$MSFT Microsoft Q4 2026 Earnings Conference Call
EARNMOAR
Call Introduction and Disclaimers 0:00
Jonathan Nielsen, vice president of investor relations, opens the call and introduces Satya Nadella, Amy Hood, Alice Jolla, and Brian Defoe. He points listeners to the investor relations website for the earnings release and financial summary slides, and reminds everyone that the discussion will include non-GAAP measures, constant currency comparisons, and forward-looking statements subject to risks described in Microsoft's SEC filings.
Record Fiscal Year Results 3:00
Satya Nadella opens by calling it a record fiscal year, with annual revenue surpassing 331 billion dollars, up 18 percent. Microsoft Cloud revenue surpassed 214 billion dollars, up 27 percent, and Azure surpassed 75 billion dollars run rate territory, up 41 percent. He frames two forward goals: making sure AI empowers every person by amplifying their agency and ambition, and helping every organization build its own continuous learning loop so it does not outsource its core intellectual property.
Expanding AI Infrastructure 3:30
Microsoft added 31 new data centers this quarter across five continents, bringing the year's total to 88, and cut the time to bring new GPU capacity online by nearly half in its largest regions. The company added another gigawatt of capacity this quarter and remains on track to roughly double overall capacity within two years. Efficiency gains include a fourfold increase in throughput for copilot workloads since the start of the year, a new sovereign cloud partnership with Mistral, and continued rollout of Microsoft's own Maya 200 chip, which delivers 30 percent better performance per dollar than the latest fleet hardware and now supports both OpenAI and MAI models. Microsoft also plans to be among the first cloud providers to deploy AMD Helios and Nvidia Vera Rubin rack-scale infrastructure, while its Cobalt CPUs power workloads for customers like Adobe, ARM, Elastic, OpenAI, Sprinklr, and TomTom.
Model Choice and Foundry Platform 6:03
Microsoft offers over 11,000 models in its catalog, including offerings from OpenAI, Anthropic, Mistral, xAI, and its own MAI family, and has seen a fivefold rise in customers using multiple model providers since the start of the year. Levi Strauss is cited as using both OpenAI and Anthropic models on Foundry to run more than a thousand domain-specific agents. Microsoft has introduced over a dozen new models, including its first reasoning model, MAI Thinking One, and is building a modular system separating context, memory, and action from any single model, which lets models be swapped for cost and resilience. Examples include MAI Code 1 Flash on GitHub Copilot achieving higher acceptance rates with 10 percent lower token usage, and efficiency gains such as an 89 percent reduction in GPU costs in Dynamics 365 and up to 84 percent in PowerPoint.
Data, Context, and Agent Governance 8:30
Microsoft's data estate is shifting from supporting human-used apps to supporting agents, with Postgres SQL revenue up 55 percent for a third straight quarter of acceleration and over 40,000 paid Fabric customers, up more than 60 percent year-over-year. A new service called Horizon DB delivers three times the throughput of self-managed Postgres deployments, and a new SDK called Rafin, used by more than 2,500 customers, powers backend services for apps including those built with Replit. Nearly 90 percent of the Fortune 500 already ground agents in enterprise context using Foundry, Fabric, and Work IQ, and a new Web IQ tool gives agents access to real-world web intelligence, already used by ChatGPT. Foundry itself now has 100,000 customers with revenue more than doubling year-over-year, while Agent 365, a governance control plane for agents, has nearly 40 million registered agents just two months after launch.
Copilot Adoption Across the Enterprise 11:31
Microsoft now has over 30 million paid Microsoft 365 Copilot seats, with net seat additions more than doubling quarter over quarter. Copilot is evolving from chat to co-work to autonomous long-running agents called autopilots, and user satisfaction scores have doubled over the past three quarters while latency dropped 25 percent this quarter. Large deployments include NHS England rolling out Copilot to 505,000 clinicians and staff after a trial showed it saved 43 minutes per employee per day, KPMG expanding across 276,000 professionals, HSBC committing to 200,000 seats, and EY deploying Copilot's new E7 suite to 400,000 employees, the largest win to date for that bundle which combines Copilot, E5, Entra, and Agent 365.
