Microsoft Corporation (MSFT.US) Q3 Earnings Call: Azure Growth Driven by Enterprise Breadth and Layer-by-Layer Efficiency Gains; Capital Expenditures Expected to Reaccelerate in the Second Half of Fiscal Year 2027

Microsoft Corporation (MSFT.US) on Wednesday reported financial results for the first quarter of fiscal year 2027. Data showed that the company's revenue for the quarter reached US$92.279 billion, up 19% year over year; GAAP net income was US$36.629 billion, up 32% from US$27.747 billion in the prior-year period; GAAP diluted earnings per share were US$4.92, up from US$3.72 in the prior-year period, a year-over-year increase of 32%. Non-GAAP adjusted net income reached US$36.422 billion, up 18% year over year; adjusted earnings per share were US$4.89, up from US$4.13 in the prior-year period, a year-over-year increase of 18%. Microsoft Corporation said the non-GAAP results exclude the impact of its investment in OpenAI. On profitability, Microsoft Corporation's first-quarter operating income reached US$45.915 billion, up 21% year over year, or 22% in constant currency; operating margin was approximately 49.8%, up from 48.9% in the prior-year period. Gross profit for the quarter was US$62.559 billion, up approximately 17% year over year, with a gross margin of approximately 67.8%. The following is the full transcript of Microsoft Corporation's earnings call (translated with the assistance of AI tools): I. Executive Remarks Satya Nadella, Chairman and CEO We are off to a fast start to the fiscal year. Microsoft Corporation's cloud revenue surpassed US$63 billion, accelerating to 30% year over year. The company is focused on three things: designing AI that empowers people, ensuring every organization has sovereignty over its own learning, and building trust through controls that ensure the safety and security of agents. Next, let me talk about how the company is building a new tech stack around these goals, starting with Copilot. This quarter the company shipped its largest Copilot update to date, bringing chat, collaboration, autopilot and code into a unified experience. This is an integrated product experience spanning commercial and consumer, with the full Office suite built in. The company views Copilot as the new operating system for work. Customers are increasingly using Copilot to get real work done. Automated task execution has nearly doubled since July; daily active users grew nearly 50% quarter over quarter and more than 5x year over year; user satisfaction rose nearly 20% quarter over quarter, another all-time high; 90% of Copilot queries use auto mode, dynamically routing tasks across models to optimize among quality, speed and cost. These improvements are translating into larger purchases. The number of customers with more than 50,000 Copilot seats grew more than 8x year over year, including Capgemini, HCL Tech, Hitachi Group, Truist Financial and the UK Revenue and Customs Department. Just last week, Volkswagen announced plans to deploy Copilot to more than 330,000 employees, one of the largest deployments in Europe to date. The E7 suite (combining Copilot E5, Entra and Agent 365) has strong momentum, with thousands of customers across industries purchasing to date, including Amadeus, BlackRock, Inc. (BlackRock), Fiserv and the New South Wales Ministry of Health. The company's business model is expanding from per-seat subscriptions to include consumption models, thereby expanding the total addressable market (TAM) and average revenue per user (ARPU). Subscriptions bring increasingly powerful everyday intelligence to each user, typically funded by IT, integrating and rolling out frontier capabilities across the organization; and as agents take on higher-value work, consumption models open up demand funded by business units, with spending tied directly to the value those agents deliver. The number of Copilot customers actively consuming credits through usage-based billing grew more than 75% this quarter. In software development, all major coding agents have landed on GitHub. Over the past six months, pull requests (PRs) generated by agents grew 6x, and more than one-third of PRs on the platform are now opened by agents. The company is focused on improving reliability to handle the scale of agent activity on GitHub. At this morning's GitHub Universe conference, the company showed how it is expanding compute for the agent wave and re-architecting core Git; it also announced general availability of Copilot Fusion, bringing multi-model orchestration to software development, delivering up to 40% speed gains and 77% cost reductions. Since usage-based billing took effect, credit consumption for agent features on GitHub has grown more than 1x. The company is also helping developers secure code, turning AI-generated code into high-quality software. GitHub Advanced Security revenue grew more than 50% year over year, with customers using its AI capabilities to scan entire codebases and remediate vulnerabilities in real time. In business applications, the company is extending its vision of "Copilot as the new operating system for work," embedding Dynamics 365 business logic and capabilities directly