Skip to content

Microsoft FY2026 Q4 财报:市场认可的不是少花钱,而是算力已经能快速变现

报告日期:30 Jul 2026

报告期:截至 30 Jun 2026 的季度及财政年度

市场数据口径:29 Jul 2026 美股收盘及盘后交易

引用口径:财务数据默认单位为十亿美元。电话会完整英文问答的公开来源见文末官方电话会链接;本文按投资问题重组,并在每组原文后提供中文解释与评论。

结论

微软这份财报获得市场认可,核心原因不是资本开支下降,而是公司第一次同时给出了较完整的“投入—使用—收入”证据链:

需求继续超过可用产能
→ 新增或优化出的 CPU/GPU 产能可以在当季迅速售出
→ Azure Q4 增长 43%
→ FY2027 Q1 预计进一步加速至约 45%
→ Microsoft 365 Copilot 付费席位从上季度超过 20m 增至超过 30m
→ 商业 RPO 增至 678b,且本季度环比新增部分全部来自 frontier model companies 之外

这条证据链比“AI 会带来长期机会”更有价值。它说明微软当前的资本开支并不是纯粹建立远期选择权,而是至少有一部分产能正在被即时变现,并且 Azure、M365 Copilot、GitHub Copilot、Foundry 和 Agent 365 已经出现不同形式的收入或使用量增长。(FY2026 Q4 财报FY2026 Q4 电话会

财务结果也明显超过市场预期:

  • 收入 90.0b,同比增长 18%,高于 FactSet 预期的 87.62b。
  • non-GAAP EPS 为 4.74,同比增长 23%,高于市场预期约 4.24。
  • Azure and other cloud services 收入同比增长 43%。
  • 下一季度公司收入指引为 89.85b–90.95b,同比增长 16%–17%;Azure 按固定汇率预计增长约 45%。
  • 盘后较后时点股价约上涨 9% 至 426.03 美元;Yahoo Finance 抓取值为 425.01 美元,较 390.54 美元收盘价上涨约 8.8%。

数据来源:Microsoft FY2026 Q4 财报官方电话会及指引AP 引述的 FactSet 预期和盘后行情Yahoo Finance

但这不是一份没有风险的财报。

Q4 总 CapEx 为 41.0b,其中现金购置 PP&E 为 35.8b;自由现金流为 19.6b,同比下降约 23%。FY2027 Q1 CapEx 预计超过 50b,FY2027 全年 CapEx 仍将同比增长。Microsoft Cloud 毛利率为 65%,继续受到 Azure 收入占比上升、AI 基础设施和产品使用量增长的压力。(FY2026 Q4 电话会

因此,这份财报改变的是证明责任。管理层在 Azure Q&A 中明确表示,当前需求继续超过可用供给。市场因而愿意暂时接受更高的资本开支。接下来需要证明的,则是这种供不应求能否持续到足以覆盖 GPU/CPU 更新、折旧、租赁、能源和融资成本,并最终让自由现金流重新快于收入增长。完整英文问答见本文电话会部分及微软官方逐字稿

本季度发生了什么

财务摘要

指标 FY2026 Q4 FY2025 Q4 同比变化 观察
Revenue 90.01b 76.44b +18% 高于市场预期约 2.39b
Gross margin 60.48b 52.43b +15% 增速低于收入
Operating income 40.60b 34.32b +18% 与收入同速增长
Operating margin 45.1% 44.9% +0.2ppt 基本稳定
GAAP net income 35.77b 27.23b +31% 含投资收益
non-GAAP net income 35.29b 28.81b +22% 仅剔除 OpenAI 投资影响
GAAP diluted EPS 4.81 3.65 +32% 含 OpenAI 投资净收益
non-GAAP diluted EPS 4.74 3.86 +23% 仍含 Anthropic 等离散项目
Cash flow from operations 55.44b 42.65b +30% 云业务账单和回款强劲
Cash paid for PP&E 35.80b 17.08b +110% 现金基础设施投入翻倍以上
Free cash flow 19.64b 25.57b -23% 收入和利润增长尚未转化为 FCF
Total CapEx 41.0b 约 24.1b +70% 包含 5.6b finance leases

数据来源:FY2026 Q4 财报FY2026 Q4 电话会。市场预期来自 AP;总 CapEx 同比增幅来自 Axios 对官方披露的整理。Free cash flow 为经营现金流减现金购置 PP&E。

最重要的分化是:

Revenue +18%
Operating income +18%
Operating cash flow +30%


Cash paid for PP&E +110%
Free cash flow -23%

微软仍然拥有极强的经营现金创造能力,但 AI 基础设施正在比收入更快地吸收现金。

全年结果

指标 FY2026 FY2025 同比变化
Revenue 331.84b 281.72b +18%
Operating income 155.24b 128.53b +21%
GAAP net income 133.75b 101.83b +31%
non-GAAP net income 128.79b 105.45b +22%
Operating cash flow 182.94b 136.16b +34%
Cash paid for PP&E 115.95b 64.55b +80%
Free cash flow 66.99b 71.61b -6%
Cash、cash equivalents 及短期投资 76.84b 94.57b -19%
PP&E, net 313.08b 204.97b +53%

数据来源:FY2026 Q4 财报的全年损益表、资产负债表和现金流量表

全年收入增长 18%、运营利润增长 21%,说明利润表仍有经营杠杆;但自由现金流下降约 6%,说明现金流量表尚未进入收获期。

微软的资产负债表也正在发生结构性变化:

现金及短期投资减少约 17.7b
PP&E 净额增加约 108.1b

公司正把流动性转换为云和 AI 生产资产。这个变化本身不是负面,但意味着投资判断越来越依赖这些资产未来的利用率和经济寿命。

业务拆分

Intelligent Cloud:这是真正推动重估的业务

Intelligent Cloud 收入 39.31b,同比增长 32%;运营利润 15.96b,同比增长 31%,运营利润率约 40.6%,与去年基本持平。Azure and other cloud services 增长 43%。(FY2026 Q4 财报

更重要的是,管理层并没有把 Azure 超预期简单归因于市场需求,而是明确列出两类供给侧改善:

  • CPU 和 GPU fleet 的效率提升;
  • 新设备从到货到上线的流程缩短,使产能更早交付。

由于需求高于供给,这些改善释放的产能无需等待新的销售周期,而是在当季被客户消化。完整英文问答见本文电话会部分及微软官方逐字稿

这句话是本次财报最重要的投资信息。它把 CapEx 与收入之间的因果链缩短了:

