Microsoft conference call: AI demand exceeds expectations, capacity may be limited after June, tariffs mentioned only once, economic pressure coincides with the acceleration of cloud computing penetration

On Wednesday after the U.S. stock market closed, Microsoft Corporation's latest financial report showed that the expansion of its cloud division helped the company exceed sales and profit expectations, with total revenue growing by 13% to reach $70.1 billion and adjusted earnings per share of $3.46. Azure cloud business revenue grew by 33%, surpassing expectations. In this financial report, Microsoft's strong profit figures and upwardly revised guidance conveyed positive signals, with the stock price rising over 9% in after-hours trading on Wednesday. It is noteworthy that during this conference call, Microsoft mentioned the potential impact of tariffs only once and hardly mentioned Sam/OpenAI. Here are the highlights from the earnings call: AI demand exceeding expectations may lead to capacity constraints after June. Initially hoping to achieve supply-demand balance by the end of the fourth quarter, we are indeed seeing demand grow throughout the quarter, so we will experience some supply shortages and tightness by year-end. This quarter's Azure performance exceeded expectations mainly due to non-AI services (enterprise customers adopting acceleration), with the excess contribution from AI coming from our early delivery of some GPU capacity; AI service growth for the next quarter is expected to maintain current levels. As model diversity reduces reliance on single hardware, the gross margin of the AI business has surpassed that of traditional cloud transformation during the same period. This year, we will continue to invest $80 billion in data center construction, with half of that directed towards domestic data centers in the United States. Microsoft mentioned the tariff issue only once, limited to the opening remarks: Windows OEM and device revenue grew by 3% year-on-year, exceeding expectations, as tariff uncertainties throughout the quarter kept inventory levels high. The essence of cloud services is software-driven efficiency—virtualization technology has increased server utilization by tenfold, and system optimization in the AI era will replicate this path. Historical experience shows that periods of economic pressure are precisely when cloud computing penetration accelerates, as companies need to optimize costs through the cloud; however, AI tools (like Copilot) directly enhance employee efficiency, making this demand counter-cyclical. Here is the full transcript of the conference call: Host: Hello everyone, welcome to Microsoft’s Q3 fiscal year 2025 earnings call. All participants are currently in listen-only mode. The Q&A session will follow the formal presentation (operational instructions). This meeting will be recorded. Now, I would like to invite Vice President of Investor Relations Jonathan Neilson to speak. Jonathan Neilson (Vice President of Investor Relations): Good afternoon, thank you all for participating. Joining me today are: Chairman and CEO Satya Nadella, Chief Financial Officer Amy Hood, Chief Accounting Officer Alice Jolla, and Company Secretary and Deputy General Counsel Keith Dolliver You can check our financial report press releases and financial summary slides on the Microsoft Investor Relations website. These materials are intended to supplement today's remarks and provide a reconciliation of GAAP and non-GAAP financial metrics. More detailed financial outlook slides will be updated on the website during the meeting. This meeting will discuss certain non-GAAP metrics, which should not be viewed as a substitute for or superior to GAAP metrics, but rather as additional explanatory items to help investors gain a deeper understanding of the company's performance this quarter and the impact of related events on financial results. Unless otherwise specified, all growth comparisons are based on year-over-year data. If the fixed exchange rate growth rate is the same as the original data, only the original growth rate will be mentioned. Satya Nadella (Chairman and CEO): Thank you, Jonathan. This quarter set a record, thanks to the continued strong performance of Microsoft Cloud, with revenue exceeding $42 billion, growing 22% in constant currency. Cloud and AI have become core resources for businesses to expand output, reduce costs, and accelerate growth. Here are some examples starting from infrastructure: We continue to expand data center capacity. In just this quarter, we opened new data centers in 10 countries across four continents. Thanks to multiple compound growth laws, model performance doubles every six months. We continuously optimize every layer from data center design, hardware and chips to system software and model optimization to reduce costs and improve performance. This is reflected in our supply chain: the delivery cycle for new GPUs has shortened by nearly 20%; AI performance (per unit of power) in mixed clusters has improved by nearly 30%; and the cost per token has decreased by over 50%. In terms of cloud migration, we are seeing accelerated demand for Azure from customers across various industries (from Abercrombie & Fitch to Coca-Cola and ServiceNow). For key VMware, SAP, and Oracle workloads