AWS 发明了按需贩售运算基础设施的模式，至今仍是市占最大的云端供应商；但若只看自 2022 年末以来新增进积压的承诺金额，Amazon 反而落后于排名第二的 Microsoft，且略微落后于 Google。Google 过去被视为云端市场的「陪跑者」，如今借由其前沿 AI 技术，并以自研 AI 晶片出租给 AI 公司使用，成为更可信的竞争者。这些积压指标仍有可比性限制：Microsoft 的口径包含 Windows 与 Office 等更广泛产品的承诺，Google 的口径也纳入其生产力套件等；即便如此，当其走势与营收不同步时，它仍被视为偏领先的指标。
投资人面临的核心风险是承诺未必能如期转为现金与可认列收入：疫情期间企业曾与云端供应商重谈合约、把支出往后挪；而在当前环境中，单一大客户（常见为 OpenAI）的巨额交易就可能单独改变整体积压。Oracle 的积压自 2022 年末以来暴增 4,600 亿美元，其中几乎全部来自 OpenAI 承诺在未来 Oracle 资料中心租用容量的 3,000 亿美元；但 Oracle 股价此后下跌约四分之一（约 25%），市场质疑其获利性与是否能完整落地。更多大额承诺也在增加：OpenAI 表示在一年最后三个月将额外支出 2,500 亿美元于 Azure、以及 380 亿美元于 AWS；Anthropic 则承诺向 Azure 购买 300 亿美元容量；市场对这些承诺进入各公司积压后的反应，将成为衡量投资人是否担忧 AI 泡沫的重要温度计。
During earnings season, attention around Big Tech and cloud businesses often centers on growth rates, such as how fast sales are rising at Amazon Web Services (AWS), Microsoft’s Azure, and Google Cloud Platform. Another key metric that is less visible in press releases is the multiyear-deal backlog that will be earned over time: remaining performance obligations. This figure reflects shifting competitive dynamics as artificial intelligence companies such as OpenAI and Anthropic make enormous long-term commitments to the cloud providers that run their models, a shift that accelerated after ChatGPT’s late-2022 arrival.
AWS invented on-demand infrastructure selling and still holds the largest cloud market share, but the dollar value of commitments added to backlog since late 2022 has trailed Microsoft, the No. 2 cloud player, and sits slightly behind Google. Google, once seen as a cloud-market also-ran, has leveraged cutting-edge AI capabilities and demand for renting its custom AI chips to become a more credible rival. The backlog numbers have comparability caveats—Microsoft’s includes commitments tied to broad franchises like Windows and Office, and Google’s includes its productivity suite—but the metric is still treated as a leading indicator, especially when it moves differently than revenue.
The investor risk is that reported commitments may not convert into cash and recognizable revenue on schedule: during the pandemic, customers renegotiated to push cloud spending into future years, and today a single huge customer deal—often OpenAI—can swing totals. Oracle’s backlog has surged by $460 billion since late 2022, almost entirely driven by a $300 billion OpenAI commitment for future Oracle data-center capacity, yet Oracle shares have fallen by about a quarter (about 25%) as investors question profitability and whether the full deal will materialize. More large figures are arriving: OpenAI said it would spend an additional $250 billion on Azure and $38 billion on AWS in the last three months of the year, and Anthropic committed to $30 billion of Azure capacity; market reactions as these commitments enter reported backlog will signal sentiment about a potential AI bubble.