过去的担忧:
投入更多 CapEx → 未来也许有需求

本季度给出的证据:
提升现有 fleet 效率或提前上线产能 → 当季即产生额外 Azure 收入

因此,43% 增长不只是需求强,也包含微软基础设施运营效率的改善。下一季度 Azure 固定汇率增长指引约 45%,说明管理层认为这一改善至少在短期仍可延续。(FY2026 Q4 电话会

需要保留的风险是,微软仍然没有披露 Azure AI 与非 AI Azure 的收入拆分。43% 是 Azure and other cloud services 的整体增速,不能全部归因于生成式 AI。

Microsoft 365:Copilot 从试点开始进入大规模部署

Productivity and Business Processes 收入 37.85b,同比增长 14%;运营利润 21.90b,同比增长 15%,运营利润率约 57.9%。(FY2026 Q4 财报

Microsoft 365 Commercial cloud:

  • reported revenue 增长 14%;
  • 调整去年一次性确认影响后增长 16%;
  • commercial paid seats 增长 6%;
  • M365 Copilot paid seats 超过 30m;
  • Copilot 净新增付费席位环比超过翻倍;
  • 超过 50,000 席位的客户数量同比增长超过 7 倍;
  • 将 Copilot 部署给多数知识员工的企业客户数量环比增长接近 75%;
  • E7 推出两个月,已有数百家企业购买数百万席位。

数据来源:FY2026 Q4 电话会准备发言

这批数据解决了两个此前不能区分的问题:

  1. 付费席位增加可能只是企业购买后尚未使用;
  2. 使用增长可能只发生在少数试点用户中。

管理层披露的补充指标——客户从部署到超过 80% 月活所需时间从数月缩短到数日、每用户对话数接近翻倍、周均参与度接近 Outlook 和 Teams——至少表明一部分大客户正在从试点转向广泛部署。(FY2026 Q4 电话会准备发言及 M365 Q&A

“seat + usage”是比 30m 席位更重要的信息

商业模式的核心问题是:如果 AI 提高单个员工的产出并减少传统席位,微软能否在 seat 之外建立 metered usage 收入,使 AI 越强反而带来更高 ARPU?

本季度管理层第一次把答案讲得非常直接:商业模式正在从单纯按席位收费转向席位加使用量。完整英文问答见本文电话会部分及微软官方逐字稿

具体变化包括:

  • Cowork 已加入 usage-based billing,已有数千家客户付费使用;
  • Dynamics 365 正从 seat 转向 seat plus consumption;
  • GitHub Copilot 引入 usage-based billing 后,出现 Business、Enterprise 席位增长和新增 consumption revenue;
  • GitHub Copilot revenue 环比加速超过 60%;
  • E7 将 Copilot、E5、Entra 和 Agent 365 打包,推动 premium ARPU;
  • 管理层预计 M365 Commercial cloud 收入增速将在 FY2027 内加速。

数据来源:FY2026 Q4 电话会

这为以下商业模式提供了证据:微软不是只把 AI 当作每用户每月固定价格的附加项,而是试图同时收取访问权和实际任务消耗费用。

它扩大了收入上限,但也带来新的执行风险:

  • 客户只有在任务确实产生 ROI 时才会持续增加 consumption;
  • usage 收入会比固定席位收入更波动;
  • 高使用量也会同步提高推理成本;
  • 如果模型价格快速下降,微软需要靠 workflow、数据、治理和分发能力维持毛利。

Foundry、Agent 365 和数据层:微软试图占据模型之上的价值

电话会披露:

  • Foundry 客户达到 100,000,收入同比超过翻倍;
  • 达到年化 1tn token 使用量的 Foundry 客户数量同比增长 4 倍;
  • Agent 365 推出两个月,已注册接近 40m agents;
  • Fabric 付费客户超过 40,000,同比增长超过 60%;
  • 同时使用 Foundry 和 Fabric 的客户超过 17,000,同比增长 60%;
  • PostgreSQL 收入增长 55%,连续第三个季度加速;
  • 使用多家模型供应商的客户数量自年初以来增长 5 倍。

数据来源:FY2026 Q4 电话会准备发言

这些指标不应直接相加为收入。注册 agent、Foundry 客户、Fabric 客户和 token run-rate 客户是不同口径,可能彼此重叠。

但它们共同支持一个方向:微软希望把模型变成可替换的底层投入,把持久记忆、企业上下文、权限、数据、治理、评估、工具调用和工作流留在 Azure/M365/GitHub/Dynamics 的平台层。完整英文问答见本文电话会部分及微软官方逐字稿

这既是微软降低 OpenAI 单一依赖的技术架构,也是商业策略:

如果模型不可替换:
价值容易集中到 frontier model provider

如果模型可替换:
客户可以按质量、成本、延迟和合规选择模型
微软则对算力、数据、上下文、治理和应用层持续收费

这个战略与微软的资产禀赋匹配,因为微软同时拥有 Azure 基础设施、企业身份与安全、M365 数据和工作入口、GitHub 开发者入口以及 Dynamics 业务流程。

但目前仍不能证明模型层会完全商品化。OpenAI、Anthropic、Google 或其他模型供应商仍可能通过更好的模型能力、原生应用或自有 agent 平台夺取更大份额。

More Personal Computing:财报中的明显弱项

More Personal Computing 收入 12.85b,同比下降 4%;运营利润 2.75b,同比下降 14%,运营利润率约 21.4%。(FY2026 Q4 财报

其中:

  • Windows OEM and Devices 收入下降 7%;
  • Windows OEM 下降 5%;
  • Xbox content and services 下降 10%;
  • Search advertising ex-TAC 增长 10%。

FY2027 的压力还会继续:

  • Windows OEM and Devices 全年收入预计下降 high-teens;
  • FY2027 Q1 预计下降 low-twenties;
  • Xbox content and services Q1 预计下降 mid-single digits;
  • PC 市场受到组件涨价、设备售价上升、去年 Windows 10 停止支持高基数和渠道库存影响。

数据来源:FY2026 Q4 电话会业绩和指引部分

市场没有因为这部分疲弱否定财报,是因为 Intelligent Cloud 与 M365 的增量远大于 MPC 的下降。但它提醒投资者,微软并不是所有业务都在受益于 AI。

EPS beat 需要打多少折扣

财报列出的 non-GAAP EPS 为 4.74,高于市场预期约 4.24,表面 beat 约 0.50 美元或 11.8%。(FY2026 Q4 财报AP

但 management 明确说明,相对于四月给出的内部指引,本季度有若干离散项目合计贡献 0.27 美元 EPS:

  • Anthropic 投资产生 3.2b gain;
  • Voluntary Retirement Program 费用低于预期;
  • 部分被 severance expense 和 Xbox impairment 抵消。

