Meta正积极发展云端运算业务，计划透过出售算力及AI模型存取权来创造营收。此消息由Bloomberg率先报导后，周三科技股剧烈波动：Meta股价飙升8.8%，创下五个月来最佳单日表现，Amazon与Google等云端供应商同步受惠；反之，Nvidia与Advanced Micro Devices等晶片股下跌，而新兴云端公司CoreWeave更暴跌14%，反映市场对Meta加入竞争的担忧。市场普遍将此解读为对晶片及算力供应链的利空讯号——身为AI基础设施最积极投资者之一的Meta若开始出售算力，似乎暗示其产能已然过剩。

然而，实际情况并非如此简单。Meta近期才与Crusoe及CoreWeave等伙伴签订合作协议，并持续扩建自有资料中心。执行长Mark Zuckerberg公开阐述其策略：在AI竞赛中宁可过度投资也不愿投资不足，并形容此举为「积极前置布建产能」。Google近期限制Meta存取其Gemini模型一事更印证了算力几乎不存在「过多」的问题。云端业务实质上为Zuckerberg提供了一道安全阀——他在去年十月及今年五月均坦承存在过度投资的可能性，并表示若算力确实过剩，随时可将多余产能出售给需求看似永无止境的市场。Jefferies分析师Brent Thill亦回应称，Meta并非退出AI竞赛，而是将早期积极的产能承诺转化为策略性价值创造选项。

更深层来看，Meta进军云端市场有多重策略意涵。首先，这是将巨额AI投资转化为即时获利的途径——Amazon与Google已证明可靠的云端服务每季可产生数百亿美元营收。其次，若Meta以较低价格销售AI token，可能迫使竞争对手降价，而Meta本身作为token的大量买方亦将从全行业降价中获益。然而，Meta的进入势必冲击现有业者，尤其是CoreWeave等专门向大型AI公司销售算力的新兴云端企业，不仅面临客户竞争加剧，整体市场定价亦可能受到压力。尽管Meta短期内不太可能放缓基础设施建设步伐，但在投资者日益关注回报的背景下，Zuckerberg或许正在未雨绸缪——若其消费者端AI产品（包括聊天机器人与智能助理）未能如预期般成功，云端业务将成为一份明智的保险策略。

Meta is actively developing a cloud computing business to monetize its computing power and AI model access. The news, first reported by Bloomberg, triggered significant market reactions on Wednesday: Meta shares surged 8.8%, marking their best single-day performance in five months, while fellow cloud providers Amazon and Google also benefited. Conversely, chip stocks like Nvidia and Advanced Micro Devices declined, and neocloud company CoreWeave plummeted 14% on fears of heightened competition. The market broadly interpreted Meta's move as bearish for the chip and computing-power supply chain—the notion that Meta, one of the most aggressive spenders on AI infrastructure, would sell computing capacity suggested it had accumulated too much.

In reality, however, the situation is more nuanced. Meta recently signed deals with partners such as Crusoe and CoreWeave and continues to expand its own data center fleet. CEO Mark Zuckerberg has openly articulated a strategy of overinvesting rather than underinvesting in the AI race, describing it as an effort to "aggressively front-load building capacity." Google's recent decision to cap Meta's access to its Gemini model further underscores that excess capacity is virtually nonexistent. The cloud business effectively provides Zuckerberg with an escape hatch—he acknowledged the possibility of overbuilding last October and again in May, stating that surplus capacity could always be sold into a market with seemingly insatiable demand. Jefferies analyst Brent Thill echoed this view, arguing that Meta is not stepping away from the AI race but rather converting early, aggressive capacity commitments into a strategic value-creation option.

On a deeper level, Meta's cloud entry carries multiple strategic implications. First, it offers a pathway to convert massive AI investments into immediate revenue—Amazon and Google have demonstrated that reliable cloud offerings can generate tens of billions in quarterly revenue. Second, if Meta sells AI tokens at lower prices, competitors may be forced to cut their own rates, and Meta itself, as a heavy buyer of tokens, would also benefit from industry-wide price reductions. Inevitably, however, Meta's entrance will create losers, particularly neoclouds like CoreWeave that sell computing power to major AI companies, facing both intensified customer competition and potential pricing pressure across the broader market. While Meta is unlikely to slow its infrastructure push anytime soon, Zuckerberg may be building a hedge before it is needed amid growing investor focus on returns—a smart insurance policy should its consumer AI efforts, including chatbots and agents, fail to deliver the expected success.