Developer Tools, Security, and Vertical Apps 15:30
GitHub Copilot now has 50 million users, and after shifting to usage-based billing, its revenue accelerated more than 60 percent quarter over quarter, with GitHub overall reaching 225 million users and one in three pull requests now involving an agent. In security, Purview has audited over 50 billion Copilot interactions, up nearly 360 percent year-over-year, and a new tool called Project Perception simulates attacks and drives remediation using teams of agents. In healthcare, Microsoft is on pace to automate over 100 million patient encounters this year, including 28 million this quarter, and Mass General Brigham's rollout of Dragon Copilot to over 4,000 providers cut burnout by 21 percent in a study. Dynamics 365 is being rebuilt for an agent-first world, exposing over 650,000 actions across business functions, and Microsoft Discovery is now broadly available for scientific and engineering workflows with early customers including BHP, GSK, and Pacific Northwest National Lab.
Frontier Core and Devices Business 18:01
This month Microsoft launched Frontier Core, an outcome-driven engineering organization embedding 6,000 industry and engineering experts with customers, building on a year of testing that included over 330 projects across 164 customers, such as helping Novo Nordisk build a compliant clinical data agent. On devices, Nadella says Xbox needs a business reset across content, platform, and operations, with a return to growth targeted for fiscal 2027, while Windows is being positioned as a secure edge AI platform. Bing and Edge have gained search share for five straight years, and LinkedIn has posted double-digit member growth for a fifth consecutive year, with recruiter AI seats up 140 percent quarter over quarter.
Financial Results and Segment Performance 20:00
Amy Hood reports fiscal year revenue of over 331 billion dollars, up 18 percent, with operating income growing 21 percent to more than 155 billion dollars. Quarterly revenue was 90 billion dollars, up 18 percent, with earnings per share of 4.74 dollars, up 23 percent when adjusted for the OpenAI investment impact; results included a 3.2 billion dollar gain from the Anthropic investment. Capital expenditures were 41 billion dollars, cash flow from operations rose 30 percent to 55.4 billion dollars, and the company returned 10.2 billion dollars to shareholders this quarter, over 43 billion for the year. Commercial remaining performance obligation grew 84 percent to 678 billion dollars. By segment, Productivity and Business Processes revenue was 37.8 billion dollars, up 14 percent; Intelligent Cloud revenue was 39.3 billion dollars, up 32 percent, with Azure growing 43 percent; and More Personal Computing revenue was 22.9 billion dollars, down 4 percent, with Xbox content and services revenue down 10 percent.
Depreciation change and capex outlook 29:30
Starting in fiscal 2027, Microsoft is extending the useful life of its data centers and office buildings from 15 to 25 years, reflecting how long these assets actually stay in operating history and use. This mainly changes the timing of future depreciation and gives only a minimal benefit to FY27 operating income. The bigger effect is on capital expenditures, because more future data center leases will now be classified as operating leases rather than finance leases, and only finance leases count as capex. Outside of this accounting shift, calendar year 2026 capex plans are unchanged, but the lease reclassification adjusts the expected figure to approximately 175 billion dollars.
Full year FY27 guidance 31:00
Microsoft expects M365 commercial products and server products revenue to decline in the mid single digits for the year as it laps stronger prior transactional purchasing, while Windows OEM and devices revenue should decline in the high teens due to weaker PC demand and higher component costs. Foreign exchange is expected to reduce full year revenue growth by less than one point. At the company level, Microsoft expects another year of double digit revenue and operating income growth, with operating expenses growing mid to high single digits and capital expenditures rising year over year to meet demand. Full year operating margins should be down less than a point, the company expects to stay free cash flow positive, and the FY27 effective tax rate is expected to be approximately 20 percent.
First quarter segment guidance 32:31
For the first quarter, Microsoft expects productivity and business processes revenue of 36.7 to 37 billion dollars, with M365 commercial cloud growing about 16 percent in constant currency on an adjusted basis. M365 consumer cloud should grow in the mid teens, LinkedIn in the high single digits, and Dynamics 365 in the low teens. Intelligent cloud revenue is expected at 40.95 to 41.25 billion dollars, with Azure growing approximately 45 percent in constant currency, though demand continues to exceed supply. More personal computing revenue is expected between 12.2 and 12.7 billion dollars, with Windows OEM and devices declining in the low 20s and Xbox content and services declining in the mid single digits. Total company revenue is expected between 89.85 and 90.95 billion dollars, representing 16 to 17 percent growth, with capital expenditures expected to exceed 50 billion dollars including the lease reclassification impact.