同时:

  • GAAP EPS 4.81 包含 OpenAI 投资带来的 0.07 美元净收益;
  • non-GAAP EPS 4.74 只剔除了 OpenAI 投资影响,并没有剔除 Anthropic gain 和其他离散项目。

数据来源:FY2026 Q4 财报及电话会

如果机械地从 4.74 中扣除 0.27:

观察口径 EPS ≈ 4.47

相对 4.24 市场预期:
仍高约 0.23,或约 5.4%

这不是公司公布的 non-GAAP 指标,也不能把所有离散项目简单视为非经营性;它只是用于说明本季度 EPS beat 的一部分来自投资和费用时点。

因此,更稳妥的判断是:

  • 收入 beat 和 Azure 加速是真实的;
  • operating income 增长 18%,与收入同速,也是真实的;
  • EPS beat 的幅度被离散项目放大;
  • 市场上涨更应该归因于 Azure、Copilot、RPO 和下一季度指引,而不是 4.81 的 headline EPS。

CapEx:175b 不是经济意义上的下调

这是本次财报最容易被误读的部分。

三个不同口径

口径 数字 含义
FY2026 Q4 total CapEx 41.0b 包含 finance leases
FY2026 Q4 cash paid for PP&E 35.8b 现金流量表中的资本支出
FY2026 Q4 finance leases 5.6b 主要是大型数据中心场地

数据来源:FY2026 Q4 电话会

总 CapEx 中:

  • 约三分之二是 short-lived assets,主要为 CPU 和 GPU;
  • 其余约三分之一为 long-lived assets;
  • FY2027 Q1 total CapEx 预计超过 50b;
  • FY2027 全年 CapEx 预计继续同比增长。

为什么 calendar 2026 从约 190b 变成约 175b

微软从 FY2027 开始将数据中心和办公楼预计使用寿命从 15 年延长至 25 年。这个变化使更多未来数据中心租赁从 finance lease 转为 operating lease:

  • finance leases 计入公司披露的 CapEx;
  • operating leases 不计入该 CapEx;
  • 经济上仍然要支付租金,现金会在未来通过经营现金流流出;
  • 管理层明确表示,剔除这一会计分类影响后,calendar 2026 的投资预期没有改变;
  • 新口径下约为 175b,而不是此前约 190b。

数据来源:FY2026 Q4 电话会会计口径说明

所以不能把 15b 的表面下降直接解释为:

需求减弱
微软减少了 15b 数据中心建设

更准确的解释是:

一部分未来数据中心使用权
从资产和融资租赁口径
转为经营租赁口径

这可能降低未来披露的 CapEx,却不一定改善经济成本或现金流。相反,operating lease payments 增加会压低经营现金流。

两种同时成立的解释

正面解释:

  • 两三分之二 CapEx 是交付周期较短的 CPU/GPU,若需求变化,可以减少后续采购;
  • 土地和数据中心 shell 的建设时点也可以调整;
  • 微软有 Azure、OpenAI、M365、GitHub、Dynamics 和其他第一方应用共同消化产能;
  • 当前新增产能可以迅速变现,降低闲置风险。

负面解释:

  • short-lived assets 意味着更快的技术更新和折旧压力;
  • 组件价格上升可能推高未来 COGS;
  • Q1 CapEx 超过 50b,说明现金开支仍在上升;
  • FY2027 只承诺 free cash flow positive,而没有承诺 FCF 增长或恢复到历史利润转换率;
  • 经营租赁重分类会使 headline CapEx 变小,却把现金压力分散到未来经营现金流。

因此,175b 并不是足以单独支撑看多的数字。市场真正认可的是,在这种投入强度下,Azure 和 Copilot 已经显示出较强的收入响应。

电话会:市场为什么从担心 CapEx 转向认可执行

以下按投资问题重组,而不是按发言顺序摘录。每节先保留分析师问题的完整逻辑,再解释管理层回答,最后给出本文评论。

1. Azure 为什么加速:需求更强,还是微软执行更好

完整原始问答

Brent Thill, Jefferies: Thanks. Amy, impressive acceleration in Azure up to 43 going to mid-40s. I guess the questions around the underlying drivers, what you and Satya are seeing in terms of just what’s driving this and many of the questions around capacity constraints, are we just still in the same environment or is this Microsoft just executing better, given the constraints we’re all seeing? Thanks.

Amy Hood: Thanks, Brent. First, there are still constraints in the system. I think we’ve continued to say, I think now, for a number of quarters, that demand continues to exceed available supply, and that certainly remains true. You can even see it, I think, in some of the pricing that’s occurring in the spot market for assets.

When you think about being able to deliver better, the first thing we focus on, and I tried to talk a little bit about it in my prepared remarks, is efficiency, being able to get more out of everything that we’ve got in the fleet. That applies to efficiency gains in the CPU fleet. It’s going to be efficiency gains in the GPU fleet.

We saw a good work this quarter, in particular, from our engineering teams to make as much of that available as we could. And because of the supply demand imbalance we’ve been talking about, when we can make efficiency gains, they are quickly monetized in quarter. And I think that dynamic certainly impacted the quarter positively.

I would also say some of the process improvements we’ve made to make sure both CPUs and GPUs, just the lead time from how quickly we can get things to simplify it tremendously plugged in, was also improved over the past 90 days. And so, those improvements, again, are very quickly monetized when we’re able to do that.

And at the scale that we’re operating in terms of across the entire hyperscale fleet, making efficiency improvements that can be quickly monetized does result in acceleration in the quarter. It’s part of also what we expect to see and talk about with Q1.

分析师问题

Jefferies 的 Brent Thill 指出,Azure 增速上升至 43%,下一季度又指向 mid-40s。他要求管理层区分两个因素:外部需求和供给约束是否仍与此前相同,还是微软自身在既有限制下执行得更好,从而推动了加速。

管理层回答

Amy Hood 表示,系统仍然受限,需求高于可用供给的状态没有改变,现货资产价格也能反映这种紧张。

本季度的增量主要来自两类执行改善。第一,工程团队提高现有 CPU 和 GPU fleet 的效率,使同一批资产可以提供更多可售算力。第二,微软缩短 CPU/GPU 到货、安装和上线的流程,使新产能更早进入生产。

因为供需不平衡仍然存在,这些效率和流程改善一旦释放产能,就能在本季度立即被客户使用。微软的 hyperscale fleet 足够大,因此即使单位效率只提高一点,也会形成可观察的收入增量。管理层预计这一机制也会支持 Q1。

评论

这个回答比单纯说“AI demand remains strong”更可信,因为它解释了 43% 如何形成:不是只有客户下单,也包括微软把物理资产更快转化为可计费服务。

它也解释了为什么市场愿意接受 Q1 超过 50b 的 CapEx。只要 marginal capacity 仍能快速售出,资本开支就更接近受需求约束的增长投入,而不是供给先行的投机建设。

风险在于这种逻辑高度依赖供不应求。如果 Azure 增速下降的同时,管理层不再强调 capacity constraint,市场会迅速重新计算闲置率、折旧和 ROIC。

原文位置:FY2026 Q4 电话会,Brent Thill 问答

2. 如果 AI 产能过剩,微软如何保护自己

完整原始问答

Mark Moerdler, Bernstein Research: Thank you very much for taking my question, and congratulations on the solid – it’s a really great quarter.

