Microsoft acknowledged that rising prices of hardware components such as GPUs will push up annual capital expenditures by approximately $25 billion, and the tight supply of cloud computing power is expected to persist through at least the end of 2026. However, the CFO noted that the company's AI business profit margin is better than during the previous cloud transformation phase, a point that may be underestimated by the market. Meanwhile, Microsoft announced a shift in the business model for GitHub Copilot, linking pricing to usage and value.
Microsoft held its Q3 earnings call for the fiscal year 2026 on April 29 Eastern Time, with CEO Satya Nadella and CFO Amy Hood in attendance to deliver remarks.
During the call, Nadella was straightforward: 'This has been a record-breaking third quarter.' He stated that Microsoft Cloud revenue exceeded $54.5 billion, representing a 29% year-over-year increase, while the annualized revenue run rate (ARR) for AI surpassed $37 billion, marking a 123% year-over-year growth.
Nadella characterized the current moment as a pivotal juncture in technological history: 'We are at the beginning of one of the most significant platform shifts, where agents will become the dominant workload and transform value creation across the entire economy.' Following the call, Microsoft shares rose slightly by 0.3% in after-hours trading.
Hardware Price Increases and GPU Shortages: $190 Billion for the Year, $25 Billion Attributable to Rising Prices
This represents the most closely watched incremental information from the call.
Hood disclosed that capital expenditures in Q4 are expected to exceed $40 billion, a significant sequential increase, with approximately $5 billion attributable to component price hikes. Capital expenditures for this quarter totaled $31.9 billion, with about two-thirds allocated to short-term assets such as GPUs and CPUs.
For the full calendar year 2026, Microsoft anticipates capital expenditures to reach approximately $190 billion, with about $25 billion stemming from higher component pricing.
Hood admitted that despite accelerating delivery efforts, supply constraints will persist throughout the year: 'Even with these additional investments and our efforts to expedite the deployment of GPUs, CPUs, and storage, we expect to remain supply-constrained throughout 2026.'
She further elaborated: 'Demand remains broad-based and continues to outpace supply.' Nevertheless, Hood remained optimistic about yields: 'We are confident in the returns on these investments, driven by stronger demand signals, increasing product usage, and the efficiency improvements we have implemented across our platform.'
Profit Margin: AI Business Outperforms During Cloud Transformation Phase
In response to the question of whether 'high investment in AI squeezes profits,' Hood provided an assessment that exceeded market expectations:
"The profit margin of our AI business is actually better than what we observed during our cloud transformation phase. This might be an area where the market had underestimated us."
She attributed this resilience to several factors: consumption and usage-based pricing models that better capture AI value; royalty-free IP usage rights from OpenAI collaboration; cost optimization of infrastructure through proprietary chips (Maia 200, Cobalt); and ongoing improvements in software and hardware efficiency.
Management expects a year-over-year improvement of approximately one percentage point in operating profit margin for the full fiscal year — a result achieved despite continuous heavy investments in AI infrastructure throughout the year and including a one-time cost of approximately $900 million for voluntary retirement programs.
Copilot Acceleration: Over 20 Million Seats, Growth Rate of 250%
Nadella disclosed that M365 Copilot paid seats have exceeded 20 million, with net new seat additions growing 250% year over year this quarter, marking 'the fastest growth rate since product launch.'
Significant expansion among large clients was also evident: Accenture now has more than 740,000 seats, making it the largest single client to date; Bayer, Johnson & Johnson, MERCEDES-BENZ GROUP AG UNSP ADR EACH REP 0.25 ORD SHS, and Roche have all committed to purchasing 90,000 seats or more.
Nadella used an intuitive comparison to describe usage intensity: "The depth of weekly user engagement is now on par with Outlook — an increasing number of users are making Copilot a daily habit." He also mentioned that per-user Copilot query volumes increased by nearly 20% quarter over quarter, and monthly active usage of first-party AI Agents has grown sixfold year-to-date.
Business Model Transformation: From 'Per Seat' to 'Seat + Usage'
At the business model level, both Nadella and Hood clearly pointed out that the pricing logic is shifting from purely 'per seat' to 'seat plus usage.'
Nadella stated: 'Any of our per-user businesses—whether productivity, programming, or security—will transition into a per-user plus per-usage model. This is the most accurate way to understand it.'
This quarter, Microsoft announced that GitHub Copilot will transition to a usage-based and value-driven pricing model, effective June 1.
He also addressed the core question of 'who will ultimately pay for these AI expenses': 'Ultimately, the money will come from an enterprise’s evaluation and outcomes—where agents, representing or collaborating with users, create real value. Whether it's customer service, personal productivity, team efficiency, or a specific business process—a certain cost is decreasing, or a certain revenue is increasing. That is what drives usage.'
Azure Growth Rate: Q4 Guidance at 39%-40%, with Expectations of Mild Acceleration in the Second Half
Azure’s revenue growth this quarter reached 40% (39% on a constant currency basis), surpassing previous expectations. Hood explained that this was primarily due to 'delivering capacity earlier in the quarter, which allowed consumption of both AI and non-AI services to increase.'
For the next quarter (Q4), Microsoft guided Azure’s growth rate to be 39% to 40% (on a constant currency basis), roughly flat compared to the current quarter.
More noteworthy is the longer-term outlook: Hood stated that despite supply constraints, 'we expect Azure’s growth rate in the second half of calendar year 2026 to show mild acceleration compared to the first half.'
Commercial Bookings: Excluding OpenAI, Growth Reached 7%; RPO Surpassed $62.7 Billion
Hood disclosed that excluding the impact of OpenAI, commercial bookings grew by 7%; including OpenAI, they declined by 4% to 6% year-over-year.
Remaining Performance Obligations (RPO) in commercial contracts increased to $627 billion, representing a 99% year-over-year growth including OpenAI, with a weighted average contract term of approximately 2.5 years. The portion expected to be recognized as revenue within the next 12 months grew by 39% year-over-year, while the portion beyond 12 months surged by 138%.
Hood explained to investors the rationale behind the short-term pressure on order figures: 'As we transition from our historical per-seat business model to a seat-plus-consumption model, the nature of orders will also evolve. Not all of it will flow through the traditional order logic; part of it will be billed directly based on usage.'
Reconstruction of the OpenAI Agreement: Royalty-Free IP, Revenue Sharing Extended to 2030
During the earnings call, Microsoft publicly addressed the adjustments to its agreement with OpenAI for the first time.
