Microsoft conference call: Order surge, Azure in short supply, data center tightness expected to last until 2026

Microsoft achieved double-digit growth in revenue and profit in the first quarter of fiscal year 2026, but the capacity bottleneck of its cloud computing business Azure is becoming a key constraint on growth. According to a previous article by Wall Street Insight, Microsoft’s revenue in the third quarter increased by 18% year-on-year, maintaining the highest growth rate in a year and a half achieved in the second quarter, while EPS growth slowed to 13%, still exceeding analyst expectations; revenue from Azure and other cloud services grew by 39%, matching the highest growth rate in two and a half years achieved in the second quarter, but still below some buyers' optimistic expectations; capital expenditures in the third quarter reached a new high, up 60% from the previous record and over 74% year-on-year; investment in OpenAI impacted net profit by nearly $3.1 billion in the third quarter. Microsoft Chairman and CEO Satya Nadella and Microsoft Executive Vice President and CFO Amy Hood communicated with analysts during the subsequent earnings call. Hood stated in the conference call that the demand for Azure services "far exceeds our current capacity," and that spending is expected to increase in the current fiscal quarter. Microsoft's capital expenditures in the first quarter reached $34.9 billion, including leasing costs, indicating an increase of about $10 billion in spending on server farms. However, Hood mentioned in the call with analysts, "I thought we would catch up (with demand), but we did not." Microsoft expects the tightness in data centers to persist until 2026. In response to this challenge, Microsoft is taking proactive measures to alleviate the pressure on data center capacity. The company plans to double the total footprint of its data centers in the next two years and continue optimizing the performance and efficiency of existing data centers. Additionally, Microsoft is continuously exploring new technologies and solutions to improve the utilization and flexibility of data centers. Regarding stock performance, Morgan Stanley analysts mentioned during the Q&A session that despite Microsoft's strong performance, its stock performance has lagged behind the market, questioning whether AGI will change Microsoft's strong positioning in the market or even weaken this advantage. In response, Satya stated that the new agreement with OpenAI brings certainty to intellectual property relations, and the deployment of AI systems in the real world brings value to customers and providers, with Microsoft aiming to create significant value for customers by leveraging advancements in AI models. The following is a summary of the key points from the conference call: Explosive Demand and Huge Contracts Microsoft's performance growth is primarily driven by strong demand for its AI platform and Copilot series products. The earnings report shows that Microsoft’s cloud revenue has exceeded $49 billion, a year-on-year increase of 26%. A key indicator for future revenue—remaining performance obligations (RPO)—surged over 50%, reaching nearly $400 billion, with a weighted average performance period of only two years, indicating that a large number of contracts will convert to revenue in the short term. CEO Satya Nadella revealed during the call that the company has reached a new final agreement with OpenAI, marking a new chapter in their collaboration. According to the agreement, OpenAI has signed an incremental contract for Azure services worth $250 billion In addition, Microsoft's exclusive IP rights and API exclusivity on Azure will last until the realization of AGI (Artificial General Intelligence) or 2030. The adoption rate of Microsoft's AI tools among enterprise customers is astonishing. The number of customers using Azure AI has reached 80,000, including 80% of Fortune 500 companies. Microsoft 365 Copilot, released just 9 months ago, is being used by over 90% of Fortune 500 companies. Meanwhile, GitHub Copilot, aimed at developers, has over 26 million users. The widespread deployment of these first-party and third-party applications is continuously driving the overall growth of Azure. Surge in Capital Expenditure and Supply Bottlenecks To meet unprecedented demand, Microsoft is making capital investments on an unprecedented scale. In the first quarter, Microsoft's capital expenditure (including leases) reached $34.9 billion, 60% higher than the previous record in the last quarter and up over 74% year-on-year. Amy Hood explained that about half of this amount is used to procure short-term assets like GPUs and CPUs to meet the needs of the Azure platform, first-party applications, and AI research and development. Despite the massive investment, capacity shortages still exist. Amy Hood admitted, "I thought we would catch up (with demand), but we haven't." When Deutsche Bank analyst Brad Zelnick inquired about the impact of capacity shortages on revenue, Hood responded that while it is difficult to quantify precisely, it indeed "directly affects Azure," as the company needs to prioritize the computing power requirements of high-priority businesses like Microsoft 365 Copilot. To address this bottleneck, Microsoft is accelerating the construction