Satya, Amy, sentiment around AI remains incredibly volatile with concerns about oversupply coming, as well as concerns about component pricing increasing impacting margins.

Amy, two related questions: How does Microsoft protect itself if there really is overcapacity and overbuilding of data centers or overbuilding of chips, etcetera? And on the flip side of that, how do you manage through the hardware price increases that we’re seeing, the component pricing, and that it doesn’t just drive you to either massively drive up the price of your offerings or negatively impact your margins? Thank you.

Amy Hood: Thanks, Mark. The questions are a little bit related, but I’ll start with maybe the first.

Currently, the situation is obviously that demand exceeds available supply in a relatively extreme moment, but when you start to think about over the duration, I try to remind people, a lot of the expense, especially you see it in CapEx, you’ve seen our CapEx really pivot toward what I would call and do call short-lived assets, which really, that’s CPUs and GPUs that have relatively shorter lead times. And so, if the demand environment changes, you just slow down what is, in fact, the largest component and the driver of COGS.

The investment into land and data center builds is actually quite flexible. It’s a smaller percentage of the overall cost structure, and timing can be changed on much of that, especially on the builds, or you can stagger the timing of the build out of, as I was saying, some of the GPUs and CPUs that you plan to put in.

And so, when you think about being able to manage through that, hyperscalers have been doing that for quite a long time in terms of having the flexibility and the understanding of manage those changes in demand.

And the other thing is that’s important, Mark, is you just have an incredibly diverse book of business by geo, by segment, by industry. And I feel like when you look even at our backlog or what we added in RPO this quarter, it is from the breadth of really, the Microsoft product portfolio as well as our customer portfolio.

When you have the ability to late bind some of the more expensive components in short term, you have a big book of business that’s flexible. You have a big first-party app business that also uses the capacity that you’re building out in addition to your Azure platform. It does allow us to have a lot more flexibility to manage through those.

When it comes to the pricing question, I think that’s really impacting everybody equivalently in so many ways. What we’ve been trying to do, of course, at this point is to just make sure that we’re doing the best efficiency work we can so we can continue to give customers great value. We’re reminding people that frankly, the cloud offers tremendous benefits versus having to make these purchases as servers on-prem yourself, or the price increases are even more hard for customers. The cloud still provides a great ROI in those types of situations.

And we’re adding this capacity, to your point, but a lot of this obviously, is also being sold in newer contracts. And we’re able to have the pricing reflect it, but keep value where we – listen, for the long term, you want to have pricing work for customers and for you. And so, we’re trying to stay focused on that as well.

Satya Nadella: And if I just add to Amy’s comments, I thought Amy captured it well. All of us are reading this 1873 as the book to be read. And so, in my mind, I think you’ve got to get the product shape right. That’s a lot of what we are focused on. You have to get the portfolio right. Amy talked about how what we’re doing, whether it’s in Copilot or the Super App, bringing all the form factors or all the way to Azure, and the agent-first primitives in Azure. You have to really get that portfolio to all come together.

The mix of customers is super important. You have to recognize the breadth, the geo mix, the segment mix, the workload mix. And you’ve got to really think about all of those when you’re even building capacity. And then you’ve got to run an efficient railroad. At the end of the day, Amy talked a little bit about, even in the last quarter, how we’ve improved on the efficiency front. It’s not something that will just show up at the end. You have to monotonically work at it.

And so, we are very focused on all those. And then we know that there will be ups and downs of what is the cycle here, but the secular shift is clear. And we’re very bullish about us coming up with the right mix of business and the right margin structure, and most importantly, with the right value for our customers.

分析师问题

Bernstein 的 Mark Moerdler 将当前 AI 投资争议的两端放在一起:一方面,市场担心数据中心和芯片最终供给过剩;另一方面,组件价格上涨又可能压低毛利或迫使微软提高产品价格。他要求管理层说明如何同时应对需求回落和成本上升。

管理层回答

Amy Hood 的回答分为四层。

第一,当前仍处于需求明显高于供给的阶段,但最大 CapEx 组成已转向交付周期较短的 CPU 和 GPU。如果需求发生变化,公司可以放慢这部分采购,而不必继续完成原有速度的扩张。

第二,土地和数据中心建设只占总成本的一部分,建设时点和在场地中安装 CPU/GPU 的节奏可以错开,因此 long-lived infrastructure 并非完全刚性。

第三,微软的客户和工作负载足够分散,覆盖不同地域、行业和客户类型;Azure 之外还有大量第一方应用可以使用同一基础设施。昂贵组件可以较晚决定具体分配给哪个 workload。

第四,组件涨价可以通过 fleet 效率、云相对本地服务器的经济性以及新合同定价部分消化,但公司仍需在自身毛利和客户 ROI 之间保持平衡。

Satya 补充,管理这轮周期不能只看单一产品,需要同时优化产品形态、业务组合、客户和地域组合、workload mix 以及 hyperscale 运营效率。

评论

这个回答说明微软的防御不是“AI 不会过剩”,而是“即使过剩,公司也有更灵活的采购节奏和更多内部消化渠道”。

这是合理但尚未被下行周期验证的主张。两三分之二 short-lived assets 既是灵活性,也是风险:可以停止买下一批,但已经买入的 GPU 仍会折旧;技术更新越快,旧资产的经济寿命可能越短。

原文位置:FY2026 Q4 电话会,Mark Moerdler 问答

3. Copilot 是从试点扩张,还是只是更多试点

完整原始问答

Adam Wood, Morgan Stanley: Hi, good evening. Thanks for taking the question. And also, congrats on a very strong end of the year.

I wanted to maybe just ask about M365 Copilot. Obviously, very strong quarter there with over 30 million paid seats and a strong acceleration. Could you just talk a little bit about how you’re seeing customers move from pilots to broader deployments here? Is this still a pilot-driven motion or are we seeing a lot more broader deployments?