into Copilot to help every organization build its own "infinite SaaS factory." This quarter, the company introduced 30 new Copilot skills in Dynamics 365, enabling customers to invoke CRM and ERP capabilities directly within workflows. In recruiting, more than 10 million LinkedIn members use AI-powered job search every day, and 25,000 customers use the company's agent-based hiring solutions. Recruiters using the hiring assistant find interview-quality candidates 33% faster while reviewing 83% fewer resumes to find qualified candidates. In healthcare, the company automated 30 million patient visits this quarter, up 70% year over year, helping clinicians reduce administrative work and spend more time with patients. Ten healthcare providers, including Advocate Health, Mass General Brigham, Mercy and Intermountain Health, have each recorded more than 1 million visits using the company's solutions since the start of the calendar year. In scientific research, the company's forward-deployed engineer (FDE) team works with customers to accelerate scientific progress using Discovery. A prime example is Saint-Gobain, which is using Discovery to accelerate the development of low-carbon building materials. Beneath these first-party agent experiences is a unified platform. First is Foundry, which is increasingly an important driver of growth. Foundry helps organizations build learning loops grounded in their business context and deploy agents securely at scale. The company now has more than 120,000 Foundry customers, with revenue growing more than 3x year over year. Foundry's model choice and feature set, combined with its deep integration with the company's stack, are creating differentiation. Azure OpenAI benefited directly last quarter from major advances in OpenAI models; combined with Foundry's built-in model choice, customers can build solutions that optimize among quality, cost and performance. In fact, the number of customers building agent solutions with multiple models grew 5x year over year, and revenue from open-source models on Foundry grew 10x. The company's first-party models also continue to advance rapidly. The company's goal is to push the "cost-to-outcome frontier," delivering comparable performance within three to six months of a leading model's release at a cost far below its training cost, passing those cost benefits on to customers. This is already reflected in today's products: MAI-Cyber now powers more than 40% of scans in the MDASH code scanner, reducing the cost per scan by 18%; MAI-Transcribe is the world's most accurate and cheapest transcription model, powering Dragon Copilot's multilingual workflows; MAI-Code-Flash had 3 million users on GitHub alone last month, helping developers use AI credits more efficiently; Microsoft-Decision-1 is the company's new model for structured decisions, outperforming LLMs and other decision models on both latency and quality, with more than 4,000 customers using it in its first month, including Atos Group, Equinor and Raiffeisen Bank International (RBI). The company recently completed pre-training of MAI-Base-2, which in tests was 4 to 12 times more compute-efficient in training than other open-weight models; this foundation model will be used for subsequent reasoning, code and cybersecurity MAI models. Beyond models, the company is also opening to Foundry customers the frameworks that power its own Copilot experiences, offering a Microsoft Corporation-managed way to build agents that can reason, call tools and complete complex long-running tasks while preserving customer model choice. Another key layer is data. The company now has 44,000 paid Fabric customers, up more than 50% year over year. The company is bringing Foundry and Fabric more closely together so customers can build their own agents next to their data. More than 19,000 customers now use both Foundry and Fabric, enabling enterprises to connect and operate agents across real-time operational, analytical and unstructured data. Customers are also rapidly deploying the company's AI-optimized databases, including PostgreSQL, SQL DB Hyperscale and Cosmos DB, to give agents fast and secure access to data and context. PostgreSQL growth accelerated for the fourth consecutive quarter, and the number of PostgreSQL customers also using Foundry grew 90% year over year. The company also has one of the most important databases of any enterprisethe knowledge base beneath Microsoft 365, spanning people, their relationships, documents, emails, meetings, chats, projects and other assets. Work IQ makes this context available to every agent workload. With Microsoft IQ, the company brings together Work IQ, Fabric IQ and Foundry IQ into a unified context for agent systems. Nearly 60,000 customers are accessing context on top of Microsoft IQ, up 25% quarter over quarter. As enterprises deploy more agents, they also need to govern them at scale. Through Agent 365, the company provides an AI observability platform that lets customers manage, secure and optimize the performance and cost of their agents. There are now more than 50 million agents in the registry, up more than 30% quarter over quarter. Finally, as