of its "globally scaled cloud and AI factories." The company plans to nearly double the total footprint of its data centers over the next two years and has announced the construction of the world's most powerful AI data center in Wisconsin, with a total scale of 2,000 megawatts. At the same time, the company has deployed the world's first large-scale NVIDIA GB300 cluster and has increased the token throughput of core models like GPT-4o by over 30% through software optimization, maximizing the efficiency of existing facilities. Addressing Investor Concerns About "AI Bubble" and Investment Risks In the face of massive capital expenditures and seemingly endless demand, concerns about an "AI bubble" and investment risks have emerged among investors. During the conference call, Bernstein Research analyst Mark Moerdler directly asked how to ensure that there won't be overbuilding to meet current demand. In response, Amy Hood emphasized that the nearly $400 billion RPO, which measures the total value of contractual obligations that companies have yet to fulfill by a specific date, can help investors predict future revenue potential. Additionally, RPO represents signed contracts, indicating real, strong demand rather than speculation. She pointed out that the company's capital expenditure strategy is prudent and aligned with contract cycles, such as the lifespan of short-term assets like GPUs and CPUs being roughly consistent with the duration of related contracts, thereby controlling risks When asked about the concentration risk of a single customer, Nadella elaborated on his strategic resolve. He stated that Microsoft's goal is to build a "highly replaceable" global fleet to serve a diverse customer base, including first-party, third-party, enterprise, and digital-native companies. Nadella revealed that the company selectively accepts demand, choosing to reject those that are overly concentrated and do not align with long-term strategic goals. He candidly remarked, "Every time we say 'no,' I feel better the next day." This statement aims to convey a clear message to the market: Microsoft is pursuing sustainable, balanced long-term growth rather than reckless short-term expansion. New OpenAI Agreement and Financial Impact Microsoft's deep partnership with OpenAI is at the core of its AI strategy and a focal point for investors. Nadella stated that the new agreement brings "more certainty" to the intellectual property relationship between the two parties, solidifying Microsoft's strategic advantage. He disclosed that Microsoft's investment value in OpenAI has increased by approximately tenfold. However, this collaboration has also brought significant financial impacts. According to information from the earnings call, in the first quarter, Microsoft recorded a loss under "other income and expenses," primarily due to recognizing its share of losses in OpenAI under the equity method. UBS analyst Karl Keirstead noted that this figure far exceeds previous amounts, and Amy Hood explicitly stated that the increase in losses "is entirely due to our recognition of OpenAI's losses under the equity method," emphasizing that the first quarter's performance was not affected by the newly signed agreement this week. Here is the full transcript of the earnings call: Chairman and CEO Satya Nadella: This is a strong start to our fiscal year. Microsoft Cloud revenue exceeded $49 billion, growing 26% year-over-year, with commercial RPO growing over 50% to nearly $400 billion, and a weighted average duration of only two years. We are seeing increasing and spreading demand for our AI platform and the Copilot family, driving our investments in capital and talent. In terms of infrastructure, we are building a globally scaled cloud and AI factory that maximizes token value per watt per dollar while supporting the data sovereignty needs of customers in various countries. We are rapidly innovating within the Copilot family across high-value areas such as information work, coding, security, science, health, and consumer. As you saw yesterday, we reached a new definitive agreement with OpenAI, marking a new chapter in our industry's most successful partnership and investment. This is a great milestone for both companies, and we will continue to benefit from each other's multifaceted developments. Currently, our investment has grown by approximately tenfold. OpenAI has signed incremental contracts for Azure services worth $250 billion. Our exclusive IP rights and API exclusivity for Azure will last until AGI or 2030, with model and product IP rights extended to 2032 We are also vibrant, driving AI innovation through talent and computational investment, which has had a significant impact on the real world. In terms of AI platforms and Copilot and agents, we have the largest fleet of data centers in the AI era and are scaling up at an unprecedented rate. This year, we increased our total AI capacity by over 80% and plan to nearly double the total footprint of our data centers in the next two years, reflecting the demand signals we are seeing. This quarter, we announced the world's most powerful AI data center—Fairwater in Wisconsin, expected to go live next year, with a total scale of 2 gigawatts. Additionally, we have deployed the world's first large-scale NVIDIA GB300 cluster. We are