被收购时的 500 多人，期间人员流失极少；而如今，一名 AI 创业者一年内获得的产品发布、用户规模和资本密度，被形容为相当于过去五年的积累。数字表明，忠诚度正在被速度、规模和价格重新定价。

Since mid-2025, Silicon Valley has seen at least three major AI acqui-hire deals: Meta invested more than $14 billion in Scale AI and recruited its CEO; Google paid $2.4 billion to license Windsurf’s technology and absorb its founders and research teams into DeepMind; and Nvidia wagered $20 billion on Groq’s inference technology while hiring its leadership. These figures highlight how rapidly deal sizes have escalated and how talent itself has become the primary asset. At the same time, frontier labs such as OpenAI and Anthropic have experienced intensified churn, with researchers moving back and forth within cycles of less than two years.

Investors describe this shift as the “unbundling” of the tech startup. Historically, founders and early employees often stayed until a liquidity event, typically after a four-year vesting period. In the generative AI boom, startups are capital-rich and talent-driven, and are often valued as collections of transferable people and IP. Money is a key incentive: in 2025 Meta reportedly offered top AI researchers compensation packages worth tens or even hundreds of millions of dollars. Opportunity costs have also risen, with more computer science PhD candidates leaving programs for industry roles over the past five years.

Investor behavior is adapting accordingly, with contracts increasingly requiring board approval for major IP licensing. Historical contrasts underscore the scale of change: Thinking Machines Corporation grew from about 50 employees in 1986 to over 500 by its 1996 acquisition, with minimal attrition. Today, one year at an AI startup is likened to five years in earlier tech eras, given the speed of product launches, capital inflows, and user adoption. The data suggest loyalty has been repriced by velocity, scale, and compensation.