agents' access to enterprise data and systems expands, security becomes even more critical. The company's approach is to separate "intelligence" from "permissions," enforcing identity, permissions, monitoring and containment outside the model. Last week, the company expanded Perception's new capabilities to help customers protect against and contain agent insiders, including agent discovery across on-premises and cloud environments, dynamic permissions that enforce least-privilege access for each task, and continuous risk assessment. The company also introduced the Guardian agent team to continuously monitor agent activity and prevent dangerous behavior before it happens. Perception has received positive feedback from preview customers, including KPMG, Nationwide Building Society, Northern Trust Corporation (Northern Trust) and Salentis. More broadly, the company is helping customers secure and govern AI use across the enterprise. To date, customers have used Microsoft Purview to audit more than 7.6 billion agent activities, up 5x quarter over quarter. Underpinning all of this is infrastructure. The company is building fungible compute pools against its growing backlog so it can dynamically allocate capacity to the highest-value opportunities across compute, data, Foundry and Copilot. Accelerating demand led the company to add 24 data centers across five continents this quarter and to bring capacity online faster, with "doctor lifetimes" (i.e., delivery lead times) for the new GB300 in its largest regions shortened by about 20% quarter over quarter; these regions account for the bulk of GPU deployments. At the same time, the company is extracting more value from existing infrastructure by optimizing data center design, chip systems, software and models. For example, throughput for Copilot workloads improved more than 50% quarter over quarter. The company continues to expand its compute pools with the latest innovations from NVIDIA Corporation (NVIDIA), AMD and its own Maia systems, and has taken delivery of its first NVIDIA Corporation Vera Rubin GPUs and is very excited about them. Maia 200 has achieved up to 1.7x more tokens per watt in Copilot's AI experiences and teams, and Maia 300 will drive further gains, with 40% more tokens per watt than leading GPU systems. Growth extends beyond GPUs. The company's homegrown cloud-native Cobalt CPU continues to see strong adoption, improving price-performance for general-purpose cloud workloads. Azure growth overall was broad across enterprise and small and mid-sized customers, as well as between core cloud services and AI infrastructure. Finally, devices and consumer. In Windows, the company is pursuing its vision of "bringing hybrid intelligence to every desk and every home, making every PC a secure agent container." With more powerful models running locally, there is an opportunity to make intelligence effectively unmetered to a greater degree. The company is working with the broader hardware ecosystem to bring these capabilities to a new generation of PCs based on AMD Ryzen AI 400 series, Intel Corporation Core Ultra series 3 and Qualcomm Snapdragon X2 series. Earlier this month, the company also announced new PCs powered by NVIDIA Corporation RTX Spark and DGX station for Windows. With these new devices, users can run last year's frontier models on a laptop and last quarter's models on a desktop. The company is integrating Windows with Agent 365 so agents can run on devices within a secure boundary. The company introduced Microsoft Execution Containers (MXC)a policy-driven execution layer that governs and contains agents on Windows and other platforms, integrated with Agent 365 for governance and observability. Agents such as OpenAI Codex, Anthropic Claude Code and Meta Muse are adopting MXC, and customers including Meta have begun deploying it in their enterprise environments. In search and advertising, Bing and Edge gained share again. At LinkedIn, engagement deepened across 1.4 billion members, with feed content viewing time up 15% year over year. In gaming, the company is focused on making great games and reaching players on the platforms they chooseconsole, PC, mobile and cloud. Minecraft delivered a record quarter, with engagement and revenue both at all-time highs; World of Warcraft held the largest test in the franchise's history. The company enters the holiday season with its strongest first-three-quarter lineup in years, including "Modern Warfare IV," released last week to the best reviews for a Call of Duty title in more than a decade, and "Grand Theft Auto 6" (GTA 6), coming to Xbox next month. In summary, based on the demand the company sees, the strength of its products and the efficiencies it is driving across the stack, management has never been more confident in the long-term returns on its investments. I'll now hand the call over to Amy. Amy Hood, Executive Vice President and CFO Results exceeded expectations on revenue, operating income and earnings per share, driven by strong demand for Microsoft Corporation cloud. Revenue for the quarter was US$92.3 billion, up 19% year over year. Gross profit dollars grew 17%, with a gross margin of 