Nadella stated, 'Overall, we are pleased with our partnership with OpenAI. We have a cutting-edge model, royalty-free terms, full IP rights, and the ability to utilize it up to the 32nd generation. We fully intend to maximize this opportunity.'
Hood added two key financial points: First, the revenue-sharing arrangement will continue until 2030, providing us with significant predictability and a real advantage. Second, Microsoft has eliminated royalty payments to OpenAI—IP is acquired royalty-free, removing the RF revenue share owed to OpenAI.
Outlook for the Next Fiscal Year: Double-Digit Growth in Revenue and Operating Profit, Continued Reduction in Workforce
Hood provided clear guidance for the upcoming fiscal year (FY27), forecasting double-digit growth in both revenue and operating profit.
The growth rate of operating expenses will remain in the mid-to-high single digits, while the company's workforce will decrease year-over-year. She emphasized that this reflects the company’s ongoing efforts to enhance operational efficiency, operate at a faster pace, and improve agility.
Summary of Microsoft's Third Quarter Earnings Call for Fiscal Year 2026
Meeting Date: April 29, 2026 Company Name: Microsoft Meeting Type: Fiscal Year 2026 Third Quarter Earnings Conference Call
I. Presentation Session
Moderator (Operator):
Welcome to Microsoft’s Fiscal Year 2026 Third Quarter Earnings Conference Call. All participants are currently in listen-only mode, and a question-and-answer session will follow the formal presentation. This conference call will be recorded in its entirety. It is now my pleasure to hand over the microphone to Jonathan Neilson, Vice President of Investor Relations. Please begin your remarks.
Jonathan Neilson (Vice President of Investor Relations):
Good afternoon, and thank you for joining us today. Presenting alongside me on this conference call are Chairman and Chief Executive Officer Satya Nadella, Chief Financial Officer Amy Hood, Chief Accounting Officer Alice Jolla, and Deputy General Counsel and Corporate Secretary Brian DeFoe.
On the Microsoft Investor Relations website, you can review the earnings press release and financial summary slides for this quarter. These materials are intended to supplement the statements made during today's conference call and include reconciliations of GAAP to non-GAAP financial metrics. A more detailed outlook slide will also be published on the Microsoft Investor Relations website simultaneously with the earnings guidance provided during this call.
This conference call will discuss several non-GAAP measures. The non-GAAP financial metrics provided should not be viewed as substitutes or superior to financial performance measures prepared in accordance with GAAP; rather, they are presented as supplementary information to assist investors in gaining a more comprehensive understanding of the company's third-quarter performance, as well as the impact of specific items and events on financial results.
Unless otherwise stated, all growth comparisons mentioned during this conference call are relative to the same period last year. Where data is available, we will also provide growth rates at constant currency to evaluate the actual performance of business segments excluding the effects of exchange rate fluctuations. If the growth rate at constant currency matches the reported growth rate, only the reported growth rate will be cited. Immediately following the conclusion of the conference call, we will post the prepared remarks on our official website until the full transcript is released.
This conference call will be webcast live and recorded. Should you ask a question, your inquiry will be included in the live transmission, transcript, and future recordings. You may replay this conference or review the transcript on the Microsoft Investor Relations website.
This conference call will include forward-looking statements, which are predictions, expectations, or other expressions regarding future events. These statements are based on current expectations and assumptions and are subject to risks and uncertainties. Actual results may differ materially due to factors discussed in today's earnings press release, comments made during this conference call, and risk factors outlined in our annual report on Form 10-K, quarterly reports on Form 10-Q, and other filings submitted to the U.S. Securities and Exchange Commission. We do not undertake any obligation to update any forward-looking statements.
Next, I will hand the microphone over to Satya.
Satya Nadella (Chairman and Chief Executive Officer):
Thank you very much, Jonathan. This quarter set a historical record, driven by the continued strong performance of Microsoft’s cloud business—cloud revenue exceeded $54 billion, representing a 29% year-over-year growth. Our AI business achieved an annualized recurring revenue (ARR) of over $37 billion, growing 123% year over year.
We are at the starting point of one of the most profound platform shifts. As AI agents become widely adopted and gradually emerge as the dominant workload, the entire technology stack will undergo a profound transformation, extending time dimensions and redefining value creation across the economic system.
To seize this opportunity, we are advancing with full force around two key priorities:
First, building world-class cloud and AI infrastructure to meet the demands of the agent computing era;
Second, creating high-value agent systems in core areas such as productivity, programming, and security.
These two layers reinforce each other. We are focused on delivering competitive value and differentiated experiences for our customers, helping them maximize business outcomes.
At the infrastructure level
We are optimizing every layer of the technology stack, covering data center design, chips, system software, model architecture, and their optimization, and have achieved significant operational results:
Since the beginning of the year, we have reduced the downtime of new GPUs in key regions by nearly 20%.
The Fairwater data center in Wisconsin went online six weeks ahead of schedule, allowing us to recognize revenue earlier.
Thanks to the synergy of hardware and software optimization, the inference throughput of Copilot's most frequently used model has increased by 40%.
This quarter, we added over 1 gigawatt (GW) of capacity, with overall scale expected to double within two years.
We are actively accelerating capacity expansion based on demand signals and have announced new data center investments across four continents.
Meanwhile, we continue to advance fleet modernization through proprietary technological innovations alongside the latest achievements from NVIDIA and AMD. Millions of servers are equipped with our self-developed networking, security, and virtualization chips (including Azure Boost), as well as our self-developed CPUs and accelerators.
The Maia 200 AI Accelerator delivers more than a 30% improvement in compute power per dollar compared to the latest silicon chips in the fleet and is now operational in data centers in Iowa and Arizona.
The Cobalt server CPU has been deployed in nearly half of the data center regions, supporting large-scale workloads for customers such as Databricks, Siemens, and Snowflake. As major clients expand their AI deployments, an increasing number of customers are choosing to run cross-platform workloads on Cobalt, prompting us to significantly increase Cobalt supply to meet demand.
Agent Application Platform Layer
Above the infrastructure layer lies the intelligent agent application platform layer. At its core is model selection — we offer the most extensive range of models among all hyperscale cloud service providers, allowing customers to freely choose from options such as OpenAI, Anthropic, and open-source models based on workload requirements.