And then when we think about the monetization of the product in terms of additional seats, migration to higher value SKUs like E7, and then consumption, what do you see is the main monetization or the main driver of monetization from here, please? Thank you.

Satya Nadella: No, thank you, Adam, for that question. Let me start and then Amy can add.

I think, yeah, it starts with, again, that product shape. As you can see, even within the quarter, the product shape has changed pretty dramatically. We now have chat, Cowork, autopilot, code all coming to essentially, what is going to become this flagship Super App that various roles can use it.

And if you think about even the usage side, that’s the place where, again, lots of interesting data there, which is time to usage has drastically come down. What used to be months is days from when a license is bought to usage. The usage intensity itself has gone up significantly. I mean, we’re talking about a usage intensity that’s at the same level of what is an everyday communication tool like Outlook or Teams.

The second thing I’d say is the overall enterprise wiring of this, it’s not like a tool that’s isolated somewhere, but it’s wired in whether it is – you brought up E7. It’s wired into the governance pieces with Agent 365 so that you have your IT Ops, SecOps, FinOps all wired in, as well as it’s all the business processes.

For example, your CRM system, your ERP system, all of them are just skills and plug-ins that go into core work. You’re able to take that enterprise-wide workflow and wire it into the Super App. That increases usage so it all compounds.

And then the other one is the business model. We now have this perceived business model. And so, we also now have the usage business model. It’s seat plus usage. We’re already seeing the ARPU growth that comes from things like E7, but really as we deliver more value to customer and customer outcomes at the enterprise level.

In fact, if I think about historically, Office compared to what Copilot is, is much more narrower. This is the first time where you really have an enterprise-wide tool, which has a both per-seat and usage-based pricing. The TAM is much more expansive. We’re going to be very, very focused on driving customer value and then expanding with it.

Amy Hood: Yeah, Adam, and I think I talked a little bit about it in my prepared remarks, but I do think what we’ve been seeing is over the course of this year, some of the growth in ARPU was from E5 plus the Copilot license that Satya is talking about.

We’ll see a little bit more from E7 really has a lot of interesting value in the Agent 365 component, in particular, where Satya is talking about, I mean, having SecOps and FinOps, think about in general, everyone is going to need both observability of token spend and the manageability of token spend for all business processes. And that is what E7 brings.

And so, I think it was only in market for a part of the quarter, and I think we were quite encouraged by the value customers saw in that SKU. I think we’ll continue to focus on that through the year.

And then finally, what Satya is talking about is this building TAM that grows through the year. And as I think about that expansive, expanding TAM, that’s really where we’re talking about this usage and consumption growth. And so, as more of those experiences get wired in and as IT gets more involved in that process, it’ll be quite, I think, changing in terms of what people think of the M365 capabilities.

Satya Nadella: It would be fun for you, Adam. I think one of your colleagues put out an ROIC document. I took that document to Copilot, which is a PDF, and I said, “Build me a new Power BI dashboard, essentially.” But here is the thing. It built a rich semantic model that went into my Fabric with OneLake that brought all the data in from the external sources. In fact, it was current with all the SEC filings of all the MAG7. And then on top of that, the repo itself is in GitHub, but the artifact is sitting in my Copilot as a site.

That, to me, is a classic example of an enterprise-wide workflow. I, as a knowledge worker, could go create a dashboard. The data engineer can go to Fabric and find the artifact. The professional developer can go to the repo and find it in GitHub. And by the way, it’s all registered with Agent 365. That’s a little bit of what Amy is describing as the coming together of a new way to work, even while at the same time, bringing IT, security and manageability of it.

分析师问题

Morgan Stanley 的 Adam Wood 看到 M365 Copilot 超过 30m 付费席位和明显加速后,追问客户是否真正从 pilot 进入 broad deployment;并要求管理层在新增席位、E7 premium SKU 和 consumption 三种变现方式中指出未来主要驱动力。

管理层回答

Satya 先从产品形态解释使用增长:Copilot 正把 chat、Cowork、Autopilot 和 Code 合并为跨角色的旗舰入口。客户从购买 license 到形成高使用量的时间已经从数月缩短至数日,使用强度接近日常通信工具。

第二层是 enterprise wiring。Copilot 不只是独立聊天工具,而是与 Agent 365 的 IT、security、finance governance,以及 CRM、ERP、Fabric、GitHub 等业务和数据系统连接。更多 workflow 进入同一入口后,使用量可以互相强化。

第三层是商业模式。微软同时使用 per-seat 和 usage-based pricing;E7 提高 ARPU,Agent 365 提供 token spend 的可观察性和管理能力,Cowork 等任务则带来 consumption。

Amy Hood 补充,过去一年的 ARPU 增长首先来自 E5 加 Copilot,未来 E7 与 Agent 365 会增加价值,而更广泛的 workflow 使用会逐步扩大 consumption TAM。

评论

管理层没有提供 Copilot 单独收入和毛利率,因此仍不能精确估算 30m 席位的收入贡献。

但回答给出了比席位数更好的部署指标:高使用率形成时间、对话强度、大客户部署规模、部署给多数知识员工的企业数量,以及 usage billing 的实际客户数。这些指标共同降低了“30m 只是未使用 shelfware”的概率。

从商业模式看,最重要的变化是微软明确确认 seat plus usage,而不是 30m 本身。固定席位提供可预测收入,consumption 则让微软分享 AI 完成更多任务所创造的价值。

原文位置:FY2026 Q4 电话会,Adam Wood 问答

4. 多模型战略如何让微软获利,而不是削弱 OpenAI 关系

完整原始问答

Karl Keirstead, UBS: Okay, great. Thank you, Satya. Maybe I’ll start away from the numbers and ask if you could spend a minute and elaborate on your opening comments about model choice and the protection of corporate IP. Maybe I could ask this in two parts.

First, how material do you think traction could be for open and custom models over the next year or two, knowing that many enterprises might be initially reticent to use open models?

And secondly, how exactly does Microsoft benefit from this shift, knowing that you’ve also got fairly large frontier lab exposure? Thanks so much.

Satya Nadella: Thank you, Karl. The way we are coming at this is at the end of the day, the goal is to have the firm be in control of their own destiny, in terms of what I describe as building their human capital and their token capital. At the end of the day, if a firm is a learning machine, they need their own learning machine, and that’s really the goal. And the models are an input, not some extraction of the knowledge of the enterprise.

But in some sense you have to really – at the end of the day, every firm is going to evaluate who are the providers who are helping them with their outcomes and their knowledge creation. I think that that is now fairly clear, and it’s going to become clearer by the day. This is not going to be about, come in and take all my knowledge and benefit yourself, whereas I am not getting anything out of it.