68%. Operating income grew 21%, or 22% in constant currency. Earnings per share were US$4.89, up 18%, or 19% in constant currency and adjusted for the impact of the OpenAI investment. FX impact was broadly in line with guidance. Operating expenses grew 6%, driven by continued investment in R&D compute and talent. Operating margin rose slightly year over year to 50%, driven by strong revenue growth, platform efficiency gains and focused investments across the portfolio, partially offset by the mix shift toward Azure. Adjusted for the impact of the OpenAI investment, other income and expense was negative US$234 million, slightly below expectations. Capital expenditures were US$45.2 billion, as some third-party capacity deliveries slipped into the second quarter. About three-quarters of capital expenditures went to short-lived assets (primarily CPUs and GPUs, including higher memory and storage costs) to continue building the infrastructure customers need to build and scale AI applications; the remainder went to long-lived assets. Finance leases were US$7.4 billion this quarter. As a reminder, the company recognizes the full principal value of finance leases and capital expenditures when the lease commences. Cash used for property, plant and equipment was US$39.4 billion. Operating cash flow was US$58 billion, up 29%, driven by strong cloud billing and collections. Free cash flow was US$18.6 billion, reflecting higher capital expenditures. Finally, the company returned US$10.2 billion to shareholders through dividends and share repurchases. The following are segment results, reported under the new segment structure announced in September. Agents and Infra revenue was US$76.8 billion, up 25%, reflecting strong performance across every layer of the platform. Commercial bookings declined 12% (11% in constant currency), affected by large Azure commitments in the prior-year period, partially offset by healthy growth in commercial sales momentum this quarter (including a large Azure commitment this quarter). Commercial remaining performance obligation (RPO) grew 73% to US$677 billion, driven by large Azure commitments this quarter, with a smaller sequential decline than historical seasonality. The weighted average duration of RPO is 2.3 years, with about 30% to be recognized as revenue over the next 12 months, up 29% year over year; the portion to be recognized beyond the next 12 months grew 98%. For Microsoft Corporation cloud, revenue was US$63.9 billion, accelerating to 30% (31% in constant currency), with sequential dollar additions nearly doubling year over year, reflecting strong demand for Azure and M365 commercial cloud. Azure revenue exceeded expectations, growing 48% despite the aforementioned third-party capacity delays, driven by Foundry and improved CPU monetization. Foundry revenue grew more than 3x year over year, with the beat partly from GPU pool efficiency gains that allowed the company to allocate more capacity to customers, as well as higher revenue per token that benefited the quarter. On the CPU pool, engineering work allowed the company to identify and monetize underutilized capacity this quarter, helping it serve more of the growing demand with existing capacity. The timing of such engineering efficiencies, third-party GPU lease deliveries, the company's own capacity buildout and ARPU changes from model mix can cause quarterly variability in year-over-year growth. M365 commercial cloud revenue grew 18% (19% in constant currency, adjusted for the 1-point in-period revenue recognition impact in the prior-year period); reported growth was 17% (18% in constant currency). Results exceeded expectations, driven by enhanced GitHub Copilot adoption and usage, partially offset by weaker sales execution on M365 E3 and standalone SKUs. The company continues to refine its sales motions, balancing the cadence of more traditional per-seat SKUs with newer premium E7, M365 Copilot and usage-based productivity offerings. Year-over-year growth was driven by both ARPU and seats, with ARPU growth led by premium per-seat offerings such as M365 Copilot, E5 and E7E7 paid seats have reached millionsplus usage-based offerings led by GitHub Copilot. Copilot revenue accelerated sequentially, with continued adoption across M365, GitHub and security. M365 commercial paid seats grew 5%, partly affected by the aforementioned weaker sales execution. Growth was driven by installed-base expansion across all customer segments, primarily from small and mid-sized business and frontline worker offerings. M365 consumer cloud revenue grew 17% (16% in constant currency), down sequentially as it began to lap the prior-year price increase. Productivity and server licensing revenue declined 3% (2% in constant currency), mainly due to lower transactional purchasing against a high prior-year base from product launches, partially offset by a higher mix of contracts with in-period revenue recognition. Industry solutions cloud revenue grew 10%, with growth in Dynamics 365 and LinkedIn talent solutions, though booking trends continued to slow. Frontier and Support Services revenue grew 4% (5% in constant currency), driven by support. Microsoft