More than 10,000 customers on the Foundry platform have utilized more than one model.
5,000 customers have adopted open-source models.
The number of customers using Anthropic and OpenAI models has grown by 2x quarter-over-quarter.
Bayer is integrating multiple models on Foundry to build its own internal intelligent agent platform, with over 20,000 monthly active users. Currently, more than 300 customers are expected to process over 1 trillion tokens through Foundry this year, representing a 30% growth quarter-over-quarter.
In terms of self-developed models, we focus on enhancing the differentiated capabilities of high-value Copilots and agents while reducing operating costs (COGS).
We launched MAI-Transcribe-1, an advanced speech-to-text model.
We introduced MAI-Image-2, one of the world’s leading image generation models.
These models have been deployed in first-party scenarios such as Bing Image Generation and PowerPoint. Early data indicates that Transcribe-1 improves GPU efficiency by 67%, while Image-2 achieves an efficiency gain of up to 260%. We have also made MAI models available to commercial clients like Shutterstock and WPP for the first time via Foundry.
We have made progress in innovating based on OpenAI's intellectual property, including multi-step retrieval optimization in Work IQ and Copilot, as well as adaptive reasoning for intent complexity in the Researcher function — significantly improving accuracy while drastically reducing latency.
Enterprise Data and Context Layer
Above that lies the enterprise data and context layer, encompassing Fabric, Foundry, Microsoft 365, and the security graph. We are building a unified Organizational Intelligence (IQ) layer on this foundation.
Thousands of enterprises are already accessing contextual information across these IQ layers. As AI usage grows, the context layer continues to expand, creating a virtuous flywheel effect that constantly enhances the baseline quality, relevance, and effectiveness of each intelligent agent, making our IQ layer an unparalleled context engine for organizational intelligence.
In terms of database business, this quarter saw accelerated growth on a quarter-over-quarter basis—Cosmos DB alone achieved a 50% year-over-year revenue increase driven by AI application workloads.
The number of paid Fabric customers reached 35,000, representing a 60% year-over-year increase.
The volume of data in Fabric OneLake data lake grew nearly fourfold year over year.
More than 15,000 customers are using both Foundry and Fabric, marking a 60% year-over-year increase—more enterprises are connecting agents to Fabric's integrated real-time operations, analytics, and unstructured data.
Significant progress has also been made in Foundry’s agent services, enabling customers to build stateful agents that operate persistently across time boundaries, coordinate tools and models, and form closed-loop evaluations and improvements within long-term workflows.
In the low-code/no-code tools space, nearly 90% of Fortune 500 companies have built active agents through Copilot Studio. Growth in the Copilot Credit Consumptive billing model has nearly doubled, with sequential growth approaching 2x—more customers are extending Copilot capabilities with customized agents to align with their business processes.
With Agent 365, we provide a control plane that extends an enterprise’s existing governance, identity, security, and management frameworks to the agent level. Currently, tens of thousands of companies are managing tens of millions of agents through Agent 365, and this momentum is expected to grow significantly as demand for identity governance, security, and related tools continues to rise.
High-value agent systems
We are upgrading the Copilot series from synchronous assistants to asynchronous collaborators, enabling them to perform long-term tasks in key business areas.
Knowledge work domain: This quarter, Microsoft 365 Copilot seat additions reached a new historical record, with year-over-year growth of 250%, marking the fastest growth rate since its launch. Growth has continued to accelerate on a quarter-over-quarter basis, and paid seats have exceeded 20 million. The number of customers holding more than 50,000 seats has quadrupled year over year; Accenture holds over 740,000 seats, making it the largest Copilot customer to date; Bayer, Johnson & Johnson, Mercedes-Benz Group AG, and Roche have all committed to purchasing 90,000 or more seats.
Copilot brings unique value in workplace scenarios—almost every task relies on organizational context. Work IQ anchors Copilot's responses within the full context of the organization, encompassing personnel, functions, documents, and communication content, all while remaining within the company’s secure boundaries. The volume of data supporting Work IQ now exceeds 17 exabytes (EB), growing by 35% year over year, with billions of emails, documents, and chat records added daily, hundreds of millions of Teams meetings, and millions of SharePoint sites.
As Copilot adoption increases, the dialogues and generated content between Copilot and agents feed back into Work IQ, enriching the contextual layer even further.
Over the past year, we have introduced more than 625 feature updates in Microsoft 365 Copilot, representing a 50% increase year over year. Key updates include:
Default support for multi-model access in chats, with intelligent automatic routing;
The "Review and Recommend" agent can utilize multiple models collaboratively to generate optimal responses;
Starting last week, agent mode has become the default experience for Copilot in Word, Excel, and PowerPoint;
The new Cowork feature allows users to delegate tasks to Copilot in innovative ways.
The aforementioned innovations have driven Copilot usage intensity to record highs: monthly active usage of first-party agents has increased sixfold year-to-date; the number of Copilot queries per user has grown by nearly 20% quarter-over-quarter; weekly activity now matches that of Outlook, with an increasing number of users incorporating Copilot into their daily work routines.
Business application domain: Customers are transitioning from traditional seat-based models to a hybrid model of "seats + consumption." Customer service is at the forefront of this shift, with nearly 60% of service clients already purchasing usage-based credits. For instance, HSBC uses pre-configured agents in Dynamics 365 to handle customer inquiries across products, markets, and regulatory requirements, reducing issue resolution time by over 30%. Additionally, our AI-driven agent product introduced in LinkedIn Talent Solutions, which helps recruiters automate time-consuming tasks such as searching, screening, and drafting messages, has achieved an annualized revenue run rate exceeding $450 million.
Developer domain: GitHub is experiencing unprecedented growth driven by the rapid adoption of AI-powered programming, and we are actively scaling to meet this demand.
Approximately 140,000 organizations are now using GitHub Copilot, with enterprise subscription numbers nearly tripling year-over-year.
Most users employ multiple models.
Usage of the GitHub Copilot CLI has nearly doubled month-over-month.
Earlier this week, we announced that GitHub Copilot will transition to a usage-based pricing model effective June 1, aligning pricing with actual usage and value delivered.
Security domain: As AI compresses the window between vulnerability discovery and exploitation, the dynamics of cybersecurity have fundamentally shifted. To immediately assist customers in mitigating risks, we synchronize the rollout of AI-detected vulnerability updates with Defender protections. We are also advancing the productization of new multimodal AI-driven scanning tools.