Given that direction of travel, we are very, very clear about the architectural design of the platform, which is you’ve got to keep your harness separate from the model, when the harness will ensure that your memory, your context all of that is external. That means any given model at any given time is swappable. You should and you can use frontier models. There’s no reason not to, but you also can use multiple of them.

If you look at some of the stats I gave, it’s a great example of how to use the frontier models for what they deliver, how to use low-cost models for what they deliver, and in fact, train your own model when you don’t want to use any external model itself, because after all, you have all the outputs, you have all the traces, you have all the context.

That’s really the enterprise design architecture that we are going to evangelize. We ourselves are using it. Copilot is built that way. GitHub Copilot is built that way. Our Security Copilot is built that way. And we want to democratize that design pattern so that every enterprise can use it. And within there, there will be a mix of open weights, closed weights.

And by the way, one of the things that’s least talked about is remember, if you look even at the Hugging Face incident, the biggest thing that you should take away from that is you can’t depend on any one model. You will maybe need multiple models to even remediate some challenges that get caused by one model. That’s the way to think about it, which is you can’t be subject to the refusals of one model.

There’s a lot more design space here. We talk about the frontier as if it’s one thing. The frontier is about every firm having a frontier, and the choice, the cost control and the capability that they need in order to be able to control their destiny.

Amy Hood: And I think maybe, Karl, just to add a little bit to the end of your question, which is that it’s why it’s important that the platform is built, and I think Satya mentioned this in his comments, to be able to deliver the right model for the right job on the architecture called Azure.

And so, given that we continue to see growing demand no matter what model is chosen or what model family or whether it’s run a model of your own, the Azure platform is quite efficient at delivering that. Think about that infrastructure as being pretty fungible.

分析师问题

UBS 的 Karl Keirstead 追问两点:企业未来一两年是否会真正采用 open 或 custom models;以及当微软本身对 frontier labs 有大量投资和业务敞口时,模型选择增加究竟如何让微软获益。

管理层回答

Satya 将企业 AI 定义为企业自己的“学习系统”。模型只是输入,企业不应把内部知识、工作轨迹和持续学习能力永久交给单一外部模型供应商。

因此微软的架构把 harness、memory、context 和 action space 放在模型之外。企业可以在不同任务中混用 frontier、低成本、open-weight 和自训练模型,并在质量、成本、延迟、合规或供应变化时替换底层模型。

Copilot、GitHub Copilot 和 Security Copilot 已经采用这种架构,微软希望通过 Foundry 将它推广给其他企业。Amy Hood 补充,无论客户选择哪种模型,工作负载仍可在 Azure 上运行;Azure 基础设施因此具有较强的可替代性和复用性。

评论

这套战略将微软与 OpenAI 的关系重新定义为“重要模型和客户,但不是唯一模型或唯一价值层”。

如果成立,微软可以从三处获得收益:

  • 不同模型都需要 Azure 计算和网络;
  • 企业 memory、context、data、identity 和 governance 留在微软平台;
  • Copilot、GitHub、Dynamics 等应用把模型输出转化为可计费 workflow。

它也降低了单一模型供应商提价、拒绝服务或技术落后的风险。

但需要观察客户是否真的愿意把 agent harness 和治理层标准化在 Foundry/Agent 365 上。若模型厂商自己的应用和 agent 平台更快形成网络效应,微软的“模型可替换、平台留存价值”假设可能被削弱。

原文位置:FY2026 Q4 电话会,Karl Keirstead 问答

5. 管理层如何衡量 CapEx ROIC

完整原始问答

Gabriela Borges, Goldman Sachs: Hey, good afternoon. Thank you.

Amy, I wanted to ask you about ROI. You’ve given us color on the CapEx side of the equation. You’ve given us color on the monetization side of the equation. Maybe put those two pieces together for us.

When you look at and track ROI on the CapEx decisions you’re making today, how does that compare to a year ago, and what are some of the levers that you can still pull, perhaps from the internal silicon side, for example, as a driver of incremental monetization going forward? Thank you.

Amy Hood: Thanks, Gabriela.

I don’t know that, quite frankly, my math has changed in terms of how I do it over the past year. I would say the way to think about it for me is more the confidence in the TAM expansion, the margin levers that we have in terms of both product improvements than the infrastructure improvements.

We talked about some already on the call today in terms of the levers we have to continue to get efficiencies across both the application part of the stack and then the infra part of the stack. But you’re right, we didn’t touch on all of the pieces. I think Satya actually commented on a number of them.

We still have opportunities, obviously, as we continue to look for the best price performance on silicon, including our investments in first party. The work, frankly, on model diversification also is a margin improvement opportunity. Being able to serve the best possible outcome with a more efficient, or both efficient in terms of token usage and efficient in terms of cost structure are also margin levers. All of these things contribute, obviously, to your point of increased confidence in ROIC, frankly, of the dollars that we’re investing and continue to invest going forward.

As we think about the mix of the portfolio being able to have a pretty broad pool across knowledge work, coding, security, then basically the agent layer, I’ll call that Agent 365 as kind of a cheat, but all of that also is an opportunity, and then of course what we talked about on the Azure side between model efficiency, silicon, and component efficiency, including our investments in 1P solutions there, and just the overall efficiency of running it at a hyperscale.

So we have quite a few levers to continue to see improvement that we’re focused on, but as Satya mentioned, this is the grind work. This is like every day, you just get a little better, get a little better. We actually are quite good at that grind and making sure that we can deliver that for customers.

分析师问题

Goldman Sachs 的 Gabriela Borges 要求 Amy Hood 把 CapEx 和变现放在同一框架中:今天的资本投入回报与一年前相比是否改变,以及自研芯片等工具还能如何提高未来回报。

管理层回答

Amy 没有给出具体 hurdle rate、payback period 或当前 AI ROIC。她表示,计算方法本身与一年前没有明显变化,变化的是对 TAM 扩张和可用 margin levers 的信心。

这些 margin levers 包括:

  • 应用层和基础设施层效率;
  • 第一方芯片的 price-performance;
  • 多模型路由带来的模型成本优化;
  • 减少完成同一 outcome 所需的 token;
  • M365、coding、security、Agent 365 和 Azure 之间的 workload 组合;
  • hyperscale 运营效率。

她认为这些因素提高了对投入资金 ROIC 的信心,但也承认改善来自持续、渐进的运营工作,而不是单一突破。

评论

这是电话会中仍未被回答的问题。

管理层给出了提高 ROIC 的工具,却没有给出:

  • 当前新增 GPU 的收入或利润回收期;
  • AI workload 的折旧前和折旧后毛利;
  • Copilot、Foundry 或 GitHub Copilot 的 contribution margin;
  • 190b/175b 级投资对应的中期收入和 FCF 目标;
  • 在何种利用率或需求情景下会削减采购。

市场本季度愿意接受定性回答,是因为 Azure 43%、Q1 约 45%、Copilot 席位和使用量共同提供了间接证据。如果这些运营指标减速,缺少直接 ROIC 披露会再次成为主要估值折价。

原文位置:FY2026 Q4 电话会,Gabriela Borges 问答

为什么市场认可这份财报

1. Azure 增长不是只维持,而是继续加速

市场在财报前最关注的不是微软能否保持两位数增长,而是高 CapEx 是否能带来足够的 Azure 增量。

Q4 Azure 增长 43%,下一季度指引约 45%,直接回答了这个问题。管理层还预计 FY2027 上半年 Azure 增速继续加速。(FY2026 Q4 电话会

2. CapEx 与收入之间出现了当季连接

微软说明 fleet 效率提高和产能提前上线后,额外供给在同一季度就被使用。这比总量 RPO 或长期 TAM 更接近资本回报证据。

3. Copilot 同时出现席位、使用和商业模式进展

30m 付费席位只是第一层。更有价值的是:

  • 净新增付费席位环比超过翻倍;
  • 大规模部署客户显著增加;
  • 使用强度和部署速度改善;
  • E7 带来 premium ARPU;
  • Cowork、Dynamics 和 GitHub 开始叠加 consumption 收入。

这些变化使 AI 应用收入不再只依赖固定的 30 美元席位。

4. RPO 的质量好于 84% headline

Commercial RPO 增长 84% 至 678b,但其中包含 OpenAI 的大型长期合同。更保守的口径是:

  • 剔除 OpenAI 后 RPO 仍增长 25%;
  • 本季度所有环比新增 commercial RPO 均来自 frontier model companies 之外;
  • Microsoft Cloud 全年收入中接近 90% 来自 frontier model companies 之外;
  • 约 30% RPO 预计在未来 12 个月确认收入,该部分同比增长 37%。

数据来源:FY2026 Q4 电话会

这降低了市场对增长完全依赖 OpenAI 的担忧。

5. FY2027 仍保持利润纪律

微软预计:

  • FY2027 收入和运营利润继续双位数增长;
  • operating expenses 增长 mid-to-high single digits;
  • 全年 operating margin 下降少于 1ppt;
  • FY2027 仍为 free cash flow positive;
  • Q1 公司收入增长 16%–17%,operating margin 同比基本稳定。

数据来源:FY2026 Q4 电话会指引

这并不等于 FCF 增长,但至少说明公司预计在 CapEx 上升时仍能维持利润表纪律。

市场可能高兴得太早的地方

1. 自由现金流仍在恶化

Q4 FCF margin 从去年同期约 33.4% 降至约 21.8%;全年 FCF margin 从约 25.4% 降至约 20.2%。

如果未来收入和 operating income 继续双位数增长,但 FCF 长期停滞,投资者最终仍需下调对资本回报和估值倍数的假设。

2. Cloud 毛利率没有改善

Microsoft Cloud gross margin 为 65%,同比下降,原因包括:

  • 收入组合转向 Azure;
  • AI infrastructure 持续扩张;
  • AI 产品使用量增加;
  • 部分被平台效率改善抵消。

Productivity and Business Processes 毛利率也因 Copilot 使用量增长略有下降;Intelligent Cloud 毛利率受到 Azure 和 AI infrastructure 影响。(FY2026 Q4 电话会

这说明 AI 使用增长已经产生收入,也确实产生更高 COGS。

3. 41b CapEx 中约三分之二需要更快更新

CPU/GPU 的短寿命带来采购灵活性,但也意味着:

  • 折旧更快;
  • 新一代芯片可能使旧资产经济性下降;
  • 组件价格上涨会更快进入 COGS;
  • 产能利用率下降时,利润表承压速度更快。

4. 175b 的会计口径可能改善 headline,而不是经济性

数据中心租赁从 finance lease 转向 operating lease,会降低披露 CapEx,却不会消除使用资产的经济成本。投资者需要同时观察:

  • total CapEx;
  • cash paid for PP&E;
  • finance lease additions;
  • operating lease payments;
  • depreciation;
  • operating cash flow 和 FCF。

只看单一 CapEx headline 会低估真实投入。

5. RPO 很大,但并非全部是近期收入

678b RPO 的加权平均期限为 2.3 年,只有约 30% 预计在未来 12 个月确认。其余部分持续时间更长,并受到客户实际消耗、合同结构和 OpenAI 大额承诺影响。(FY2026 Q4 电话会

6. non-GAAP EPS 仍含较大投资收益

本季度 4.74 美元 non-GAAP EPS 只剔除 OpenAI 投资影响,仍包含 Anthropic gain 等离散项目。不能把全部 EPS beat 当作核心业务盈利能力的改善。

7. 传统 PC 和游戏业务继续收缩

Windows OEM、Devices 和 Xbox 均下降,且 FY2027 指引显示 PC 压力尚未结束。Azure 和 M365 足以抵消这些问题,但它们会减少公司整体增长的安全垫。

企业 AI 商业模式:哪些问题得到回答

理解企业 AI 商业模式,需要回答三个问题:

  1. 企业 AI workflow 的需求是否足够普遍;
  2. 客户是否愿意在 seat 之外按 usage 或 outcome 付费;
  3. 微软是否能通过 Azure、M365、GitHub、Dynamics、身份、安全和治理捕获价值。

得到验证的部分

原有观察 FY2026 Q4 新证据 判断
Copilot paid seats 需从 20m 快速向上爬升 超过 30m,净新增席位环比超过翻倍 明显验证
固定 seat 不足以支撑全部 AI TAM Cowork、Dynamics、GitHub 明确加入 usage billing 明显验证
企业 workflow 是微软的主要价值捕获点 Copilot 与 Agent 365、Fabric、GitHub、CRM、ERP 连接 方向验证
CapEx 只有在 capacity constrained 时更容易成立 需求仍高于供给,新增产能当季变现 短期验证
需要观察 Azure、Copilot 和 Cloud margin Azure 与 Copilot 强,但 Cloud margin 降至 65% 收入侧验证、利润侧未验证
模型成本下降可能扩大 workflow 经济性 多模型路由、自研模型和芯片带来多项单位成本改善 初步验证