Corporation cloud gross margin declined year over year to 66%, driven by the mix shift toward Azure, partially offset by product-level margin improvement led by Azure. Segment gross profit dollars grew 20% (21% in constant currency). Operating expenses grew 10%, driven by the aforementioned R&D investments. Operating income grew 24% (25% in constant currency), with operating margin declining slightly year over year to 53%, as strong revenue growth, platform efficiency gains and focused investments were offset by the mix shift toward Azure. For devices and consumer, revenue was US$15.5 billion, down 3%. Search and advertising revenue ex-TAC grew 12%, exceeding expectations, driven by higher volume and price across LinkedIn, Edge and Bing. Xbox revenue declined 2%, with Xbox content and services revenue down 2% (1% in constant currency), driven by declines in third-party content and Xbox Game Pass, partially offset by growth in first-party content. Windows OEM and devices revenue declined 22%. Windows OEM revenue was below expectations, down 27% (28% in constant currency), driven by demand from the prior-year Windows 10 end-of-support, OEM and channel partner inventory destocking, and soft PC market demand, with inventory levels still elevated. Segment gross profit dollars were roughly flat year over year, and operating expenses declined 3%, driven by Xbox, partially offset by the aforementioned R&D investments. Operating income grew 2% (3% in constant currency), with operating margin rising year over year to 35%, driven by lower amortization of intangibles related to the Activision acquisition, partially offset by a lower mix of Windows OEM revenue. The following is the outlook for the second quarter, in US dollars unless otherwise noted. Based on current rates, the company expects FX to have no significant impact on total revenue, cost of sales, operating expense growth or revenue growth in either segment. On segment guidance: Agents and Infra revenue is expected to be US$80.1 billion to US$80.7 billion, growth of 24% to 25%. On commercial bookings, the company expects healthy growth in commercial sales momentum, but results will be affected by the significant Azure commitments signed in the prior-year period; the timing of such significant Azure commitments will cause significant volatility in bookings, making quarterly growth less reflective of underlying sales execution. For Azure, revenue growth is expected to be 49% to 50% in constant currency, with continued efficiency gains in CPU and GPU pools partially offsetting some third-party GPU-as-a-service contracts delayed into the second half; customer demand continues to exceed available supply. Beyond the second quarter, year-over-year growth will continue to reflect the timing and magnitude of engineering efficiencies delivered in the quarter, new capacity, and the higher prior-year base from accelerated growth in the second half. For M365 commercial cloud, growth is expected to be about 19% in constant currency, with FX expected to reduce reported growth by less than 1 point; momentum in premium per-seat and usage-based offerings is expected to continue, and execution on E3 and standalone SKUs is expected to improve; growth will be driven primarily by ARPU, with seats also contributing, and full-year revenue growth is expected to continue to accelerate. M365 consumer cloud revenue is expected to grow low double digits, down sequentially, due to the aforementioned high base from price increases. Productivity and server licensing revenue is expected to decline mid-to-high single digits, down sequentially, as the mix of contracts with in-period revenue recognition is expected to decline; quarterly growth may vary significantly due to contract mix, product launch timing and other transactional purchasing trends. Industry solutions cloud is expected to grow mid-to-high single digits, still affected by the aforementioned booking trends. Frontier and Support Services is expected to grow low-to-mid single digits. Microsoft Corporation cloud gross margin is expected to be about 65%, down year over year, as continued product-level margin improvement is offset by the mix shift toward Azure. For devices and consumer, revenue is expected to be US$15.9 billion to US$16.4 billion. Search and advertising revenue ex-TAC is expected to grow low double digits, driven by consistent volume and price trends, with total search and advertising revenue expected to grow at a similar rate. Xbox content and services is expected to grow mid-to-high single digits, up sequentially, driven by the timing of first-party content releases; given the full-year content release cadence, growth is expected to be strongest in the second quarter, slow in the third quarter and slow further in the fourth quarter. Second-quarter hardware revenue is expected to decline year over year. Windows OEM and devices revenue is expected to decline around 30% (mid-30s). Growth will continue to be affected by elevated inventory levels, PC market demand and the high prior-year base from Windows 10 end-of-support. As in prior quarters, the range of possible outcomes remains wider than usual. The company expects