for customers, we remain the preferred cloud platform, with regional availability leading any other hyperscale cloud service provider. We are also excited about the new frontiers of quantum technology in the cloud. In addition to deploying quantum stacks to partner machines, we have made tangible progress on the path to building practical quantum computers with the launch of Majorana 1. In the field of data and analytics, we have deeply integrated the AI platform with the data stack. PostgreSQL usage has accelerated for three consecutive quarters, with 60% of Fortune 500 companies (such as BMW and BNY Mellon) adopting it. Cosmos DB revenue growth has accelerated again, remaining the preferred database for globally distributed NoSQL workloads at any scale, with customers including industry leaders like CarMax, DocuSign, NTT Data, and OpenAI. This quarter, the consumption of analytics services has also accelerated. Microsoft Fabric has over 21,000 paying customers (an 80% year-over-year increase), integrating data warehousing, data science, real-time intelligence, and Power BI into an end-to-end solution. Real-time intelligence is now the fastest-growing workload in Fabric (40% of customers have adopted it within 5 months of launch). Overall, over 50% of Fabric customers (such as Amorepacific, the State of Louisiana, and Petrobras) are using three or more workloads Our multi-cloud data lake OneLake has seen its data volume grow more than six times in the past year. AI Platforms and Tools: Foundry is a factory for intelligent agents and AI applications, now used by over 70,000 enterprises (from Atomic Work, Epic, Fujitsu, GainSight to H&R Block, LG Electronics) for designing, customizing, and managing their AI applications and agents. This quarter, we processed over 100 trillion tokens (a fivefold year-on-year increase), reaching a record 50 trillion tokens just last month. Four months after launch, over 10,000 enterprises are using our new agent services to build, deploy, and scale agents. This quarter, we provided customers with an industry-leading fine-tuning toolkit and introduced OpenAI's latest models as well as new models from Cohere, DeepSeek, Meta, Mistral, and Stability into Foundry. We also expanded the Phi series of small models (SLMs), adding multimodal and mini models. The total downloads of Phi reached 38 million. The research team made further breakthroughs with BitNet b1.58 (a large language model with 1 billion parameters that can run solely on CPU), which is set to launch on Foundry. Developer Tools: We upgraded GitHub Copilot from "pair programmer" to "peer programmer." With the agent mode in VSCode, Copilot can now iterate code, identify errors, and automatically fix them. This feature complements other Copilot agents such as Autofix (which helps fix vulnerabilities) and Code Review Agent (which has reviewed over 8 million pull requests). We are previewing the first SUI agent for executing development tasks asynchronously. Overall, the number of GitHub Copilot users exceeds 15 million (a year-on-year increase of over four times). Digital-native enterprises like Twilio and traditional companies like Cisco, HPE, Skyscanner, and Target have all chosen GitHub Copilot to provide AI support for developers throughout the entire lifecycle. Visual Studio and VSCode are the most popular editors globally (with over 50 million monthly active users). Power Platform is the leading low-code platform for AI creators. Power Platform has 56 million monthly active users (a year-on-year increase of 27%), and its AI capabilities are increasingly used to build applications and automate processes. Future Work Methods: Microsoft 365 Copilot is designed to facilitate human-machine collaboration. Currently, hundreds of thousands of customers across various industries worldwide are using Copilot (a threefold year-on-year increase). The overall transaction scale continues to expand, with record numbers of customers adding seats this quarter. We are continuously expanding features: last week we released a major update that integrates agents, notes, search, and creation functions into a new work framework The brand new "Researcher" and "Analyst" deep reasoning intelligent agents can analyze massive amounts of network and enterprise data, providing high-skill expertise on demand directly within Copilot. In addition to general knowledge work, we are introducing agents for each role and business process. The sales agent converts contacts into qualified leads and helps representatives quickly familiarize themselves with new customers through the "Sales Chat" feature. The customer service agent can intercept customer inquiries and assist service representatives in resolving issues more quickly. Through Copilot Studio, customers can expand Copilot and build their own agents without coding or with low-code. Over 230,000 enterprises (including 90% of the Fortune 500) have used Copilot Studio. With the deep reasoning and agent processes of Copilot Studio, customers can build agents that handle complex tasks and deterministic scenarios (such as document processing and financial approvals), and create computer-use agents that perform actions on desktop and web application UIs. Customers can now convert SharePoint sites into agents with one click. In just this