The number of Security Copilot customers has doubled year-over-year.
This quarter, the data security classification agent independently handled over 2 million distinct alerts.
To date, 35 billion Copilot interactions have been audited through Purview, representing a sevenfold increase year-over-year.
Consumer business: We are undertaking foundational work in Windows, Xbox, Bing, and Edge to regain user trust and deepen engagement. Recently, we have focused on core elements, enhancing product quality to better serve our key users.
Specific progress includes:
In Windows: performance improvements for low-memory devices, optimization of the update experience, and a focus on core functionalities most important to users;
In Xbox: renewed focus on the core gaming community, with last week’s adjustment to Game Pass serving as a positive response to user feedback;
The number of monthly active Windows devices has exceeded 1.6 billion;
Edge browser market share has grown for 20 consecutive quarters;
Bing's monthly active users surpassed 1 billion for the first time;
LinkedIn now has 1.3 billion members, with continued growth in meaningful interactions, serving as a leading B2B sales and advertising channel for small, medium, and large enterprises;
Xbox achieved record highs this quarter in both monthly active users and gaming streaming hours;
Microsoft 365 now has nearly 95 million consumer subscribers, and early satisfaction data shows that user satisfaction is improving after setting the agent mode as the default.
In summary, we are also actively advancing the transformation of our own working methods. Our guiding goal remains unchanged: to create value for customers with the highest quality and first-class innovation. This is also the source of my confidence in the company’s and customers’ growth prospects in the next phase.
Next, I will hand over the microphone to Amy to introduce the financial performance and outlook.
Amy Hood (Executive Vice President and Chief Financial Officer):
Thank you, Satya. Good afternoon, everyone.
This quarter, our performance in revenue, operating profit, and earnings per share exceeded expectations, driven by strong demand and efficient execution.
Overall Performance
As Satya mentioned, the annualized revenue run rate of our AI business exceeded $37 billion this quarter, representing a year-over-year increase of 123%, with the pace of innovation continuing to accelerate.
Revenue for this quarter was $82.9 billion, an increase of 18% year-over-year, or 15% in constant currency.
Gross profit in dollar terms increased by 16% year-over-year, or 13% in constant currency.
Operating profit increased by 20% year-over-year, or 16% in constant currency;
Earnings per share (EPS) amounted to $4.27, up 21% year-over-year, or 18% in constant currency (excluding the impact of investment in OpenAI).
The foreign exchange impact was largely in line with corporate-level guidance. The overall gross margin stood at 68%, reflecting a year-over-year decline primarily due to ongoing investments in AI infrastructure and rising usage of AI products, partly offset by continuous efficiency improvements in businesses such as Azure and M365 commercial cloud.
Operating expenses grew by 9% year-over-year, or 8% in constant currency, mainly driven by sustained investments in AI, including computing power, talent, and data for R&D purposes to support full-line product development. This quarter’s growth was also influenced by a relatively low base from the prior year, particularly in sales and marketing expenses as well as general administrative costs.
The operating profit margin rose slightly year-over-year to 46%. The company's total headcount declined year-over-year, reflecting our strategic focus on building an efficient and agile workforce.
Excluding the impact of investment in OpenAI, other net income amounted to $961 million, primarily driven by investment gains, partially offset by foreign exchange translation losses.
Capital expenditure
Capital expenditure totaled $31.9 billion, declining sequentially, primarily due to normal fluctuations in cloud infrastructure construction and differences in the timing of leased asset deliveries. Approximately two-thirds of this quarter’s capital expenditure was allocated to short-term assets (mainly GPUs and CPUs), with the remainder directed toward long-term assets expected to support commercial operations over the next 15 years or more.
Total lease financing for the quarter amounted to $4.7 billion, mainly for large-scale data center sites. Cash payments for fixed assets and equipment were approximately $30.9 billion, broadly in line with capital expenditure (lease financing impacts and timing differences between goods receipt and payment largely offset each other).
Cash flow
Operating cash flow was $46.7 billion, a year-over-year increase of 26%, primarily driven by strong cloud billing and collections, partly offset by higher operating lease payments. Free cash flow amounted to $15.8 billion, reflecting elevated levels of capital expenditures.
This quarter, $10.2 billion was returned to shareholders through dividends and share repurchases.
Commercial Business Performance
In terms of commercial orders, excluding the impact of OpenAI, there was a year-over-year increase of 7%, supported by robust execution in core annuity sales. Including OpenAI’s influence, orders declined by 4% year-over-year, or 6% at constant currency.
Commercial Remaining Performance Obligations (RPO) grew by 26% year-over-year, consistent with historical seasonal trends. Including OpenAI RPO, it reached $627 billion, representing a 99% year-over-year increase, with a weighted average contract term of approximately 2.5 years. About 25% is expected to be recognized as revenue within the next 12 months, up 39% year-over-year; the portion recognized beyond 12 months increased by 138%.
Microsoft Cloud revenue was $54.5 billion, growing 29% year-over-year, or 25% at constant currency, reflecting strong demand for the Azure platform and first-party AI applications and services. Microsoft Cloud gross margin was slightly better than expected at 66%, declining year-over-year mainly due to ongoing investments in AI, partially offset by efficiency improvements.
Segment Performance
Productivity and Business Processes segment: Revenue was $35 billion, an increase of 17% year-over-year, or 13% at constant currency.
M365 Commercial cloud revenue grew 19% year-over-year, or 15% at constant currency, surpassing expectations – robust execution and continuous product quality enhancements accelerated Copilot seat additions this quarter, with paid seats exceeding 20 million. Average revenue per user (ARPU) growth was driven by E5 and M365 Copilot. M365 Commercial paid seats increased 6% year-over-year, expanding across all customer segments, primarily among small and medium enterprises and frontline worker groups.
M365 Commercial Products revenue grew 1% year-over-year, but declined 3% at constant currency, showing a sequential slowdown primarily due to the normalization of one-time purchases of Office 2024 as expected.
M365 consumer cloud revenue grew by 33% year-over-year, or 29% in constant currency, again driven by ARPU growth. M365 consumer subscriptions increased by 7%.
LinkedIn revenue grew by 12% year-over-year, or 9% in constant currency, with all business lines achieving growth.