尚未得到验证的部分

  • Copilot 是否能从 30m 进一步达到 50m、100m 或更高;
  • seat plus usage 的实际 ARPU、收入和 contribution margin;
  • Azure AI 单独的收入、增长和毛利;
  • Agent 365 的 40m 注册 agents 中有多少持续使用并付费;
  • 175b–190b 投资规模对应的中期 ROIC;
  • FCF 何时恢复增长;
  • GPU/CPU 折旧进入稳态后,利润增长是否仍能覆盖。

需要修正的地方

以此前 calendar 2026 CapEx 约 190b 作为经济投入基准,本季度新口径约 175b,但这主要来自 lease classification,而不是实际投资计划下降。

因此,不能简单把折旧模型中的 190b 全部改为 175b。更合理的做法是:

资产购买和 finance lease
→ 进入 PP&E、折旧和 CapEx

operating lease
→ 不进入 headline CapEx
→ 但通过未来租赁费用和经营现金流体现

未来模型应同时计算资产折旧和 operating lease obligations,而不是只对披露 CapEx 乘折旧率。

多空框架

Bull case

  • Azure 在 FY2027 上半年保持 mid-40s 或更高增长;
  • capacity constraint 延续,新增算力仍可快速变现;
  • M365 Copilot paid seats 在未来四个季度向 50m–60m 推进;
  • E7、Cowork、Dynamics 和 GitHub consumption 推动 M365 Commercial cloud 增速加速;
  • Cloud gross margin 稳定或回升;
  • FY2027 operating margin 下降少于 1ppt,FCF 在下半年恢复增长;
  • RPO 增长更多来自广泛企业客户,而不是 frontier labs;
  • 多模型和第一方芯片持续降低单位 outcome 成本。

在这个情景中,微软证明自己不仅是 AI 基础设施提供商,也是企业 agent workflow 的分发、治理和计费平台。高 CapEx 可以被更大的收入池和长期平台毛利覆盖。

Base case

  • Azure 保持约 40%–45% 增长后逐步放缓;
  • Copilot paid seats 稳步增长,但 usage revenue 仍小;
  • CapEx 继续上升,Cloud gross margin 小幅下降;
  • operating income 维持双位数增长,FCF 低速增长或大致持平;
  • Windows 和 Xbox 拖累由云与 M365 抵消。

在这个情景中,微软仍是优质增长公司,但估值上行取决于市场愿意给多长时间等待 FCF 转化。

Bear case

  • Azure 增速明显低于指引,同时不再供不应求;
  • Copilot 席位增长来自折扣或 bundle,实际 usage 和 ARPU 不增长;
  • usage 收入增加,但推理成本增长更快;
  • Cloud gross margin持续下降;
  • Q1 超过 50b CapEx 后,FY2027 FCF 继续下降;
  • operating lease 重分类掩盖真实基础设施承诺;
  • 大客户推迟消费,RPO 无法按预期转化为收入;
  • 模型供应商在应用和 agent 平台层绕过微软。

在这个情景中,微软会从高 FCF 软件平台向低自由现金流、持续更新硬件的重资产云平台进一步迁移,估值倍数应下降。

接下来四个季度最需要观察什么

指标 当前基线 正面信号 负面信号
Azure growth Q4 +43%,Q1 guide 约 +45% CC H1 继续加速或维持 mid-40s 低于 40% 且供给约束缓解
M365 Copilot paid seats >30m 继续向 50m 快速推进 净新增显著放缓
Copilot usage 对话数/用户接近翻倍 使用强度、80% MAU 客户继续提高 席位增长但 usage 停滞
Seat + usage revenue 已在 Cowork、Dynamics、GitHub 推出 披露具体收入或 ARPU 只有产品发布,没有收入证据
Commercial RPO ex-OpenAI +25% 保持 20% 以上且更广泛 headline 高、剔除 OpenAI 后明显减速
Microsoft Cloud GM 65% 稳定或回升 连续下降且效率无法抵消
Total CapEx Q4 41b;Q1 >50b Azure/AI 收入同步加速 CapEx 上升而增长放缓
Cash PP&E Q4 35.8b OCF 增长更快 持续快于 OCF 增长
FCF Q4 19.6b,-23% FY2027 恢复增长 继续下降
Operating lease payments 已开始压低 OCF 透明披露、总体投入稳定 headline CapEx 降但租赁现金流激增
Foundry/Agent 365 100k customers / 40m registered agents 出现付费、usage、retention 指标 注册量大但无变现
AI ROIC disclosure 仍为定性 披露回收期、毛利或利用率 继续只谈 TAM 和效率工具

可证伪条件

当前偏正面的判断建立在“高 CapEx 正在被高需求和快速变现支持”之上。

以下情况出现两项以上,应明显降低判断:

  1. Azure 增速连续两个季度低于 40%,且管理层不再强调需求高于供给;
  2. M365 Copilot paid seats 增长明显放缓,或不再披露;
  3. 大规模部署增加,但每用户使用量和 consumption revenue 不增长;
  4. Microsoft Cloud gross margin 连续下降超过预期;
  5. FY2027 CapEx 上升,但 operating cash flow 和 FCF 同时下降;
  6. 剔除 OpenAI 后的 bookings 或 RPO 增长降至低双位数;
  7. operating lease payments 大幅增长,使 175b 新口径失去可比性;
  8. 微软无法说明第一方芯片、多模型路由和 fleet 效率如何改善单位经济;
  9. AI 产品收入增长仍主要依赖投资收益、bundle 或价格上涨,而非真实 usage;
  10. FY2027 operating margin 下降超过管理层所说的 1ppt。

最终判断

微软与 Meta 在同一天展示了两种不同的 AI 投资回报叙事。

两家公司都在大幅增加基础设施投入,也都面临折旧、算力、人才和自由现金流压力。但微软给出了更接近可验证收入的证据:

Azure 43%
下一季度约 45%
新增算力当季变现
M365 Copilot >30m paid seats
seat + usage
RPO ex-OpenAI +25%
新增 RPO 来自更广泛客户

因此,市场认可微软并不是因为它认为 41b CapEx 不重要,而是因为微软暂时证明了:

花出去的钱
正在更快地变成可售算力、付费席位、使用量和合同负债

这份财报使微软 AI thesis 从“需求很可能存在”前进到“部分需求已经规模化并可以变现”。

但它还没有前进到最后一步:

规模化收入
是否最终能够产生高于资本成本的自由现金流回报

当前判断可以偏正面,但未来不能只看 Azure 和 Copilot 增速。真正决定估值的组合仍然是:

Azure / Copilot / usage revenue 增长
减去
折旧 / 租赁 / 推理成本 / CapEx
最终形成的 FCF

市场这次选择先相信收入和执行,下一阶段则需要现金流来完成证明。

Comments