building a replaceable fleet and continuously modernizing it, covering all stages of the AI lifecycle, from pre-training to post-training, to synthetic data generation and inference, extending to GenAI workloads, recommendation engines, databases, and streaming. We maximize fleet performance and efficiency through cross-chip, system, and software optimization. It is this combination of interchangeability and continuous optimization that enables us to provide the best ROI and TCO for our customers and ourselves. For example, this quarter we increased the token throughput of GPT-4o and GPT-5 (the two most widely used models) by over 30% per GPU. We also have the most comprehensive digital sovereignty platform, with Azure customers in 33 countries building their own cloud and AI capabilities domestically to meet local data residency requirements. For instance, in Germany, OpenAI and SAP will leverage Azure to provide new AI solutions for the public sector. On this infrastructure, we built Azure AI Foundry to help customers build their own AI applications and agents. We have 80,000 customers, including 80% of the Fortune 500 companies. We provide developers and enterprises access to over 11,000 models, more than any other vendor, including OpenAI's GPT-5 and xAI's Grok-4 this quarter. For example, Ralph Lauren uses Foundry to build conversational shopping experiences in its applications, allowing customers to describe their needs and receive personalized recommendations; OpenEvidence uses Foundry to create AI-driven clinical assistants that present relevant medical information to doctors and streamline medical record processes. Speaking of our first-party models, we are excited about the performance of the new MAI model for text, speech, and image generation, which made its debut on industry leaderboards. Our SLM family continues to progress, with cumulative downloads exceeding 60 million, tripling year-over-year. In addition to the models in Foundry, we provide all the tools developers need to design, customize, and manage large-scale AI applications and agents. The newly launched Microsoft Agent Framework helps developers orchestrate multi-agent systems, achieving compliance, observability, and out-of-the-box deep integration For example, KPMG uses this framework to modernize the audit process, connecting agents with internal corporate data and achieving governance and observability. These large-scale AI deployments are driving overall growth for Azure, which has once again expanded its market share this quarter. In terms of our own built applications and agents, the monthly active users of AI capabilities have reached 900 million. The monthly active users of first-party Copilot across information work, coding, security, science, health, and consumer sectors exceed 150 million. In the information work sector, we continue to innovate Microsoft 365 Copilot, integrating chat and agent workflows into everyday tools like Outlook, Word, Excel, PowerPoint, and Teams. Just 9 months after its release, tens of millions of customers have begun using the chat feature, with a quarter-on-quarter growth of 50%, and usage intensity continues to rise. The agent mode launched this quarter can generate expert-level Word documents, Excel spreadsheets, and PowerPoint presentations from a single prompt and iteratively deliver the final product, similar to the agent mode in coding tools. We are excited about the early responses, including third-party benchmarks rating it as best-in-class. In addition to personal productivity, the multi-user and team mode of Copilot has also gone live, allowing users to invite colleagues to join Copilot conversations. Collaborative agents (such as coordinators and project managers) can prepare meeting agendas, take notes, capture decisions, and initiate tasks. The Copilot agent ecosystem continues to grow, with top ISVs like Adobe, Asana, GREL, Lexus, Nexus, SAP, ServiceNow, Snowflake, and Workday building their own agents and connecting to the Copilot system. Customers are also using Copilot Studio to build agents for critical business processes and integrate them into Copilot. The total number of agent users has doubled compared to the previous quarter. The App Builder released yesterday is a new Copilot agent that allows anyone to quickly create and deploy task-specific applications and agents within the Microsoft 365 environment. These innovations are driving momentum. The adoption rate of Microsoft 365 Copilot among customers exceeds that of any other new suite, with over 90% of Fortune 500 companies using the product. This quarter, Accenture, Bristol-Myers Squibb, EY Global, and the UK’s HM Revenue and Customs purchased over 15,000 seats, while Lloyd’s Banking Group deployed 30,000 seats, saving an average of 46 minutes per employee per day. The vast majority of enterprise customers are also expected to make additional purchases. Partner PWC added 155,000 seats this quarter, currently deploying over 200,000 seats globally. Within six months, PWC employees interacted with Microsoft 365 Copilot over 30 million times, saving millions of work hours In the coding field, GitHub Copilot is the most popular AI programming assistant, with over 26 million users. Tens of thousands of developers at AMD receive hundreds of thousands of lines of code