Dynamics 365 revenue grew by 22% year-over-year, or 17% in constant currency, continuing to gain market share across workloads. Order growth was impacted by weaker renewals as customers balance spending between traditional seat-based models and emerging 'seat-plus-consumption' models.
Segment gross profit in USD terms grew by 18% year-over-year, or 13% in constant currency, with a slight improvement in gross margin primarily driven by efficiency gains, partly offset by ongoing investments in AI, including the impact of Copilot adoption and increased usage.
Operating expenses grew by 11% year-over-year, or 9% in constant currency, mainly due to the previously mentioned shared R&D investments in AI and increased Copilot advertising spend, reflecting a low base effect from the prior year. Operating profit grew by 21% year-over-year, or 14% in constant currency, with operating margin improving to 60% year-over-year.
Intelligent Cloud segment: Revenue reached $34.7 billion, growing by 30% year-over-year, or 28% in constant currency.
Azure and other cloud services revenue grew by 40% year-over-year, or 39% in constant currency, further accelerating on top of last year's growth, surpassing expectations. Capacity delivery ahead of schedule this quarter boosted consumption for both AI and non-AI services. Strong demand continued to exceed available capacity across workloads, customer segments, and regions.
On-premises server business revenue experienced a slight year-over-year increase, but declined by 3% in constant currency as customers continue migrating to cloud services.
Segment gross profit in USD terms grew by 19% year-over-year, or 18% in constant currency; gross margin declined year-over-year primarily due to ongoing AI investments and increased GitHub Copilot usage, partly offset by Azure efficiency improvements. Operating expenses grew by 9% year-over-year, or 7% in constant currency, mainly driven by the aforementioned shared R&D investments in AI. Operating profit grew by 24% year-over-year, or 23% in constant currency, with an operating margin of 40%.
More Personal Computing segment: Revenue amounted to $13.2 billion, declining by 1% year-over-year, or 3% in constant currency.
Windows OEM and device revenue decreased by 2% year-over-year, or 3% at constant currency. Windows OEM experienced modest growth and outperformed expectations as OEMs and channel partners continued to stockpile inventory amid rising memory prices.
Search advertising revenue (excluding traffic acquisition costs) increased by 12% year-over-year, or 9% at constant currency, primarily driven by higher search volume and increased revenue per search for Edge and Bing.
Gaming revenue declined by 7% year-over-year, or 9% at constant currency. Xbox content and services revenue fell by 5% year-over-year, or 7% at constant currency, due to a high base effect from strong first-party content performance in the same period last year.
Segment gross profit grew by 6% in dollar terms year-over-year, or 4% at constant currency; gross margin improved year-over-year, primarily driven by a shift in sales mix toward higher-margin businesses. Operating expenses increased by 7% year-over-year on a low base from the previous year, or 6% at constant currency, mainly due to impairment charges and related expenses in the gaming business and aforementioned shared R&D investments. Operating profit grew by 4% year-over-year, or 1% at constant currency, with operating margin improving to 28% year-over-year.
Outlook for the Fourth Quarter
Unless otherwise specified, the following outlook is denominated in U.S. dollars.
Based on current exchange rates, foreign currency is expected to have a positive impact of approximately 1 percentage point on revenue growth for both the "Productivity and Business Processes" and "More Personal Computing" segments, with no significant impact on the "Intelligent Cloud" segment. The overall impact on total company revenue is expected to be less than 1 percentage point. Foreign currency is expected to increase cost of goods sold (COGS) growth by about 1 percentage point, with no impact on operating expense growth.
Commercial Business: Excluding the impact of OpenAI, commercial bookings are expected to achieve healthy growth on top of expiring contracts, with solid execution in core annuity sales but facing a significantly higher base from the same period last year. Microsoft Cloud gross margin is projected to be around 64%, declining year-over-year, primarily due to ongoing AI investments and increased usage of GitHub Copilot. This week, we announced a business model transition for GitHub Copilot, linking pricing to usage and value, effective June 1 of this year.
Segment Guidance:
Productivity and Business Processes Segment: Revenue is expected to range between $37 billion and $37.3 billion, representing year-over-year growth of 12% to 13%. M365 Commercial Cloud adjusted revenue is projected to grow 15% to 16% at constant currency; excluding the 2-percentage-point impact from revenue recognition in the same period last year, reported growth at constant currency is expected to be 13% to 14%. Building on the momentum of Copilot in the third quarter, net paid seats are expected to increase sequentially, driving continued ARPU growth. M365 Commercial product revenue is expected to achieve mid-single-digit growth, facing a high base effect from stronger-than-expected one-time Office 2024 purchases in the same period last year. M365 Consumer Cloud revenue growth is projected to be in the low 20% range, moderating sequentially as comparisons begin against last year’s price increases, with growth again driven by both ARPU and subscription volume. LinkedIn revenue is expected to grow by approximately 10%. Dynamics 365 revenue growth is projected to be in the low double digits, moderating sequentially due to a high base from the same period last year and the aforementioned order trends.
Intelligent Cloud Segment: Revenue is expected to range from USD 37.95 billion to USD 38.25 billion, representing a year-over-year growth of 27% to 28%. Regarding Azure, we remain focused on accelerating capacity delivery and improving fleet efficiency. Fixed-currency growth for the fourth quarter is projected at 39% to 40%, facing a strong base effect from accelerated growth in the same period last year. Broad and sustained customer demand continues to outpace supply, and we are striving to balance the allocation of new capacity—prioritizing high-return areas such as Azure, first-party applications, R&D, and the replacement of legacy servers. Revenue from on-premises server business is anticipated to decline by a single-digit percentage as customers continue migrating to the cloud.
More Personal Computing Segment: Facing strong base effects from the prior year, complex dynamics in the PC market influenced by memory pricing, and a strategic shift toward delivering quality and value to consumers, revenue is forecasted to be between USD 11.75 billion and USD 12.25 billion. Windows OEM revenue is expected to decline by a high-teens percentage, with approximately 6 percentage points attributable to the high base from the end of Windows 10 support last year, another 6 percentage points due to anticipated inventory drawdowns this quarter, and an additional 6 percentage points stemming from market contraction caused by rising PC prices driven by increased memory costs. The potential outcome range is wider than usual. Consequently, Windows OEM and device revenue is projected to decline in the mid-to-high teens. Search advertising revenue (excluding traffic acquisition costs) is expected to grow in the high single-digit range, driven by revenue per search and search volume, with Bing and Edge continuing to gain market share. Xbox content and services revenue is anticipated to decrease in the low double-digit range, reflecting a strong base from robust first-party content performance last year and recent price adjustments for Xbox Game Pass focusing on value delivery. Hardware revenue is projected to decline year over year.
