行业研究公司研究宏观策略财报招股书会议纪要 seedance2.0 低空经济 DeepSeek AIGC 大模型

本季度云服务：超大规模云服务商在2025年第四季度表现如何？

信息技术 2026-04-14 伯恩斯坦 Andy Yang 杨敏

云计算市场概况

超大规模市场已成为支持生成式 AI 浪潮的“挖金热”，尽管这些企业主要以 CPU 驱动为主。股票价格也受到 AI 情绪的驱动。随着 AI 成为收入中越来越重要的贡献者，投资者仍然关注 CAPEX 增长以及相关利润率/回报率。

主要发现

亚马逊 AWS：AWS 表现强劲，4 季度同比增长 24%，环比新增约 26 亿美元净云收入，创历史新高。AWS 的 AI 收入跑率在 2026 年第一季度超过 150 亿美元，半导体内部芯片业务（以 Graviton 和 Trainium 为首）现在年营收超过 200 亿美元，同比增长三位数。AWS 的 AI 产品收入连续第十个季度实现三位数增长，对外部云收入的贡献逐季增加。Bedrock 现在已成为一个价值数十亿美元的年营收业务，客户支出环比增长 60%。AWS 正在积极扩大容量，2025 年增加的数据中心容量比任何其他公司都多。展望未来，非 AI 工作负载的持续增长、AI 需求的增长以及容量的快速变现应该能够帮助 AWS 争取到越来越多的企业 IT 支出，同时保持健康低至中等的 30% 左右的运营利润率。
微软 Azure：Azure 本季度增长 38%（报告为 39%），超过预期。尽管略微低于买方预期，但管理层表示，如果将所有新容量分配给 Azure，Azure 的增长将超过 40%。管理层认为，将资源优先分配给第一方应用程序和研发，而不是短期 Azure 增长，是公司长期价值创造的正确做法。Azure 的客户仍然是企业，这与亚马逊和谷歌形成了一些差异。此外，微软并没有将 LLM 模型训练作为一项业务来追求，因此对任何 AI 空气袋或弱点更加免疫。
谷歌云：谷歌云收入同比增长 48%，达到约 700 亿美元年营收，远超市场预期，得益于 2400 亿美元的积压订单和跨核心 GCP（尤其是在基础设施、网络安全和数据分析服务方面）以及快速增长的 AI 采用的广泛优势。近 75% 的云客户现在使用谷歌的垂直集成 AI 堆栈，Gemini Enterprise 在短短四个月内就达到了 800 万个付费席位，同时新平台 AntiGravity 也显示出强劲的早期采用率。云运营利润率在 4 季度大幅增长至 30%（环比增长 640 个基点），反映了价格上涨、强劲的收入表现以及公司对运营效率的关注。
阿里巴巴云：阿里巴巴云收入同比增长 36.4%，达到 433 亿人民币，调整后 EBITA 利润率保持在 9.0%，与上一季度相同。外部收入同比增长约 35%，快于 9 月份的 29%。AI 相关云产品收入连续第十个季度实现三位数增长，对外部云收入的贡献逐季增加。阿里巴巴的管理层指出，3 月份季度同比增长超过 40%，预计将继续加速。展望未来，阿里巴巴云的五年目标是外部云和 AI 收入达到 1000 亿元人民币，这意味着年复合增长率约为 40%……尽管没有提及所需的 CAPEX。
甲骨文 OCI：甲骨文 OCI 继续比其四大竞争对手增长更快，并且仍在加速。3 季度 OCI 收入达到 49 亿美元，同比增长 84%（合并报表为 81%）。总 RPO 增加 290 亿美元至 5530 亿美元，同比增长 325%，主要受大规模 AI 合同的推动。cRPO 增长 64%，约占 RPO 的 12%。甲骨文 OCI 的增长速度有可能在 2029-2030 年使 OCI 达到或超过 GCP 的水平。

关键数据

AWS：4 季度同比增长 24%，环比新增约 26 亿美元净云收入，创历史新高。AWS 积压订单达到 2440 亿美元，同比增长 22%，环比增长 40%。
Azure：2FQ26 同比增长 38%（合并报表为 39%），超过预期。Azure 积压订单达到 1420 亿美元年营收，同比增长 110%，环比增长 2.5 个基点。
谷歌云：4 季度收入同比增长 48%，达到约 700 亿美元年营收。谷歌云积压订单达到 2400 亿美元，同比增长 55%，环比增长 20%。
阿里巴巴云：12 月份季度收入同比增长 36.4%，达到 433 亿人民币。调整后 EBITA 利润率保持在 9.0%，与上一季度相同。
甲骨文 OCI：3 季度收入达到 49 亿美元，同比增长 84%（合并报表为 81%）。总 RPO 增加 290 亿美元至 5530 亿美元，同比增长 325%。

研究结论

AI 将在建设阶段至少对 IaaS/PaaS 的毛利率产生负面影响，甚至可能更长时间。
供应/需求问题以及 GPU 的实际使用寿命是 AI 容量建设中的两个主要问题。
亚马逊 AWS 正在积极扩大容量，2025 年增加的数据中心容量比任何其他公司都多。
微软 Azure 的增长势头强劲，但管理层选择将资源优先分配给第一方应用程序和研发，而不是短期 Azure 增长。
谷歌云的 AI 采用率迅速提高，近 75% 的云客户现在使用谷歌的垂直集成 AI 堆栈。
阿里巴巴云的 AI 相关云产品收入连续第十个季度实现三位数增长，对外部云收入的贡献逐季增加。
甲骨文 OCI 的增长速度有可能在 2029-2030 年使 OCI 达到或超过 GCP 的水平。