suggestions each month, saving months of development time. The total number of GitHub users exceeds 180 million, with the platform growing at the fastest historical rate, adding one developer every second, and 80% of new developers start using Copilot within the first week. The rise of AI coding agents has driven record usage, with over 500 million pull requests merged in the past year. Yesterday's GitHub Universe conference introduced the Agent Headquarters, where GitHub Copilot integrates xAI and OSS internal models, providing single-task control to launch, manage, and review agents, with each agent running from its own branch, equipped with built-in controls, observability, and governance. We are building a similar security system that integrates over 30 agents into the Copilot system, including Entra, Defender, Purview, and Intune. For example, through the Phishing Triage Agent and Defender, analysts have improved the efficiency of detecting malicious behavior by 6.5 times. In the health sector, Dragon Copilot helps providers automate critical processes, assisting in recording over 17 million patient visits this quarter, a nearly fivefold year-on-year increase. More than 650 healthcare institutions have adopted environmental listening technology, including the University of Michigan Health Center, with over 1,000 doctors actively using it. In terms of AI consumer experience, we are excited about the progress of Copilot, starting with Windows. Every Windows 11 PC is an AI PC. Two weeks ago, we launched a new way to converse naturally with computers, including a visual Copilot wake word that can see screen content and engage in real-time dialogue. In the Edge browser, built-in AI capabilities have been introduced for the first time, automatically executing multi-step workflows and helping users continue interrupted tasks. Edge has maintained its market share for 18 consecutive quarters. The Bing overview page now includes embedded conversational features, with search market share increasing, and Copilot consumer application daily active users growing nearly 50% compared to the previous quarter. We made several updates last week, including Groups, turning Copilot into a shared experience; we also launched a premium Copilot subscription service through Microsoft 365 Premium, combining office applications with advanced Copilot features, allowing users to flexibly and securely use AI. In gaming, Copilot provides a voice-first immersive experience on PC, mobile, and the new Xbox Ally. Besides the Copilot and AI platform family, the entire portfolio maintains strong momentum. Cloud migration is accelerating, with Fabric revenue growing by 60%, outpacing other data and analytics platforms in the industry, and there are currently 28,000 paid Fabric customers SQL DB's revenue grew nearly 75% at a massive scale, while Cosmos DB grew by 50%. Dynamics 365's market share increased. The security end-to-end stack processes 100 trillion signals daily, with 1 billion monthly active Entra users, 16 billion interactions with Purview Audit Copilot, a quarterly growth of 72%, and 40,000 Sentinel customers. LinkedIn has nearly 1.3 billion members, covering every endpoint in the gaming sector, focusing on high-profit content and services. Games such as "Guardians," "Ninja Gaiden 4," and "The Outer Worlds 2" have been launched, with "Minecraft" reaching a record high of 155 million monthly active users, and overall content and service revenue hitting a record this quarter. The response to Xbox Ally has been enthusiastic, setting a new record for PC players. The global cloud and AI factories, along with high-value areas like Copilot, are driving widespread dissemination and real-world impact. We will continue to increase AI investments, including capital and talent, to seize the enormous opportunities ahead. Now, I will hand over to Amy to discuss financial performance and outlook. I look forward to rejoining to answer everyone's questions. Amy Hood, Executive Vice President and Chief Financial Officer: Thank you, Satya. Good afternoon, everyone. First, as Satya mentioned, we were pleased to announce the next phase of our partnership with OpenAI yesterday. They continue to choose Microsoft to support their workloads, and we will continue to drive innovation that meets real-world needs together. Our first-quarter performance was unaffected by the agreement signed this week. Let's look at the specific quarterly data: the fiscal year started strong, with revenue, operating income, and earnings per share all exceeding expectations. Many business areas continue to gain market share, demonstrating our leadership position in key markets. This quarter, revenue was $77.7 billion, growing 18% and 17% at constant currency. Gross profit increased by 18% and 16% in dollars and at constant currency, respectively, while operating income grew by 24% and 22% at constant currency. Earnings per share were $4.13, growing 23% and 21% at constant currency after adjusting for the impact of OpenAI investments. The foreign exchange impact was largely in line with guidance. Driven by AI investments, the company's gross margin was 69%, slightly down year-over-year, including the impact of expanding AI infrastructure and increased usage of AI products. This decline was offset by ongoing efficiency