Company Overall: Revenue is expected to range from USD 86.7 billion to USD 87.8 billion, representing year-over-year growth of 13% to 15%, with accelerated growth in commercial businesses partly offset by consumer business performance. The fourth-quarter outlook for cost of goods sold (COGS) and operating expenses includes a one-time cost of approximately USD 900 million related to the recently announced voluntary retirement program. As a result, COGS is projected to be between USD 29.4 billion and USD 29.6 billion, growing 22% to 23% (including approximately USD 350 million from the retirement program), while operating expenses are forecasted to be between USD 19.3 billion and USD 19.4 billion, increasing by about 7% (including approximately USD 550 million from the retirement program). Despite sustained investments throughout the year to expand AI platform, application, and service capacity and accounting for the aforementioned one-time costs, operating margin for the full year (FY26) is expected to improve by approximately 1 percentage point year over year.
Excluding the impact of OpenAI investment, net other income and expenses are projected to be approximately negative USD 100 million (interest income will be exceeded by interest expenses, including those related to data center financing leases). The adjusted effective tax rate for the fourth quarter is forecasted to be around 19%.
Capital Expenditures: Expected to exceed USD 40 billion, up sequentially, including the impact of approximately USD 5 billion in component price increases and volatility from finance lease accounting recognized entirely in the current period upon lease commencement. The proportion of short-term assets is expected to remain broadly in line with the third quarter.
For the full year 2026, capital expenditures are projected to be approximately USD 190 billion, including the impact of about USD 25 billion in component price increases. Given stronger demand signals, increasing product usage, and efficiency gains achieved at the platform level, we are confident in the returns from these investments. Nevertheless, capacity constraints are expected to persist through at least 2026. Despite these constraints and the need to continuously balance the allocation of new supply, Azure's growth in the second half of calendar year 2026 is anticipated to show moderate acceleration compared to the first half.
Outlook for the Next Fiscal Year
First, we will continue evolving our operational model to enhance execution speed and flexibility, with headcount expected to decline year over year. Operating expense growth is forecasted to be in the mid-to-high single-digit range, reflecting ongoing R&D investments, including in compute, data, and talent for AI, to accelerate product innovation.
Second, it is important to note that we will face high base effects from the prior year, including factors such as the end of Windows 10 support, elevated OEM inventory levels, and increased one-time purchases of Office and server products.
Third, we remain focused on building platforms that enable customers to build and run AI solutions while continuously innovating in first-party AI applications and services. As a result, FY27 is expected to deliver double-digit revenue and operating profit growth once again.
In summary, as we enter the final quarter of this fiscal year, we remain committed to driving continuous innovation to help our customers create new business value.
Next, we will move into the Q&A session. Jonathan, please begin.
II. Q&A Session
Moderator (Operator):
We will now begin the question-and-answer session. (Instructions) The first question comes from Keith Weiss at Morgan Stanley. Please go ahead with your question.
Keith Weiss (Morgan Stanley):
Thank you for the opportunity to ask a question and congratulations on another very strong quarter. The data on Microsoft 365 Copilot is particularly impressive, surpassing most expectations.
I’d like to ask a macro question about demand. We have been discussing robust demand, which is evident in CIO surveys and clearly reflected in your business performance.
In the short term, could you discuss how this demand translates into commercial orders and how this conversion process is evolving? You mentioned different contract cycles between seat-based and consumption models, which might have an impact, along with considerations around renewal bases.
In the long term, perhaps Satya could comment on what continues to sustain this demand. In other words, who is ultimately paying for all of this? While we see enthusiasm for Microsoft in CIO surveys, overall IT spending forecasts are not increasing significantly, nor is GDP growth showing notable acceleration. So, where is this funding coming from, and when might we see evidence of these sources? Thank you.
Amy Hood:
Let me first address the first part of your question—regarding how these models impact orders.
This is a very important question. On the order side, normal cyclical factors have always been present, including contract expiration cycles or the signing of large multi-year Azure commitments, which have historically shown some volatility. But if we look at it from a broader perspective—which aligns with the deeper intent of your question, and Satya will also elaborate—we need to rethink what it means when the business model has long been based on per-seat billing, and suddenly, the way work gets done and efficiency improves shifts to 'one employee plus one AI agent.'
In my view, this means that the business model will transition to a combination of 'per-license plus per-usage,' and the scope of this model will be far broader than people might expect. It also means that over time, the nature of orders will change—they will still include logic tied to seat licenses, but they will also incorporate a metering system similar to Azure’s. Some usage may not appear in orders in the traditional way but will instead be billed directly based on consumption.
If this usage can deliver real value to customers (which Satya will elaborate on shortly), customers will continue to use these AI agents because they directly create value or drive growth for the business. Therefore, I believe this transformation should be viewed from a broader perspective. While it may not immediately reflect in order data in the short term, I recommend thinking about this opportunity from this angle when evaluating it.
Satya Nadella:
Amy is absolutely correct. The fundamental shift is this: any of our businesses that are billed per user—whether in productivity, coding, or security—will evolve into a hybrid model of 'per user plus per usage.' This is the most accurate way to understand it. This is already happening in the coding space and has begun to take shape. Some of our business model adjustments this quarter reflect exactly that.
It also reflects the essence of usage intensity—where does the funding for this come from? Ultimately, it comes from the real value created by enterprises through the use of agents, whether working on behalf of users or collaborating with them: be it in customer service, personal productivity, team collaboration, or a specific business process. By using agents, either costs are decreasing or revenue is increasing because workflows are being compressed and optimized.
This is precisely the pattern we are observing. When people use Copilot, they utilize chat, reasoning-enabled chat, Cowork, agent modes, and agent functionalities within Word, Excel, and PowerPoint—but all of this happens within the context of specific task trajectories. When they discover that these task trajectories compress workflows, increase revenue, or reduce costs, usage will continue to grow.
Therefore, this may no longer be the kind of purely seat-based sales campaign we saw in the past; instead, it will focus more on cultivating high-intensity users and high-intensity usage. This is the core of our attention.