quarter, customers created over 1 million custom agents through SharePoint and Copilot Studio (a quarter-on-quarter increase of 130%). Business Applications: Dynamics 365 continues to capture market share, with companies like Avaya, Brunswick, and Softcat shifting from traditional vendors to Dynamics. For example, Verizon chose the Dynamics 365 sales module to enhance the efficiency of its sales team. In the healthcare sector, Dragon Copilot has had a smooth start. Last quarter, we assisted institutions such as City of Hope, Ottawa Hospital, Tufts Medical Center, and Wellstar in recording nearly 9.5 million doctor-patient interactions (a quarter-on-quarter increase of over 50%). In manufacturing, we launched factory operations and safety intelligent agents at the Hannover Messe, with leading partners like Autodesk, PTC, and Siemens building industrial AI solutions based on our technology stack. In the retail sector, we launched agents to help clients like Bath & Body Works create personalized shopping experiences and optimize store operations. In the Windows sector: Copilot+ PCs are faster and have better battery life, making them the best in their class. We continue to win new customers with top-notch AI capabilities, providing AI applications from partners like Adobe, Canva, and Zoom. Last week, we launched exclusive AI features (such as Recall, Click to Do, and Windows Search) for all Copilot+ PCs. With the end of support for Windows 10, the deployment of Windows 11 in businesses has increased by nearly 75% year-on-year. In the security sector: Security is our top priority, and we have made significant progress on the engineering goals set out in our "Secure Future Plan" a year and a half ago. We applied these experiences to platform innovation: last month, we launched the Security Copilot agent with partners to help defenders autonomously handle high-frequency security and IT tasks based on 84 trillion threat signals daily We have also added new features for Defender, Entra, and Purview to assist enterprises in securing and governing AI deployments. Overall, we currently have 1.4 million security customers, with over 900,000 enterprises (such as Ernst & Young Global, ManpowerGroup, Trinet, Regions Bank) adopting four or more services (a year-on-year increase of 21%). The monthly active users of the identity management platform Entra exceed 900 million. Consumer Business: LinkedIn: Over 1 billion professionals connect, learn, recruit, and sell through LinkedIn, with membership numbers maintaining double-digit year-on-year growth. The platform's video viewing time increased by 36%, and the number of comments grew by 32%. More members are using AI to learn skills and seek jobs, with the number of AI-assisted learners increasing more than twofold month-on-month. We remain the leader in the recruitment market, with clients like Equinix and Verizon finding qualified candidates faster through LinkedIn Hiring Assistant. The SMB subscription service "Premium Pages" for LinkedIn Premium saw a more than 75% month-on-month increase in subscriptions this quarter. LinkedIn Marketing Solutions has accelerated revenue growth for two consecutive quarters and remains the best way to reach B2B decision-makers. Advertising Business: We are reconstructing search, browsing, content discovery, and personal assistant experiences through AI: Copilot Search and Bing are reshaping search results with AI-curated overview pages and embedded conversational features. Copilot Vision and Edge allow Copilot to respond in real-time to browsing content. Copilot Discover personalizes the MSN experience based on user interactions and preferences. The new Copilot app focuses on enhancing daily engagement and successful conversations in scenarios such as dialogue, search, shopping, and travel planning. Overall, Bing and Edge's market share continues to expand, with total advertising revenue exceeding $20 billion in the past 12 months. Gaming Business: We are continuously transforming our business and focusing on improving profit margins, reaching over 500 million monthly active users across devices. This quarter, we became a top publisher with pre-orders and pre-installs in the Xbox and PlayStation stores. PC Game Pass revenue increased by over 45% year-on-year. Through Xbox Play Anywhere, players can enjoy over 1,000 games on both console and PC. Last week, we introduced cloud gaming to LG TVs, with cloud gaming hours surpassing 150 million for the first time this quarter. We are integrating AI into Xbox: The new Copilot for Gaming is a personalized gaming companion that provides in-game assistance and expert guidance. The pioneering Muse model can generate gameplay in real-time. The "Minecraft" movie has become the annual box office champion, with new monetization pathways for the IP driving a year-on-year growth of over 75% in weekly active users for the game. Summary: We are rapidly innovating to expand business and consumer opportunities. At the Build conference in a few weeks, we will share how we are creating the most powerful AI platform for developers, stay tuned. Now, let's welcome Amy to speak. Amy Hood (Chief Financial Officer): Thank you, Satya. Good afternoon, everyone. This quarter, revenue reached $70.1 billion, a year-on-year increase of 13% (15% increase at