improvements, particularly in Azure and M365 commercial cloud businesses. Due to investments in cloud and AI engineering, including computing power and AI talent to support the development of the entire portfolio of products, operating expenses grew by 5% and 4% at constant currency, respectively. Operating profit margin rose year-over-year to 49%, with high-margin business performance exceeding expectations this quarter. After adjusting for the impact of OpenAI investments, other income and expenses were $401 million, with interest income exceeding interest expenses, which included interest payments related to data center financing leases Driven by the growth in demand for cloud and AI products, capital expenditures for this quarter amounted to $34.9 billion, with approximately half allocated to short-term assets (primarily GPUs and CPUs) to meet the increasing demand for the Azure platform, first-party applications, and AI solution development, accelerate R&D, and continue replacing decommissioned servers and network equipment. The remaining expenditure supports long-term assets that will be monetized over the next 15 years and beyond, including $11.1 billion in finance leases, primarily for large data centers. Cash paid for PP&E was $19.4 billion. The difference between total capital expenditures and cash paid for PP&E is mainly due to finance leases and the normal timing difference for goods received but not yet paid. Operating cash flow was $45.1 billion, a year-on-year increase of 32%, driven by cloud billing and collections, although partially offset by higher vendor payments. Free cash flow increased by 33% to $25.7 billion, with the increase in the finance lease portfolio and the growth in capital expenditures having minimal impact on free cash flow. A total of $10.7 billion was returned to shareholders through dividends and stock buybacks. In terms of business performance: Commercial bookings grew, exceeding expectations for Office 2024 transaction purchases. At constant currency, M365 consumer cloud revenue grew by 26% and 25%, driven by ARPU growth. M365 user subscriptions increased by 7%, surpassing 90 million. LinkedIn revenue grew by 10%, with a 9% increase at constant currency, driven by marketing solutions. The talent solutions business was impacted by a weak hiring market. Dynamics 365 revenue grew by 18%, with a 16% increase at constant currency, with continued revenue growth across workloads. Segment gross profit increased by 19% and 16% in dollars and at constant currency, respectively, with gross margin percentage improvement, benefiting from M365 commercial cloud efficiency improvements, although partially offset by increased AI investments and usage of co-pilot chat systems. Operating expenses grew by 6% and 5% at constant currency, while operating income grew by 24% and 20% at constant currency, with operating profit margin increasing by 3 percentage points year-on-year to 62%, primarily driven by high gross margins and operational leverage improvements. Intelligent Cloud business segment: Revenue was $30.9 billion, growing by 28% and 27% at constant currency. Demand for Azure and other cloud services continued to grow, with revenue increasing by 40% and 39% at constant currency. The core infrastructure business (primarily from the largest customers) exceeded expectations, and Azure AI service revenue overall met expectations, although demand still exceeded supply this quarter despite increased online capacity. On-premises server business revenue grew by 1%, with no significant change at constant currency, and Windows Server 2025 transaction purchases exceeded expectations. At constant currency, segment gross profit grew by 20% and 19%, with gross margin percentage declining year-on-year due to AI investments being partially offset by improvements in Azure efficiency. Operating expenses grew by 4%, operating income increased by 27%, and operating profit margin was 43%, slightly down year-on-year, with increased AI investments largely offset by improvements in operational leverage More Personal Computing: Revenue of $13.8 billion, up 4%. Windows OEM and device revenue increased by 6% year-over-year, exceeding expectations, driven by strong demand before the end of support for Windows 10 and high inventory levels. Search and news advertising, excluding TAC, grew by 16% and 15%, respectively, mainly driven by sales growth and continued earnings from third-party partners. Revenue in the gaming sector decreased by 2% to 3% at constant currency. Xbox content and services revenue grew by 1%, remaining relatively unchanged at constant currency. Reminder: The timing of larger long-term Azure contract revenues is unpredictable, increasing quarterly volatility. Microsoft's cloud gross margin is approximately 66%, down year-over-year due to AI investments and the transition to Azure hybrid. Segment Market Guidance: Productivity and Business Processes: Expected revenue of $33.3 billion to $33.6 billion, growth of 13%-14%. M365 commercial cloud revenue is expected to grow by 13%-14% at constant currency, with trends remaining relatively stable, and ARPU growth driven by E5 and M365 co-pilot. M365 commercial product revenue growth is expected to be in the low to mid-single digits, potentially affected by revenue recognition dynamics. M365 