IBM CEO 亲自点破 AI 投资回报的数学难题，做 AI 基础设施或模型投资的团队值得细读——连卖铲子的人都说钱可能回不来。
IBM CEO Arvind Krishna 表示 AI 并非泡沫，但估计行业需要 6 到 8 万亿美元的总资本支出用于数据中心和芯片建设。要在七年内收回这些投资，公司每年需要新增 1 到 2 万亿美元的收入，而 Krishna 认为这笔收入可能不存在。他还预测只有两到三家公司能成功构建领先的 AI 模型，其他公司只是在为一场多数人会输的竞赛花钱。评论指出，Krishna 作为 AI 基础设施服务商 IBM 的 CEO，其质疑比普通怀疑者更有分量。谷歌刚通过股权融资 800 亿美元，Oracle 裁员 3 万人以转投资本支出，而 Anthropic 提交上市申请的同一天 GitHub Copilot 的 token 计费崩溃、用户流失。
hard to make all these numbers square
hard to make all these numbers square Hedgie @HedgieMarkets 🦔IBM CEO Arvind Krishna says AI is "not a bubble" then estimates the industry needs $6 to $8 trillion in total capex for data center and chip buildout. To recover that over seven years, companies would need $1 to $2 trillion in new annual revenue. Krishna says he doesn't believe that revenue exists. He also thinks only two or three companies will succeed at building leading AI models. Everyone else is spending to stay in a race most of them will lose. My Take "Not a bubble" but $6 to $8 trillion chasing revenue that the guy doing the estimate says probably isn't there. I don't know what definition of bubble Krishna is using, but by mine that qualifies. He's the CEO of IBM, a company that sells AI infrastructure services to enterprises. If he's flagging that the payback doesn't work, and he benefits from the spending continuing, that should carry more weight than a random skeptic saying the same thing. Google just sold $80 billion in equity to fund its buildout. Oracle cut 30,000 workers to redirect salaries into capex. Anthropic filed its S-1 the same day GitHub Copilot's token billing collapsed and users fled. Krishna put a number on what everyone else is dancing around. The spending has outrun any realistic estimate of the revenue it can generate, and the companies writing the checks know it. They're just betting they're one of the two or three that survive long enough for it to pay off. Hedgie🤗 🔗 View Quoted Tweet 💬 6 🔄 0 ❤️ 12 👀 1020 📊 5 ⚡