constant currency). Gross profit grew by 11% (13% increase at constant currency), operating profit increased by 16% (19% increase at constant currency), and earnings per share were $3.46 (19% increase at constant currency). The performance exceeded expectations, thanks to the efficient execution of the sales and partner teams. Strong customer demand for cloud and AI services continues to drive business growth. AI-related revenue once again exceeded expectations. Commercial order volume grew by 18% (17% increase at constant currency), primarily driven by large contracts with OpenAI's Azure. Core renewal sales and long-term platform commitments remained stable. The commercial remaining performance obligations increased to $315 billion (34% increase at constant currency), of which about 40% will be recognized as revenue in the next 12 months (17% year-on-year growth), and the remaining portion (recognized after 12 months) grew by 47%. This quarter, renewals accounted for 98%. The impact of exchange rates on total revenue, revenue from each segment, and operating expense growth was generally in line with expectations, with total revenue growth being dragged down by 1 percentage point due to exchange rates. Microsoft Cloud revenue was $42.4 billion (exceeding expectations), with a 22% increase at constant currency. Microsoft Cloud gross margin was 69% (in line with expectations), a year-on-year decrease of 3 percentage points (affected by AI infrastructure expansion). The overall gross margin for the company was 69% (a year-on-year decrease of 1 percentage point), for the same reason. Operating expenses grew by 2% (3% increase at constant currency), below expectations (thanks to cost efficiency and some investments being pushed to Q4). The operating profit margin increased by 1 percentage point year-on-year to 46%, exceeding expectations (due to team structure optimization and management level streamlining). As of the end of March, the total number of employees increased by 2% year-on-year, with a slight decrease quarter-on-quarter. Performance by segment: Productivity and Business Processes (revenue of $29.9 billion, 13% increase at constant currency): The better-than-expected performance came from LinkedIn, M365 commercial and consumer versions. M365 commercial cloud revenue increased by 15% at constant currency (in line with expectations), with ARPU growth driven by the E5 package and Copilot. Paid M365 commercial seats increased by 7% to over 430 million, with all customer groups achieving an expansion of the install base (mainly due to growth in small and medium-sized enterprises and frontline employee services). M365 commercial product revenue increased by 8% at constant currency (exceeding expectations due to increased Office transaction purchases). M365 consumer cloud revenue increased by 12% at constant currency (exceeding expectations, driven by subscription growth from the January price increase) M365 consumer subscriptions increased by 9% to 87.7 million. LinkedIn revenue grew by 8% at constant currency (exceeding expectations, strong performance across all businesses), with the talent solutions business still affected by a weak job market. Dynamics 365 revenue increased by 18% at constant currency (in line with expectations, with continued growth across workloads). Segment gross profit increased by 13% at constant currency, with gross margin year-on-year remaining flat (impact of AI infrastructure expansion offset). Operating expenses increased by 1% (2% at constant currency), and operating profit increased by 18%. Intelligent Cloud (revenue of $26.8 billion, up 22% at constant currency): Azure and other cloud services revenue increased by 35% at constant currency (including a 16 percentage point contribution from AI services). Non-AI service growth accelerated (improvements in enterprise customers and scale operations), with AI service capacity coming online faster than expected. On-premises server business revenue decreased by 4% at constant currency (slightly below expectations, mainly due to changes in contract revenue recognition structure). Enterprise and partner services revenue increased by 6% at constant currency (slightly exceeding expectations, strong performance in enterprise support services). Segment gross profit increased by 14% at constant currency, with gross margin decreasing by 4 percentage points (impact of AI infrastructure expansion). Operating expenses increased by 7%, and operating profit increased by 18%. More Personal Computing (revenue of $13.4 billion, up 7% at constant currency): All businesses exceeded expectations. Windows OEM and device revenue increased by 3% (exceeding expectations, with tariff uncertainties keeping inventory high). Search and news advertising revenue (excluding TAC) increased by 23% at constant currency (significantly exceeding expectations, driven by third-party partnerships boosting usage, rate increases, and Bing/Edge traffic growth). Gaming revenue increased by 6% at constant currency, with Xbox content and services revenue up by 9% (strong performance from both third-party and first-party content). Segment gross profit increased by 11% at constant currency, with gross margin improving by 2 percentage points (improvements in search and gaming profitability). Operating expenses