consumer cloud revenue is expected to grow by about 20%. LinkedIn revenue is expected to grow by about 10%. Dynamics 365 revenue is expected to grow by 15%. Intelligent Cloud: Expected revenue of $32.25 billion to $32.55 billion, growth of 26%-27%. Azure revenue is expected to grow by about 37%, with demand still far exceeding available capacity. While accelerating the rollout of capacity, we balance Azure revenue growth, first-party applications, AI solution development, and the need for server replacements. Capacity is still expected to be constrained before the end of this fiscal year. On-Premises Servers: Revenue is expected to decline to low single digits. More Personal Computing: Revenue is expected to be between $13.95 billion and $14.45 billion. Windows OEM revenue is expected to decline to single digits. Device revenue is expected to decline year-over-year. Search and news advertising revenue growth is in the low double digits, with growth rates normalizing and declining quarter-over-quarter. Xbox content and services revenue is expected to decline in the low to mid-single digits year-over-year, partially offset by strong first-party performance against declines in subscription and hardware revenue. Demand signals such as bookings, RPO, and product usage are growing beyond expectations. We are investing in infrastructure, AI talent, and product innovation to expand our leadership position, focusing on delivering value to customers and achieving long-term sustainable revenue growth. With that said, let's move into the Q&A session. Q1 : Morgan Stanley analyst Keith Weiss: Great, thank you everyone for answering the questions, and congratulations on another outstanding quarter. If we look at Microsoft, this is two consecutive quarters of strong performance. When we think about this company a year ago or five years ago, we really see results that far exceed expectations. A 111% growth in commercial bookings, which almost no one could have predicted. However, the stock price performance has lagged the market. My question is: do you think this situation will change? I feel it may put pressure on the stocks. I noticed that AGI is still mentioned in your OpenAI agreement. So, Satya, when we talk about AGI or changes in application and computing architecture, do you see certain factors—whether AGI or otherwise—that might change Microsoft's strong position in the market and weaken that advantage? Is there anything you are concerned about as generative AI models continue to evolve? Chairman and CEO Satya Nadella: Thank you for your question, Keith. I think it can be answered from two perspectives. First, we are very pleased with the new agreement we reached with OpenAI, as it brings more certainty to all our intellectual property relationships and also involves the definition of AGI. Secondly, I think you touched on the core issue: how these AI systems are deployed in the real world and truly bring change and returns to customers and providers. Even if intelligent capabilities grow exponentially, the issue is that it remains uneven—people refer to it as "jagged intelligence." It may perform excellently on certain tasks, but the growth is not balanced. Therefore, what we need are systems—whether GitHub Agent HQ or M365 Copilot—should not be viewed as a single product, but rather as systems that smooth out these uneven capabilities and genuinely help enhance abilities. For example, in M365 Copilot, I can generate Excel spreadsheets. The good news is that Excel can understand OfficeJS formulas, and a spreadsheet generated by a good model is excellent. I can even enter agent mode, iterate on the model, and ensure it stays on track. After that, I can hand it over to an analyst agent, which will understand the data just like a data analyst understands Excel models. The core is that we need this type of structural approach. Even if the models are very powerful, we will remain in the "jagged intelligence" phase for a long time. In the most fundamental areas—GitHub, security, M365—we believe it is very valuable to build these systems as agents to help customers at the organizational level. By the way, this is also the goal we hope third-party customers will achieve in Foundry. Overall, I am very optimistic about the progress of AI. I believe that the AGI defined in the contract will not be realized anytime soon, but by building these systems, we can leverage the advancements in AI models to create tremendous value for customers. This is a question that requires deep understanding, and I am confident in our ability to make progress. Keith Weiss, Analyst: Great, that’s very helpful. Q2: Jefferies Analyst Brent Thill: Thank you, Amy. Regarding the surge in bookings, I think many people are a bit concerned about concentration risk. I noticed some contracts worth $100 million, but I don’t want to pry too much into the details. Can you tell us what you are seeing in the 51% RPO growth and over 110% booking growth? Does this give you confidence in the breadth and coverage of these transactions globally? Thank you.