these trends to persist through the remainder of the fiscal year. So at the company level, revenue is expected to be between US$96.0 billion and US$97.1 billion, growth of 18% to 19%. Cost of sales is expected to be US$32.7 billion to US$32.9 billion, growth of 26% to 27%; operating expenses are expected to be US$18.5 billion to US$18.6 billion, growth of 9%, driven by continued investment in R&D compute, capacity and talent. Both the cost of sales and operating expense guidance include estimates for the recently announced voluntary employee separation program. Operating margin is expected to decline slightly year over year. Adjusted for the impact of the OpenAI investment, other income and expense is expected to be approximately negative US$300 million, as interest income will be offset by interest expense (including interest related to data center finance leases). The effective tax rate for the second quarter is expected to be about 20%. On capital expenditures, the second quarter is expected to exceed US$50 billion. As a result, calendar-year spending is expected to be approximately US$170 billion, due to some third-party capacity deliveries delayed as previously mentioned. In the second half of the fiscal year, capital expenditures are expected to accelerate, primarily from CPUs (including higher memory and storage costs) and GPUs, as well as third-party capacity deliveries delayed into the second half. As infrastructure efficiency continues to improve, each dollar of capital expenditure is expected to support more revenue and higher returns. Given the strong business performance, the company expects free cash flow to remain positive for full-year FY27. To supplement cash generated from operations and maintain financial flexibility, the company will use additional financing sources when it deems appropriate. In summary, FY27 is off to a strong start, with Microsoft Corporation cloud growth accelerating. Product improvements and efficiency gains across data center infrastructure, first-party applications and agents are increasing the value the company delivers to customers and improving outcomes. Delivering value to customers gives the company a foundation for durable long-term growth. II. Q&A Session Adam Wood, Analyst, Morgan Stanley: Drivers of AI infrastructure returns Question: With capital expenditures below expectations and better cloud gross margin sequentially, as new AI capacity comes online, are the more critical drivers of returns over the next few years improvements in utilization and unit costs during the year, or the ability to monetize each workload across more layersFabric, Foundry and Copilot? Answer (Satya Nadella): All of the above. The company's approach is to build a compute pool and make it as efficient and fungible as possiblethis involves all the physical aspects of data center construction, operational discipline in bringing capacity online, and how optimized workloads run on that infrastructure. Once that is done, it is simply a matter of allocating based on demand signals. This is visible even under the new segment disclosure: the stateful database beneath Copilot is Work IQ, which is used in Foundrycustomers build, for example, multi-agent supply chain systems across unstructured and structured data on Foundry, and of course use the PostgreSQL database in Azure. In addition, all agent work is registered in Agent 365 and its security is managed by Perception. So the company views these as the same infrastructure, just reported as different product forms. The core question is where profit and leverage come fromnot from any single layer, but from every layer, and when the company threads them together, the benefits compound. Answer (Amy Hood): Sorry, Adam, if I missed the first part of the question, I at least addressed the second part. Karl Keirstead, Analyst, UBS Group AG: Drivers of the 49%-50% Azure growth guidance Question: Let me drill into one of the key metrics from this callthe December guidance for Azure growth of 49% to 50% in constant currency. There are many variables, as you can imagine: capacity coming online, allocation to third-party customers, perhaps OpenAI strength, but given how strong that number is, please discuss some of the key drivers. Answer (Amy Hood): Let me start with the first quarter, which helps frame the second quarter and beyond, and this actually ties to Adam's question. In the first quarter, the engineering teams did outstanding work delivering efficiency gains at the platform layernot just GPU efficiency, but efficiency at every layer: how applications use the platform, how Foundry uses the platform, and the Azure platform itself. These efficiencies accumulate bit by bit every week; some ideas work, some don't, and it requires sustained investment, so the impact in any single quarter isn't precisely predictable. The first quarter was very good, and part of the result showed up as the company being able to allocate more capacity to Foundry during the quarter. That's a good thing, because as noted, demand still exceeds supply; whenever the engineering teams deliver such efficiencies, the company tries to use them for Foundry, because as Satya