Amy Hood:
Keith, I would like to take this opportunity to sincerely express my gratitude to you. Working with you over these many quarters has been a rare and valuable experience. We deeply appreciate your long-term coverage and research on us. Congratulations, this should be your last earnings season tracking us.
Satya Nadella:
Thank you, Keith. It has been a great pleasure working with you.
Keith Weiss:
Thank you very much.
Karl Kilstedt (UBS):
Amy, could you provide further clarification on the capital expenditure guidance you just mentioned? It appears that capital expenditure will need to increase significantly in the second half of the year, reaching approximately $120 billion. I would like to understand your confidence level in overcoming physical component constraints to achieve this goal. Does this involve more participation from partners? What is your overall framework for allocating the new capacity between third-party and first-party usage? Thank you.
Amy Hood:
Thank you, Karl. In fact, I am quite confident that we can overcome the physical constraints at the industrial logic level of the supply chain to achieve this goal.
Part of the work involves bringing more capacity online, but much of it is near-term — ensuring CPUs, GPUs, and storage are in place to better support the demand signals we continue to observe. Price factors partly explain this, reflecting the situation in terms of volume. The impact of pricing on the capital expenditure figures also suggests a higher proportion of short-term assets.
Regarding capacity allocation, our guidance for Azure’s fixed-currency growth of 39% to 40% in Q4 indicates that we can meet demand in a balanced way at the Azure level through efficiency improvements.
In terms of Copilot usage, we have observed a significant shift in its trajectory during the third quarter—a trend that is evident in both programming and productivity domains. I am equally confident about the prospects in the security field.
Looking ahead to the first half of FY27 (i.e., the second half of the calendar year), we have gained clearer insights into further optimizing efficiency, accelerating data center delivery, and bringing them to a 'revenue-ready' state as soon as possible.
Therefore, the balancing pressure between first-party usage and Azure demand will persist. We are making every effort to expedite capacity investment, and the capital expenditure figures for the second half reflect this strategy.
Brent Hill (Jefferies):
Amy, one common concern we often hear is that AI costs will be extremely high. However, tonight, your company, Google, and Amazon all released earnings reports showing higher profit margins. What might investors be overlooking? Why could AI potentially mean better profit margins for the industry in the long run?
Amy Hood:
Thank you, Brent. We have been discussing the current phase of the AI business cycle, comparing it to the development cycle of the cloud business. Looking back at the cloud transformation period, the profit margin of our AI business was actually higher than that of the cloud business at the time, and this advantage has been sustained.
I believe what we are truly focused on is ensuring that the business model reflects how these applications are built and the value they deliver. Based on this value, it is more captured through consumption-based and usage-driven pricing models—I think this may be somewhat underestimated by the market when looking at future profit margins.
At the same time, fully leveraging our own intellectual property is crucial. IP obtained through partnerships is free for us in the long term, and we can apply it to enhance profitability. Our deep investments in first-party hardware stacks also allow us to extract profit margins from the infrastructure layer. Of course, there is the efficiency work itself—we have been in a phase of accelerated expansion, deploying capacity into production as quickly as possible; after completing this phase, we have also begun focusing on efficiency improvements, including optimizations on both the hardware and software sides, to achieve such profit margin levels.
As Keith initially pointed out, when transitioning to a usage-based model, the core principle is to ensure that we provide extremely high value to customers. Therefore, we need to focus on customer usage that creates value—when usage generates value and positive outcomes, the expansion of TAM and return on investment will be excellent.
Mark Moerdler (Bernstein Research):
Thank you for the opportunity to ask a question, and congratulations on this quarter's impressive performance, growth, and positive signals in the outlook.
I would like to delve deeper into the issue of capital expenditure and ongoing investments. Business cloud, Azure, and AI are all growing rapidly, but investors are somewhat concerned about the 'mismatch' between the growth rate of capital expenditures and revenue. Could you provide some clarification regarding the timing, and how much of the spending is allocated to equipment upgrades or first-party businesses? This would help us build confidence that, despite such large-scale capital expenditures, the core business remains very healthy and profitability will remain strong. Thank you.
Amy Hood:
Thank you, Mark. Let me start with Azure — given its scale and the accelerated growth trends we have discussed (Q4 guidance of 39% to 40%, with expectations of further acceleration in the second half of the year naturally), when you see growth rates like this on a business of this magnitude, the spending on short-term assets — which is what truly correlates with revenue, rather than the roughly one-third allocated to 15-year assets or lease contracts with certain time-based fluctuations — these factors can easily lead to confusion.
In a way, this reminds us of the previous cycle. When the Total Addressable Market (TAM) is so vast and the supply-demand gap continues to widen, the return on investment on the platform side is undoubtedly confidence-inspiring.
What you're really asking is: As the 'usage + consumption' model at the application and service layers gradually emerges, have we started to see corresponding financial benefits? Looking back at last quarter’s data, I believe we are seeing some signs of acceleration — the figures for M365 Business Cloud this quarter satisfy me, and Q4 guidance looks even more promising.
What you, Mark, and investors are asking is: When will these be reflected in revenue growth? That is the first indicator to watch. We can also observe this trend in GitHub — where revenue growth rates and usage-consumption patterns are driving top-line acceleration.
Considering the scale of capital expenditure, it takes time to translate into a 'revenue-ready' state, and we already have over $600 billion in backlog contracts — not including the accelerating growth we are seeing in Copilot seats. I am very confident about this situation. What we need to focus on is how to realize this revenue as quickly as possible. To be candid: when there is potential for faster revenue growth or efficiency gains, the priority must be to deploy capital expenditures and convert them into revenue as soon as possible.
Satya Nadella:
I would like to add that over the past two years, we have become increasingly convinced and steadfast in one aspect within the AI field—where is the Total Addressable Market (TAM), and what is the category economics of TAM? Interestingly, as we are now in 2026, the most exciting developments turn out to be plugins in Word or Excel, or CLI tools in programming. This precisely demonstrates that we hold a structural strategic position in the largest TAMs: knowledge work, programming, and security.
Combine this with the right business model (the "users + usage" model repeatedly mentioned by Amy), and consider the current volume of our contracted backlog—it paints a clear trajectory. If there is one thing we must ensure, it is securing sufficient capital expenditure capacity in time to meet the growth in usage—this will be critical. Keep in mind, model capabilities improve exponentially. Take, for example, the agent mode in Excel—it didn’t seem useful until a certain tipping point when it suddenly started working well—and this was simply because the model capability reached the necessary level. Therefore, you need to be prepared for these windows of opportunity.