increased by 1%, and operating profit increased by 23%. Overall company financials: Capital expenditures (including finance leases) were $21.4 billion (slightly below expectations, mainly due to fluctuations in data center lease delivery times). Cash expenditures for real estate, plants, and equipment were $16.7 billion, with about half of cloud and AI-related expenditures supporting long-term assets for the next 15 years or more, and the remainder for servers (CPU/GPU) deployed based on demand signals (including a $315 billion customer contract backlog). Operating cash flow was $37 billion (up 16% year-on-year), with free cash flow of $20.3 billion. Other income and expenses had a net loss of $623 million (better than expected, mainly due to net gains from derivatives and investments), with equity method investment losses slightly exceeding expectations. Effective tax rate was approximately 18%. Returned $9.7 billion to shareholders through dividends and buybacks (up 15% year-on-year). Fourth quarter outlook (unless otherwise noted, all figures are in USD): As of April, the trends in commercial business, LinkedIn, gaming, and search demand are stable, assuming this trend continues. Performance may be affected if the environment changes. Windows OEM business is expected to see inventory levels decline from Q3 highs, with the personal computing segment's guidance range widened to cover fluctuations. Against the backdrop of a weaker dollar, exchange rates are expected to boost total revenue growth by 1 percentage point (1 percentage point increase for the productivity and business processes segment, and less than 1 percentage point increase for the intelligent cloud and personal computing segments). The impact of exchange rates on cost and operating expense growth is less than 1 percentage point. Commercial order volume is expected to grow steadily (high base and increased expiring contracts), with volatility in long-term Azure contracts potentially increasing quarterly growth rate fluctuations. Microsoft's cloud gross margin is expected to be around 67% (year-on-year decline, mainly due to AI infrastructure expansion). Capital expenditures: Q4 is expected to increase sequentially, with the total for the second half maintaining the January guidance (cloud infrastructure construction and financial leasing delivery times may cause quarterly fluctuations). Segment guidance: Productivity and Business Processes: Revenue of $32.05-32.35 billion (11%-12% growth at constant currency). M365 commercial cloud revenue is expected to grow by 14% (flat compared to the previous quarter), driven by E5 and Copilot, with seat growth potentially slowing due to a high base. Single-digit growth in M365 commercial product revenue (including Windows commercial local components and Office transactional purchases, revenue recognition structure may cause fluctuations). M365 consumer cloud revenue is expected to grow in the mid-teens (driven by January price increases). LinkedIn revenue is expected to grow in the high single digits. Dynamics 365 revenue is expected to grow in the mid-to-high double digits (growth across all workloads). Intelligent Cloud: Revenue of $28.75-29.05 billion (20%-22% growth at constant currency). Azure revenue is expected to grow by 34%-35% (contract structure may cause quarterly fluctuations), with healthy growth in non-AI services, and AI demand exceeding expectations may lead to capacity constraints after June. On-premises server revenue is expected to decline in the mid-single digits (customers continue to shift to the cloud). Enterprise and partner services revenue is expected to grow in the mid-to-high single digits (driven by enterprise support services). More Personal Computing: Revenue of $12.35-12.85 billion. Windows OEM and device revenue is expected to decline in the mid-to-high single digits (OEM revenue declining in the mid-single digits under the assumption of inventory destocking; device revenue declining in the high double digits). Search and news advertising revenue (excluding TAC) is expected to grow in the double digits (driven by increased share of Bing/Edge leading to volume and rate growth), with TAC-inclusive revenue growing in the mid-double digits. Gaming revenue is expected to grow in the mid-single digits, with Xbox content and services revenue growing in the high single digits (driven by first-party content). Overall company guidance: Costs are expected to be $23.26-23.8 billion (19%-20% growth at constant currency), with operating expenses of $18-18.1 billion (5% growth at constant currency) Despite ongoing investments in AI, the operating profit margin for fiscal year 2025 is expected to rise slightly year-on-year. Other net losses amount to approximately $1.2 billion (mainly due to losses from equity method investments). The effective tax rate is approximately 19%. Fiscal Year 2026 Outlook: Capital expenditure growth will be lower than in fiscal year 2025, with an increase in the proportion of short-term assets (more directly related to revenue). Jonathan Nelson: Thank you, Amy. Now we will enter the Q&A session. Please limit each questioner to one question. Host, please repeat the instructions. Host: The first question comes from Keith Weiss of Morgan Stanley. Keith Weiss: First, congratulations on your