Nadella details Microsoft’s AI growth in annual letter as Azure tops $100 billion

In his October 21, 2026 annual letter, CEO Satya Nadella paired that argument with fiscal 2026 figures: Azure exceeded $100 billion in revenue, and Microsoft 365 Copilot passed 30 million paid seats.

From selling access to measuring completed work

Nadella’s starting point is an economic change in software. Running AI models carries a cost as usage grows, so customers will increasingly weigh the price of tokens—the units models process—against the results they produce. He calls this the “cost-to-outcome” frontier, making useful results the stated target for Microsoft’s investment across applications, models and infrastructure.

The response he describes starts with changing what software does. Microsoft evolved its work Copilot during the year from assistants that answer in a live exchange toward agents that can carry out complex, longer-running tasks. His model of delegation still gives people a role: users set the goal, while the system returns for direction and judgment when needed.

The letter presents that approach through two reporting segments: Agents and Infra, and Devices and Consumer. Nadella describes the technology strategy as model-forward and agent-first, with models supplying capabilities and agents taking workflows through to completion. The consumer side rests partly on his expectation that more AI processing will move between cloud services and users’ devices.

$331.8 billionRevenue

Microsoft reported fiscal 2026 revenue growth of 18% year over year.

$155.2 billionOperating income

Operating income increased 21% year over year, according to the letter.

Adoption reaches beyond the work assistant

The annual figures show the scale supporting that strategy. Microsoft Cloud revenue reached $214.4 billion, up 27% year over year. Azure grew 41%. Alongside the paid-seat count for Microsoft 365 Copilot, Nadella disclosed 50 million GitHub Copilot users. Those measures describe different things: the letter labels one as paid seats and the other as users.

For companies building their own AI systems, Microsoft Foundry surpassed 100,000 customers, Nadella said. Its cloud catalog contained more than 11,000 models. Foundry combines models with organizational knowledge, tools and enterprise controls, forming the platform through which Microsoft wants customers to build systems that learn from their own business context and take action.

Microsoft also added 88 datacenters during the year. Nadella says the infrastructure effort is not simply about adding capacity: it includes bringing capacity into service faster, increasing utilization and improving performance and cost throughout the system. He expressed confidence in long-term investment returns, citing demand and the efficiencies Microsoft is pursuing.

The next test is keeping knowledge portable

Nadella’s proposed mechanism separates the model from the surrounding system that supplies context, memory and permitted actions. That design, he argues, lets Microsoft choose the right model for each job while making models interchangeable. The claimed benefits are lower costs for a given outcome and greater continuity when the underlying model changes.

For customers, he sets a more demanding standard than access to a broad model catalog: an organization should be able to replace a general-purpose model without losing the expertise its system has accumulated. Nadella argues that customers should retain their intellectual property and capture the value of their own learning systems, rather than surrender it to an outside provider.

His consumer outlook extends the economic question further. More capable models running locally could make some intelligence effectively unmetered, he writes. That remains a forward-looking proposition, not a promise of free access: the letter anticipates business models combining subscriptions, usage charges, advertising, commerce and local computing to support adoption.