** Amy Hood, Executive Vice President and Chief Financial Officer: Thank you, Brent. Looking back at the RPO situation, we have a balance of nearly $400 billion, and we have been working hard to help everyone understand its breadth. It covers many products and a variety of customer sizes, and the cost balance is growing well. It is important to note that these contracts span multiple products. As Satya mentioned, we have done a lot of work in creating systems and investments. If you want to achieve this balance, it is more important to have a weighted average duration of two years. This means that most of the RPO will be realized in a relatively short time. Customers will only use it when they see value, which is why we continuously emphasize creating real value in AI platforms, solutions, applications, and systems. RPO has been established among many customers, and we are also pleased that OpenAI is part of it. Through these experiences, we have built leading systems for large-scale use that benefit all other customers. That is why we want to make the balance of RPO easier to understand and alleviate concerns about "whether it is outdated or whether the cycle is too long." At this scale, I think this is an excellent execution. Q3: Mark Moedler, Analyst at Bernstein Research: Thank you for the answer, and congratulations on your performance this quarter. Satya, Amy, I want to ask a common question from investors and the AI conference: How confident are you that software and even consumer internet businesses can monetize global investment? Or are we in a bubble? Amy, what factors will you pay attention to ensure that you do not overbuild to meet current demand? Amy Hood: I'll answer first, and Satya can add. The $400 billion short-term RPO shows that the demand for building infrastructure is very high. This is not an attempt at new business but rather the realization cycle of existing contracts. Short-term assets, such as GPUs and CPUs, typically match the contract duration, so when people consider risks, they need to be aware that asset life and contract cycles are aligned. We have not lacked GPUs or CPUs in the past few years; what we lack is the space and capacity to place them. We have spent a lot of time building this infrastructure and continue to invest in long-term assets (15 to 20 years) through financing leases. I am confident that we will efficiently utilize resources to meet demand growth. Product usage is increasing, new products are being released quickly, and customers will adopt them when they see value. We are very confident in usage patterns and booking volumes. Chairman and CEO Satya Nadella: Let me add one point. Looking ahead, there are two key things that affect capital allocation and R&D investment: Global scale token factory efficiency: We need a highly interchangeable fleet rather than building super-large data centers in a single region. Globally distributed data centers support inference, pre-training, post-training, reinforcement learning, etc., so interchangeability is crucial Fleet modernization: Annual procurement needs to follow Moore's Law while improving efficiency through software. For example, efficiency improvements of about 30% are seen with GPT-4o and GPT-5o. This applies to A100, GB200, and will also apply to GB300 in the future. Additionally, Amy mentioned that we have excellent agent systems in high-value areas, such as the Copilot system. In programming, the ARPU of Copilot is highly scalable compared to the ARPU of M365, similar to how cloud services expanded the server market in the past. The same applies to AI; while the M365 ARPU may be low, AI offers greater scaling opportunities. The tools business is similar; it was not a leading business in the past, but now programming could become one of the most scalable AI systems. The same goes for security and healthcare. On the consumer side, not only advertising but subscriptions also bring new opportunities. By integrating high-value agent systems with fleet efficiency and substitutability, we are confident in investing capital and R&D talent to seize opportunities. Analyst Mark Mader: That's great, thank you very much for these details. Q4: UBS Analyst Karl Keirstead: Thank you. This question is for Amy. Amy, I certainly don't want to lead you down a complicated accounting path, but regarding the investment in OpenAI, listed under other income, the amount is $4.1 billion. This amount is very large, and the audience may need more explanation. It far exceeds the amounts of other income in previous quarters, so it can't just be your share of losses in OpenAI. Can you clarify this, and what can we expect in subsequent quarters? Does it imply any accounting changes? Thank you. Amy Hood, Executive Vice President and Chief Financial Officer: The first quarter numbers were not affected by the new agreement, let me clarify that first. Secondly, this increased loss is entirely due to our share of OpenAI's losses recognized under the equity method. It is very clear that there are no other reasons, just the increase in OpenAI's losses. Karl Keirstead: Okay, got it, thank you. Amy Hood: Thank you, Karl. Q5: JP Morgan Analyst Mark Murphy: Thank you very much. We seem to be entering a new era where contracts signed by a few AI-native companies are very large, not only in absolute terms but sometimes relative to their own scale. For example, contract values in the hundreds of billions of dollars are 20 times their current revenue. From a philosophical perspective, how do you assess these companies' ability to fulfill their