Model choice and enterprise control 38:30
Asked about open and custom AI models, Satya Nadella explains that the goal is for each firm to control its own destiny by building both its human capital and what he calls its token capital, meaning every enterprise needs its own learning machine rather than having its knowledge extracted by a model provider. Microsoft's architecture keeps the harness, meaning memory and context, separate from the model itself, so any given model becomes swappable and enterprises can mix frontier models, low cost models, and their own trained models. He points to the Hugging Face incident as a reminder that no enterprise should depend on just one model, since multiple models may be needed to work around problems caused by any single one. Amy Hood adds that Azure is built to deliver the right model for the right job, making the infrastructure itself fungible regardless of which model a customer chooses.
Azure acceleration and capacity constraints 43:00
Asked why Azure growth accelerated to the mid-40s percent range, Amy Hood says supply still lags demand, which shows up even in spot market pricing for compute assets. The acceleration came mainly from efficiency gains across both the CPU and GPU fleet, which get monetized quickly given the current supply and demand imbalance, along with improvements in lead times for getting hardware plugged in and operational across the hyperscale fleet.
Managing oversupply and pricing risk 45:32
On concerns about potential oversupply of data centers and chips, Amy Hood explains that Microsoft's capital spending has shifted toward shorter-lived assets like CPUs and GPUs, which have shorter lead times and can be slowed down if demand changes, while land and building investment is a smaller, more flexible share of the cost structure. A diverse book of business across geography, segment, and industry, plus a large first party app business that also uses capacity, gives Microsoft flexibility to manage demand swings. On component price increases, she says cloud computing still offers better value than customers buying and running servers themselves, and pricing on new contracts is set to preserve value for both customers and Microsoft over the long term. Satya adds that getting the product shape, portfolio mix, and customer mix right, plus running an efficient operation, is what allows Microsoft to stay confident in its margin structure despite cyclical ups and downs.
Copilot adoption and monetization drivers 51:32
Asked about Microsoft 365 Copilot, which passed 30 million paid seats, Satya describes the product shape evolving into a flagship super app combining chat, Copilot Chat, and Copilot code, with usage intensity rising sharply so that time from license purchase to actual usage has dropped from months to days. He says the app is now wired into enterprise governance through Agent 365, covering IT ops, security ops, and finance, as well as into business processes like CRM and ERP systems through skills and plugins, which compounds usage. The business model now combines per seat and usage based pricing, making the addressable market broader than the original Office model. Amy Hood adds that E7's value comes largely from the Agent 365 component, giving enterprises observability and manageability of token spend across business processes, and that this expanding usage and consumption growth will be the main driver of monetization going forward.
Cybersecurity and Project Perception 57:30
Asked about the newly introduced Project Perception and what recent frontier model releases mean for Microsoft's cybersecurity business, Satya says the underlying physics of both cyber products and how companies must operate to protect themselves have changed dramatically. Project Perception applies an intelligence first, model forward approach built around red team agents that continuously search for vulnerabilities, blue team agents that triage findings, and green team agents that fix them, forming an agentic system that runs continuous cyber defense. This system draws on signals from Defender, Entra identity, network, and app security to build the context needed for real protection.
Mixing Models for Cost and Resilience 59:30
On cybersecurity, the point of a multimodel approach is not just savings but resilience. Testing in Cyber Gym showed that the MAI cyber flash one model can match top tier performance at half the cost, because it handles 90 percent of tasks while only the remaining 10 percent need a frontier model. This same mixing of the right model for the right job applies across code, cyber, and knowledge work, and it also protects operations if any single model becomes unavailable.
Question on ROI Trends 1:01:00
Asked how return on capital spending compares to a year ago, and what levers remain such as internal silicon, Amy Hood said her underlying math has not changed much. What has grown is confidence, driven by a larger addressable market and margin levers on both the product and infrastructure sides.
Remaining Margin Levers 1:02:00
She pointed to ongoing work on price performance in silicon, including first party chip investments, plus model diversification to improve token efficiency and cost structure. A broad portfolio spanning knowledge work, coding, security, and the emerging agent layer, referred to loosely as agent 365, adds further opportunity, alongside efficiency gains in Azure from silicon, components, and hyperscale operation. She described this as steady grind work, improving a little every day, and the call closed there.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