said, that is extremely important for customers broadly building the new stack. Second, token pricing creates some variability, and token pricing this quarter was higher than expectedgood margins from high-quality models did benefit the company, and that happened in the first quarter. Also, in the first-quarter numbers, the aforementioned efficiencies applied not only to the GPU pool; people may underestimate the integrated GPU, CPU, storage and memory workloads in Satya's examplesit involves all of it. As a result, the company was also able to create additional capacity within its existing pools, and that converted into results relatively quickly in the first quarter. In the second quarter, those efficiency gains carry forward, and the company needs to find more efficienciesit is confident about that because the first-quarter work is expected to continue into the second quarter, and it also needs some help from that work because some supply originally expected in the second quarter has slipped into the second half. In addition, token pricing is uncertain within the quarter. Overall, the company is pleased with the efficiency gains it sees and is comfortable guiding on that basis; other aspects are harder to predict. Mark Moerdler, Analyst, Bernstein: Sources of Azure AI strength Question: On the sources of Azure AI strength, is it enterprise customers or more digital-native/AI-native customers; as inference starts to scale and Microsoft Corporation becomes the enterprise default, how should we think about the current share of enterprise customer workloads in total workloads; and what is driving the strong growth of the Azure AI platform offeringsagain enterprises, or digital natives (clearly a mix, but details showing enterprise scaling would be more valuable)? Answer (Satya Nadella): Last week in Europe, the quality of the agent systems I saw was impressive, and they are being deployed broadly. For example, a fully automated agent-based touchless system at a delivery platform, covering automated ordering and transaction dispute management; Rolls-Royce deployed a system to manage its maintenanceits jet engines involve 30,000 parts, a fairly complex supply chain, and the planning, execution and inventory management systems are all built on Foundry and implemented as agents; Raiffeisen Bank International (RBI) demonstrated more than 20 end-to-end business processes, whether credit underwriting or others, completed more optimally. The breadth, the quality, and the multi-model naturethey start with frontier models, and in the Unilever PLC Sponsored ADR (Unilever) case, they even use MAI models trained directly on the company's traces. The maturity of model optimization, workload optimization and token optimization is encouraging. Breadth is beginning to open up. Ultimately, enterprise workloads feel like the way enterprises used to build database applications, and it is happening again in AI. In fact, some of the "novelty chasing" is fading, and enterprises are more focused on: can we build a system, can we own the weights, can we manage costs, can we have reliability, what about security. So the enterprise breadth movement is beginning. But as with everything, the enterprise business isn't quarterlyit's a decade-long endeavorand that is what the company is waiting for, rather than just four customers buying a lot of capacity and showing a few quarters of growth (that isn't a business). The company is pursuing breadth, is well positioned in breadth with large customers, and is open about the business: if the price is right, it can sell some capacity; otherwise it will patiently build a breadth business. Gabriella Borges, Analyst, Goldman Sachs Group, Inc.: Evolution of token mix and token value Question: On the "breadth" topic, management said in the prepared remarks that MAI can now match leading models within three to six months of their release, and OpenAI accounted for 11% of total revenue this quarter. Please step back and discuss the latest thinking on how the token mix and token value will evolve among frontier models, MAI and proprietary AI strategies at some mature enterprise customers. Answer (Satya Nadella): Think of the toolbox for any workload as three things. First, when a new model is released and performs well, customers can evaluate it against the outcomes they care about and adopt itthat was the case with some OpenAI models last quarter; they are excellent, and the company is excited about OpenAI's progress on Astra and the future, and it is seeing good demand. Second, customers then ask: can we use multiple frontier models at once, since their prices vary widely. In fact, in the real benchmark released this morning by Artificial Analysiscoding agents using the company's Fusionthe company ranked second or third, which is very exciting: now you can get frontier performance, 45% faster and 75% cheaper, so why not. For coding agent users, they want Anthropic and OpenAI to compete for every task, not once a quarter, and that is the value of routing and Fusion's automatic routing. Third, and importantly, many customerssuch as Unilever PLC Sponsored ADRhave their own systems and ultimately, even