commitments? How do you consider setting protective measures against concentration on a single client? Satya Nadella: Perhaps I will answer first, and then Amy can add. Mark, this goes back to what I said earlier: first, build the assets themselves to make them highly substitutable; second, recognize the strength of our portfolio. We have third-party business as well as first-party business, with third-party business distributed between enterprises and digital-native companies. Maintaining balance is very important because digital-native companies are always early adopters, have a generation of blockbuster applications, and then gradually expand into enterprise adoption cycles From the timeline perspective, the balance of third-party clients will only increase. Initially, having first-party blockbuster applications is good because it can establish scale; if they are replaceable, this is crucial. You do not want to only provide hosting for digital-native companies but rather build replaceable systems. Our decision-making process reflects this: which say "yes," and which say "no." I hope everyone understands this by now. First-party business may be the part with the greatest leverage, not just blockbuster applications. The entire portfolio of assets gives us confidence again: maximizing the use of the fleet through the portfolio. Remember, data centers and other assets are long-term assets, with multiple update cycles for any hardware. Therefore, through careful consideration and building for a broad customer base, concentration risk can be mitigated. Amy Hood: To add from another angle, Satya has explained a lot: when considering concentration risk or customer delivery, you must remember that we have a very large and flexible fleet available for first-party, third-party, and commercial cloud. CPUs, GPUs, and storage devices will only be utilized when contracts are executed. Some large contracts have phased deliveries, so there is ample lead time to understand the status. We have always considered this in our RPO balance and explained it when releasing booking numbers and RPO balances. Mark Murphy: Thank you very much. Q6: Deutsche Bank analyst Brad Zelnick: Thank you very much, and first, congratulations on a great start. Amy, can you quantify or describe the impact on revenue in the case of Azure capacity shortages? I understand the entire industry has constraints, but is there a risk of workloads flowing to other platforms? How do you mitigate this? Amy Hood: Brad, that's a great question. It's hard to precisely quantify the quarterly revenue impact, but I can provide some insights: Azure may bear most of the revenue impact because the priority is on truly high-priority demands, such as the growth of M365 Copilot and Copilot chat usage, as well as the adoption of security features and GitHub momentum. This explains the priority order of resource allocation. We have been working to mitigate this, but Azure is indeed short. There are two other priorities: ensuring that the product teams and AI talent hired over the past year and a half have sufficient capacity, which is crucial for enhancing product performance and real-world usage. This does directly impact Azure, which is where priority allocation is focused. It's hard for me to give precise numbers, but the impact could be larger. Brad Zelnick: Okay, thank you. Q7: Goldman Sachs analyst Kash Rangan: Thank you very much. Amy, first congratulations to you. Previously, you mentioned being able to accelerate Azure growth while maintaining high efficiency margins, and you did it, congratulations. **Question for Satya: Regarding the "elephant in the room," following up on Keith Weiss's question, there are rumors that another hyperscale cloud company has intervened and taken business that should have belonged to Microsoft. I would like to know your thoughts on the following criteria: is it about a certain amount of business execution, or is it a broader consideration? People may not fully understand the terminal value of Microsoft's assets at the end of the contract, while you have a complete stack and multiple monetization avenues, such as databases, foundries, and you are a platform company rather than just a hyperscale cloud company. Some may also say that you have allowed another company to enter and occupy a large business segment over a four to five-year cycle from scratch. Satya Nadella: Thank you, Kash. The core principle is to build a globally replaceable fleet, applicable to third parties, first parties, and research. When demand is too concentrated (skewed by customer, location, or type), it does not align with long-term goals. Components like AI accelerators, computing, and storage also need to be balanced; if demand is all concentrated in one area, that is not the business we want to operate in long-term. Even for third parties, there must be a balance with first-party business, as the profit stack is different. At the same time, funding must be provided for R&D and model capabilities, which will determine differentiation in the long run. I consider all these factors to ensure we accept the demand we want while saying "no" to demands that do not align with long-term interests. This is our decision-making approach, and I am very satisfied with it. Every time I say "no," I feel better the next day