未来展望

投资者希望比较和对比不同的供应商，了解每个公司的收入和利润率轨迹。
AI 将成为云计算市场的重要驱动力，但同时也将对毛利率产生负面影响。
超大规模云提供商将继续扩大容量，以满足 AI 需求的增长。
未来几年，云计算市场将继续保持强劲增长势头，但竞争也将更加激烈。

Cloud in the Quarter: How did the hyperscalers do in 4Q25? The hyperscale market has become a picks & shovels basket to support theGenerative AIwave even though they are predominantly CPU driven businesses. Even the stock prices arebeing driven by AI sentiment. Investors remain concerned about the CAPEX growth and theassociated margins / returns of Gen AI-related infrastructure investment as AI becomes amore meaningful contributor to revenues. Suddenly, the laggard in hyperscale appears theleader in AI cloud with Google’s vertically integrated solution now neck-and-neck with AWSfor incremental cloud dollars added Q/Q. Mark L. Moerdler, Ph.D.+1 917 344 8506mark.moerdler@bernsteinsg.com Mark Shmulik+1 917 344 8508mark.shmulik@bernsteinsg.com Robin Zhu+852 2123 2659robin.zhu@bernsteinsg.com Even with CAPEX stepping up, CSPs all point to capacity constraints, before it was GPUavailability, but that seems to have shifted to physical powered up datacenter capacity.All thisushers in a slew of new investor questions: Are Microsoft and Amazon going to see a returnon CAPEX? Is speed and cost to bring capacity online the new basis of competition? Do themodel wars determine winners and losers? What are the economics of Oracle’s AI build-out,and can they rely on their OpenAI commitments? Are NVIDIA and OpenAI losers to Google orkingmakers? When will AI drive increased cloud IT budgets? Bernstein's Cloud in the Quarternote, which focuses on the largest hyperscale Cloud providers, should help investors get abetter picture of the secular issues and why results may differ between providers. Firoz Valliji, CFA+1 917 344 8316firoz.valliji@bernsteinsg.com Shelly Tang, CFA+1 917 344 8342shelly.tang@bernsteinsg.com Deeksha Pandey+1 917 344 8447deeksha.pandey@bernsteinsg.com The hyperscale market is the largest tangible market opportunity in Software / Cloud /Internet ($1.3-1.5T — see Diving into the Cloud TAM) and possibly the largest in all oftech. Bernstein's Global Hyperscale Cloud Team (Global Software, U.S. Internet, and ChinaInternet) supplies this quarterly note to help pull all the data together in one place forAmazon, Microsoft, Google, Alibaba and Oracle. Wenhuan Chang+1 917 344 8546wenhuan.chang@bernsteinsg.com Updating our data through this quarter, today's note looks at the metrics each of thecompanies supply as well as our estimates through Q4 CY2025. The note also supplies ourupdated thoughts on the different companies' hyperscale businesses, where we are on AI,and how the markets and companies are progressing on what we believe is going to be avery long and valuable journey. This is the 20thnote in our series comparing and contrastingthe hyperscale vendors. This many years of historical perspective should help investorsbetter understand the trends and the opportunity. Charles Gou+852 2123 2618charles.gou@bernsteinsg.com To further help understand the long-term opportunities, we've published notes divinginto the historical trends in IaaS and separately PaaS in the U.S. and globally. We give ourthoughts on the progression of the industry (see Link). We would also call attention to ournotes that looks at the capital intensity of IaaS/PaaS based on our proprietary work oncapital intensity at Azure and AWS — What does it really cost to offer IaaS / PaaS? as well asthe impact of AI driven CAPEX in Big Tech's big spending... big returns?, Why Azure CAPEXwill slow and revenue accelerate, Microsoft Azure AI margins, Oracle: The story is betterthan we thought, with better upside potential On the Generative AI opportunity we have published numerous notes including: Can GPUintensive AI compute be a profitable business for hyperscalers?, Generative AI 101 Primer,Gen AI 201 — Inferencing, Generative AI 301 — Reasoning models, Generative AI 401:Agents, AI Infrastructure: The build-out is huge. Bonanza or bubble?, Shifting from Trainingto Inferencing impact on datacenters and 5 Takeaways on datacenter build-out. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS Microsoft (MSFT, TP $641, Outperform):Microsoft made a big bet on Azure, investing aggressively in a global footprintand technology innovation. Starting in ~CY2019 they doubled down on their existing AI investment as they prepared for andadded Gen AI capabilities, built AI datacenters and invested in their own chips. Despite a slight miss to buy side whisper numberlast quarter, Microsoft’s growth engine is in fact getting stronger given solid Azure growth and impressive RPO growth. Wealso believe Microsoft is strongly positioned in the case of an AI bubble because their first-party usage of capacity is rampingthanks to their diversified app base, along with third-party contracts which comes from not only OpenAI but a diversified largeenterprise customer base, and contract duration is well aligned with useful life of the equipment. While most of software isslowing, Azure is stable and will, we believe, accelerate supporting Microsoft’s strong double-digit rev

点击免费查看完整报告