How the company works
What it sells, how it makes money, and the anatomy of the extremes: what has to break for the result to land far from the forecast, either way. In Microsoft's own words, from its annual report for fiscal 2026.
Microsoft sells software and cloud computing to businesses and people: Office and Windows by subscription, Azure's data centers and AI models by the hour, and LinkedIn, Xbox and search advertising beside them.
How it makes money
Mostly subscriptions and cloud usage paid by businesses. The Microsoft Cloud, Microsoft 365 Commercial cloud, Azure, Dynamics 365 and the commercial part of LinkedIn, brought $214.4B in fiscal 2026, up 27%. Revenue for the year to June 2026 was $331.8B, up 18%, and operating income rose 21%. Azure and other cloud services grew 41%.
What it sells
| Business, fiscal 2026 | Revenue | Growth | Operating income |
|---|---|---|---|
| Productivity and Business ProcessesMicrosoft 365 for businesses and people, Copilot, LinkedIn, Dynamics | $140.0B | 16% | $83.9B |
| Intelligent CloudAzure, GitHub, SQL Server and Windows Server, enterprise services | $137.8B | 30% | $57.0B |
| More Personal ComputingWindows licenses for PC makers, Surface, Xbox, Bing search ads | $54.1B | −1% | $14.4B |
Segment revenue and segment operating income as Microsoft reports them; the growth is against fiscal 2025, as the annual report prints it.
Unit economics
Scale on shared data centers is the lever: Microsoft names three economies of scale, cheaper computing per unit in large data centers, demand pooled across customers and regions, and shared upkeep. Once a customer is on the platform, each added user or hour of computing costs little. The new cost is AI: additions to property and equipment were $115.9B in fiscal 2026, against cash from operations of $182.9B, and the Microsoft Cloud gross margin fell to 66% as AI capacity and usage grew. Research and development was $35.6B; 223,000 employees.
Competitive position
Leader in business software and one of the largest clouds. Its edge is the installed base, Office and Windows in most companies, sold as one platform with security and identity, and a commercial remaining performance obligation of $678B, up 84%, approximately 30% of it due within 12 months. Its AI rests partly on OpenAI, in which it holds an approximate 25% interest and rights to the models: OpenAI was $24.1B of revenue in fiscal 2026. The flip side is that the AI rivals it names, hyperscalers and AI-first companies, are often also its partners.
Biggest cost inputs
Data centers: servers and graphics processing units from few qualified suppliers, electrical power, land and leased capacity, and their depreciation. Then model training and inference, payroll for 223,000 employees, and the traffic acquisition costs of search advertising. Cost of revenue rose 21% in fiscal 2026, faster than revenue, on AI infrastructure.
Anatomy of a large upside
- AI runs on Azure. Companies move their AI work to the cloud that holds their data; Azure, up 41%, keeps growing on capacity already paid for, and the $678B of contracted revenue turns into sales as the data centers come on.
- Copilot raises the price of every seat. Microsoft 365 Commercial cloud grew 17% on revenue per user from Copilot and E5; if agents become part of how every office works, revenue per user climbs on the same installed base and margin follows.
- The spending pays at software margins. Efficiency gains in Azure and Microsoft 365 outrun the cost of AI capacity, the Microsoft Cloud gross margin turns back up, and operating income grows faster than revenue.
Anatomy of a large downside
- The build-out runs ahead of demand. Capital spending of $115.9B is built for AI demand; Microsoft warns that the cost of AI is subject to significant uncertainty, and if prices fall to competition or usage slows, depreciation stays and margins fall.
- OpenAI goes its own way. A partner that brought $24.1B of revenue and owes $6.0B also competes with Microsoft; a change in the contract, its strategy or its health hits Azure's growth and the investment.
- Power and parts set the ceiling. Data centers need electricity, land, permits and components from few suppliers; if capacity comes late, demand goes to rivals and the backlog waits.
Written from Microsoft's annual report for fiscal 2026, in its words and numbers. The extremes are the long view, what has to break for the result to land far from the forecast; the quarter's own cases, each model's, are on the quarter forecast.