after doing a lot of work with frontier models, want to train their own models, whether open-weight or, as in this case, MAI models. So all of these should be available options, and customers will choose based on pricing and needs. There is a timing element: it's not that you start somewhere and stop there; you might start with frontier models, then move to open-weight, then to MAI models, and then go back to the frontier for a few months at the next model releasethat's how dynamic workloads are. So across Foundry, Fabric, Copilot and GitHub Copilot, the company wants to build a model system that continuously delivers value to customers, rather than locking them into any one thing. Brent Thill, Analyst, Jefferies Financial Group Inc.: Sustaining 50% revenue growth and 50% operating margin Question: Azure growth is approaching 50%, and operating margin is also approaching 50%. The market often asks: how do you sustain this growth at a level of 50% on both top-line growth and bottom-line margin. Please discuss the operational improvements you continue to drive to sustain the high bottom line. Answer (Amy Hood): First, the best solution is revenue growth, so you should always start at the top of the income statement and work downalthough many assume the company works from the bottom up. What the company does at its core runs through all the commentary: it changed the segment structure because the way products and services are delivered is changingit is delivering new, more integrated products with higher value. When top-line revenue accelerates, it naturally contributes positive margins and gross profit dollars in a healthy way. Second, the company is truly focused on efficiency. When people mention "efficiency," they often think of cost, but while the results are cost-related, the real benefit of efficiency is stability, security and reliabilitydelivering the highest-quality unit at the lowest cost, which is operational excellence. So on top of product quality (which was very good last quarter), the company layers continuous operational improvement and a focus on operating expenses, ensuring they go toward building new things or delivering security and quality. The company has made very deliberate choices to focus on these two things, especially as their importance stands out in the current environment. When you do that and also happen to have a good market with breadth customers and global operations, you see good results. That was especially true this quarter: delivering a lot of efficiency gains during the quarter allowed revenue growth without additional cost, and the benefits naturally showed. Management is pleased with the elements supporting that numberthe focus on the right levers. Brad Reback, Analyst, Stifel: Unlocking M365 consumption models Question: When the company shifted GitHub from a per-seat model to a consumption model, it unlocked a lot of revenue growth. For the tools now in place on the consumption side for core M365 customers, is that unlocking similarly fast and requires no sales effort, or does it take considerable effort to get those "meters" running? Answer (Satya Nadella): It will be faster on the coding side, because the market is already more accustomed to pay-as-you-go. In fact, products like Fusion announced today at GitHub Universe offer a more attractive wayafter buying credits, you want to use them as efficiently as possible, and that is the promise of GitHub Copilot Fusion. The company is optimistic about that. As for Copilot itself, the company is in a sense injecting more and more value into the subscription with every model release, effectively integrating and rolling out frontier gains; in many cases IT funds it, and it is what they want standardized across the organization. At the same time, pay-as-you-go or consumption meters remain available for scenarios that need them. Both options are open, and the company is working to help customers understand both and use them optimally, rather than overbuying one and getting stuck maximizing tokens to no one's benefit. The company wants customers to understand the value of the subscriptionit improves continuously and is not staticwhile also being able to use the consumption meter. This is an area of growing maturity in sales motions and customer understanding, and the company is pleased with progress but still has a lot of work to do. Answer (Amy Hood): This is also why the company invests in frontier companies and pairs that with the aforementioned changes in sales motionsboth are needed, because as Satya said, the goal is for customers to fully use Copilot licenses and also understand when the consumption meter delivers value to them. The investment in the forward-deployed engineer (FDE) organization or Frontier collaboration, and the new motions in the modern work and Microsoft 365 sales teams, are all aimed at achieving this. As Satya noted, the transition outside coding will be slower, but the company still sees encouraging signs and is positive about it. Tal Liani, Analyst, BofA: Capital strategy Question: On capital commitments/capital strategy, management said in the prepared remarks that it will raise financing in