Gabriela Borges (Goldman Sachs):
Satya, Microsoft has just achieved a series of technological and commercial milestones in the past three months. I’d love to hear your reflections and review on Copilot. In the current adoption of Copilot, what’s working well, and what needs improvement? How will this influence your E7 strategy and Copilot Cowork strategy? Thank you.
Satya Nadella:
Thank you for that question.
I believe that the evolution of M365 Copilot in the realm of knowledge work represents something of a pattern—we’ve learned a lot from programming—but focusing specifically on M365 Copilot itself, the first consideration is how the form and interaction patterns of the product have evolved.
Currently, there are several forms: chat, which now includes Work IQ reasoning capabilities; this is one form. Then there are various agents—for instance, researchers and analysts used within chats, or custom agents built by customers. Beyond that, there is now editing mode.
Imagine a typical Copilot workflow: starting with a chat, asking questions, gaining insights, having it generate a document, then opening that document in Word, Excel, or PowerPoint to refine it further—essentially continuing that conversation. Finally, there’s an entirely new, complete form—you delegate tasks outright without participating interactively, but through Cowork for task delegation.
These represent the various usage patterns. One of the most intriguing phenomena is that Copilot usage frequency has become comparable to Outlook. It’s no longer a question of whether people are using it or finding it useful—it has become a daily habit for intensive use.
Another key factor that makes these forms truly valuable is intelligence itself. Intelligence comes from two dimensions: the first is the ability to integrate multiple models with context, where the context includes meetings, documents, emails, Teams content, and all SharePoint data — and this is not a static database; it is the most critical database for any company, constantly evolving every moment; the second is the framework for multi-model collaboration — similar to what we have done in GitHub and the security domain: decoupling the framework from the models, and then allowing the richness of the context to be fully expressed, because customers will use various models.
"Critique and Council" is an excellent example, as is the "rubber duck" feature in GitHub Copilot. Even in Excel, I generate using Opus and validate using Codex — this is precisely the capability you want users to have.
Combining all of this with a business model centered on "user volume + user pricing," I believe this is what we are witnessing now, and everything is unfolding seamlessly. Thank you.
Kirk Materne (Evercore ISI):
Amy, could you talk about the changes to the OpenAI agreement? Are there any aspects we should focus on from either a modeling or financial perspective compared to a few weeks ago? Satya, this also seems like an opportunity for you to continue diversifying at the model level. Regarding the new framework you’ve established with OpenAI, what are the key points worth noting? Thank you.
Satya Nadella:
Let me start by saying that overall, we are very satisfied with our partnership with OpenAI. I always place great importance on ensuring that any partnership maintains a mutually beneficial structure at all times — this is foundational to sustaining a good relationship.
Specifically, there are several core elements to this agreement:
First, intellectual property — Amy also touched upon this point. We have royalty-free usage rights for cutting-edge models, encompassing all intellectual property, extendable up to the 32nd generation, and we fully intend to leverage this right. Some of the examples I mentioned in my remarks reflect this. We are deeply appreciative of this aspect, which constitutes a significant part of the agreement.
Second, OpenAI as our client — they are one of our major clients, not only in AI accelerators but also in other computing resources. We aim to provide them with high-quality service.
Third, we hold equity in OpenAI.
Overall, as both of our organizations grow and customer expectations for model diversity increase, we have jointly driven the evolution of our partnership. I am very satisfied with where we stand currently.
Amy Hood:
I would like to add two points: the revenue-sharing arrangement extends to 2030, providing us with valuable predictability. Secondly, as Satya mentioned, royalty-free use of intellectual property and the elimination of copyright fees we previously paid to OpenAI are also significant developments worth noting.
Rishi Jaluria (RBC Capital Markets):
Thank you for taking the time to address my question. I’d like to revisit the topic of the seat-based model versus the consumption-based model that was discussed today, along with the framework for how this might evolve over the long term. I fully agree with the trends you’ve observed, which seem entirely logical.
What I’d like to explore further is your announcement of E7 — a primarily seat-based offering supplemented by some consumption components, which reinforces the seat-based model. At the same time, more and more customers continue to value the predictability of the seat-based approach — a trend we can observe from various companies encountering challenges due to uncontrolled AI usage.
Could you help us understand how you plan to balance maintaining customer predictability while continuously expanding consumption? If we fast-forward three to five years, what might the ratio between the seat-based model and the consumption-based model look like? Thank you.
Satya Nadella:
Yes, from a macro perspective, I believe these two models are not mutually exclusive. As you pointed out, customers need predictability, particularly in terms of budgeting and procurement, and seat-based pricing essentially functions as an authorization for a certain level of consumption. The correct way to interpret this logic is that seats come bundled with a baseline usage right, forming a 'consumption package' conveniently priced per seat or agent. Beyond a certain threshold, overage charges apply, transitioning into a pure consumption-based billing model. Even within the consumption-based billing framework, customers making long-term commitments will receive corresponding discounts.
This is the direction we are heading. Additionally, you mentioned how clients will assess all of this—they will do so through evaluations (Evals). They will ask: Where can the value of the Token be observed? It's that straightforward. Their focus is on: where a particular outcome is achieved, where an evaluation generates benefits, and whether the corresponding Token consumption leads to increased revenue or efficiency improvements—this will create a continuously optimizing feedback mechanism.
Regarding IT budgets, the composition of IT budgets will have to adjust accordingly—reshaped by business outcomes, coupled with reallocations from other line items on the company’s income statement (such as operating expenses) into the IT budget, collectively driving this transformation.
Jonathan Neilson (Vice President of Investor Relations):
Thank you, Rishi. That concludes the Q&A portion of today's earnings call. Thank you all for participating, and we look forward to reconnecting with you soon.
Satya Nadella: Thank you, everyone.
Amy Hood: Thank you, everyone.
Host (Operator): This concludes today's conference call. You may now disconnect, and we wish you a pleasant day. The meeting has ended.
Disclaimer: This transcript may not be 100% accurate and may contain spelling errors or other inaccuracies.
Upgraded performance updates help you stay ahead in seizing opportunities!Come and experience it >>
Editor/KOKO