近十年来，OpenAI 一直是 Microsoft 的资产：先为其云端部门品质背书，后以尖端模型让这家软体巨头在 AI 时代抢先。上周三，对其最大投资者而言，OpenAI 变成了新的东西：一项负债。在 Microsoft 的财报电话会议上，Jefferies 分析师 Brent Thill 指出，OpenAI 占 Microsoft 未来销售 backlog 的 45%，并直言市场担心其「耐久性」与「曝险」。作者称其已听过约 100 场此类电话会议，这次的提问像是在挑战 AI 荣景的核心前提：OpenAI 是否会回报。Microsoft CFO Amy Hood 表示合作「很棒」，并称这种合作让公司保持领先、站在应用创新前沿。

华尔街未被安抚。隔天，Microsoft 股价市值蒸发 $357 billion，属于历史级的抹除，且更引人注目的是其基本面结果依然扎实。这也不是投资人第一次突然、集体地重估 AI 时代的赢家与输家。Oracle 去年 9 月因 OpenAI 承诺向其购买 $300 billion 的算力而股价创新高，之后又因市场质疑该业务能否落地或获利而暴跌。Oracle 周一表示，今年将以债务与股权最多筹集 $50 billion，为其资料中心建设狂潮融资；有分析师形容这是在把自己从「相当大的洞」中挖出来。剧烈反应指向一种假设转变：哪些公司会受益于 AI、哪些会被其颠覆，以及 OpenAI 是否会产生支持者所押注的巨大营收。

Brent Thill 在采访中说「情绪已翻转」，并将其概括为两个问题：Sam Altman 是否能交付，以及其次他是否会开始「吃掉」合作伙伴的午餐。本周将透过 Google 与 Amazon 的季报再观察「谁在吃谁的三明治」。Google 在发布更强的 Gemini 版本后走强，其晶片吸引大客户，削弱 Nvidia 近乎垄断的地位，并迫使市场重视其云端业务；Alphabet 于周三公布。周四轮到 Amazon；Amazon Web Services 自 ChatGPT 出现以来增速慢于最大对手，分析师称新资料中心与自研晶片可能加速增长，但预期尚未反映；Bloomberg 汇编数据显示，分析师预测至 2027 年增长大致维持在 21%。OpenAI 的动向也相关：最近完成的季度首次纳入其与 AWS 的七年、$38 billion 算力采购协议的一部分，且据报两者正洽谈由 Amazon 对 OpenAI 投资最多 $50 billion。

For almost a decade, OpenAI was an asset for Microsoft: first by vouching for the quality of its cloud unit, and later by supplying cutting-edge models that gave the software giant a jump in the AI age. Last Wednesday, OpenAI became something new for its biggest investor: a liability. On Microsoft’s earnings call, Jefferies analyst Brent Thill pointed to OpenAI’s 45% share of Microsoft’s future-sales backlog and said the market is concerned about durability and exposure. The author, having covered roughly 100 such calls, frames the question as a blunt challenge to a core AI-boom premise: whether OpenAI will pay off. Microsoft CFO Amy Hood called it a “great partnership” that keeps the company on the cutting edge of app innovation.

Wall Street was not reassured. The next day, Microsoft shares erased $357 billion in value, a historic wipeout made more striking by solid underlying results. This was not the first abrupt, mass reassessment of AI-era winners and losers. Oracle stock hit a record last September after OpenAI committed to spend $300 billion on Oracle computing power, then cratered as investors questioned whether the business would materialize or be profitable. Oracle said Monday it would raise up to $50 billion in debt and equity this year to finance a data-center construction spree, described by one analyst as digging itself out of a considerable hole. The violent reaction signals shifting assumptions about who benefits from AI, who is disrupted, and whether OpenAI will generate the prodigious revenue its backers expect.

Thill said “the sentiment has flipped,” boiling it down to whether Sam Altman will deliver and, secondarily, whether he will start eating partners’ lunch. This week brings another read as Google and Amazon report quarterly results. Google is riding high after releasing stronger Gemini versions, with chips winning major customers, denting Nvidia’s near-monopoly, and forcing investors to take its cloud business seriously; Alphabet reports Wednesday. Amazon reports Thursday; Amazon Web Services has grown more slowly than its biggest rivals since ChatGPT arrived, and analysts say new data centers and in-house AI chips could accelerate growth, though that is not yet in expectations; Bloomberg-compiled data show forecasts of roughly 21% steady growth through 2027. OpenAI also matters: the recent quarter is the first to include a slice of its seven-year, $38 billion deal to buy computing power from AWS, and the companies are reported to be in talks that could see Amazon take a stake of as much as $50 billion in OpenAI.