outstanding performance in such an uncertain environment. Recent media reports indicate that Microsoft is adjusting its data center construction plans, but at the same time, strong AI demand is leading to supply constraints. Can you elaborate on the adjustments to the data center strategy? In particular, does the GPU surplus risk you mentioned earlier affect the current planning? Satya: We always adjust the pace of construction and leasing based on future workloads. The key is to balance demand growth, workload patterns, and regional layout. Changes in AI inference and pre-training computing, along with Moore's Law and system software optimization, all need to be included in the planning. We must ensure sufficient power in specific regions to support leasing or construction. Amy can add more. Amy: Data center decisions involve a 5-7 year cycle and require continuous balancing of demand and technological evolution. This quarter, the team delivered capacity ahead of schedule, improving efficiency. Although supply and demand remain tight at the end of Q4, progress is positive. Brent Thill (Jefferies): You mentioned that the demand for cloud migration is accelerating. Can you elaborate on the driving factors? Satya Nadella: This involves four overlapping effects: Basic migration: Traditional workloads like Windows Server/SQL Server continue to migrate to the cloud (cloud cost advantages are prominent). Growth in data services: Products like Cosmos DB and PostgreSQL are adopted by 60% of the Fortune 500, with a surge in demand for Fabric real-time analytics. Cloud-native expansion: Non-AI computing and storage demand remains healthy. AI synergy: Taking OpenAI as an example, it not only uses AI accelerators but also relies on Cosmos DB and basic computing resources—there is a resource ratio of 1:3 between AI and non-AI workloads. Mark Moerdler (Bernstein): If the economy enters a recession, can Microsoft's current business structure withstand the impact? Nadella: Our multi-layer product matrix (SaaS applications + infrastructure) essentially helps enterprises achieve "more output with less input." Historical experience shows that economic pressure periods are precisely when cloud computing penetration accelerates (enterprises need to optimize costs through the cloud); AI tools (like Copilot) directly enhance employee efficiency, and this type of demand is counter-cyclical Karl Keirstead (UBS): The 16 percentage point growth in Azure comes from AI services. Is this primarily driven by the demand for ChatGPT inference? Will this contribution rate be higher in Q4? Amy Hood: I need to clarify: The outperformance of Azure this quarter mainly comes from non-AI services (accelerated adoption by enterprise customers). The excess contribution from AI comes from our early delivery of some GPU capacity. The growth of AI services in Q4 is expected to maintain the current level, but it should be noted that the new capacity may not fully meet demand. Kash Rangan (Goldman Sachs): As the AI business expands, can the growth rate of capital expenditures slow while maintaining Azure growth? Amy Hood: The key lies in the technology compounding effect: Software optimization continuously improves the efficiency of existing hardware (such as optimizing CPU utilization with the Phi model). The diversity of models reduces reliance on a single hardware. The gross margin of the AI business is already higher than that of traditional cloud transformation during the same period. Nadella: The essence of cloud services is software-driven efficiency—virtualization technology has increased server utilization by ten times, and system optimization in the AI era will replicate this path. Mark Murphy (JP Morgan): Has the "DeepSeek moment" (a leap in model efficiency) extended the lifespan of GPUs? Will there be adjustments to asset depreciation policies? Nadella: Software advancements have indeed extended the lifecycle value of hardware (such as optimizing inference to tap into the potential of existing GPUs), but specific accounting treatments require long-term data support. The current focus is on improving asset utilization through system-level optimization. Amy Hood: Adjustments to depreciation policies need to be cautious. We currently prioritize extending hardware value through software upgrades rather than modifying financial assumptions. Kirk Batern (Evercore ISI): Is the accelerated growth of non-AI Azure services sustainable? Amy Hood: The driving factors include enterprise customers (especially Fortune 500) accelerating their cloud adoption, cross-industry penetration of data services (such as Fabric), and improved execution by partners (issues in Q2 have been partially resolved). The trend is sustainable, but we need to continue monitoring the adoption pace among small and medium-sized enterprises. Alex Zukin (Wolfe Research): Is the boundary between AI and non-AI workloads becoming increasingly blurred? Does this change the Azure growth curve? Amy Hood: It is indeed difficult to strictly distinguish— for example, digital-native enterprises run both AI training and traditional workloads simultaneously. As AI applications deepen, the synergy between the two types of resources will strengthen Azure's overall value proposition, rather than relying solely on a single growth pole