Microsoft CEO Satya Nadella’s 2026 Shareholder Letter: Building the “Frontier Ecosystem” and “Token Cost vs. Business Outcome” as the New Software Economics

As we enter the “era of Super Intelligence,” Microsoft’s central question is not how large the opportunity is, but how to turn technology into real economic results. Its approach is to build a “frontier ecosystem”: one where the value others create on Microsoft’s platform exceeds the value Microsoft keeps for itself. At the same time, everything is optimized around a new metric: “cost-to-outcome” — how much business result each unit of token spend can deliver. Inference costs have made marginal cost a key variable in software economics once again. https://linkedin.com/pulse/my-annual-letter-building-frontier-ecosystem-satya-nadella-iujec/…

Financial Results (Fiscal Year 2026) Revenue of $331.8 billion, up 18% year over year; Microsoft Cloud revenue of $214.4 billion, up 27%; Azure growth of 41%, surpassing $100 billion for the first time; operating income of $155.2 billion, up 21%. Microsoft 365 Copilot paid seats exceeded 30 million, and GitHub Copilot users reached 50 million.

Strategic Takeaways: Model-Forward, Agent-First; Financials Reorganized into Two Segments

Agents and Infra (Agents and Infrastructure) · The interaction model is shifting from synchronous chat to asynchronous agents — in his words, “macro delegate, micro steer”: people define the goal, the system executes, and returns for direction when needed. Microsoft 365 Copilot has already pivoted to asynchronous agents capable of completing long-running tasks; the new Copilot app unifies chat, collaboration, coding, and autopilot-style tasks into a single entry point. Taken together, the apps are positioned as a “new operating system for work.” · Harness and model separation: context, memory, and executable actions all live in the harness, so models can be swapped at any time. Selecting the right model for the right task also strengthens business continuity. Foundry offers more than 11,000 models (OpenAI, Anthropic, Mistral, SpaceXSI, and the in-house MAI family) with over 100,000 customers. · Customers own their own “learning loop”: he argues that you can replace a “generalist” model without losing the “veteran employee”-style experience accumulated within the enterprise — and that this is the test of control and sovereignty in this era. Microsoft IQ serves as the context engine, with Work IQ covering more than 17 EB of email, document, and meeting data; Agent 365 acts as the agent control plane, managing agents the way you’d manage employees, complete with FinOps cost management. · Infrastructure: added 88 datacenters this fiscal year and completed the Fairwater superfactory; the in-house Maia 300 chip delivers 40% more tokens per watt than mainstream GPU systems, and Cobalt 200 improves performance by over 50%. The share of in-house silicon in fleets will keep growing.

Devices and Consumer Inference will switch automatically between cloud and endpoint. As local models grow more capable, “intelligence is expected to become unbilled by volume.” The Windows installed base of 1.6 billion monthly active devices provides the deployment foundation, driving stronger PCs and a new generation of “agent-first” devices. LinkedIn has 1.3 billion members; Xbox has 1 billion annual players.

Responsibility and Culture · Responsibility: upgraded Responsible AI standards and partnered with AI safety agencies in the U.S. and U.K.; phishing-resistant multi-factor authentication coverage reached 99.97%; fixed more than 550,000 high-severity vulnerabilities; electricity matched 100% with renewable energy; using Quincy, Washington — where datacenters are located — as an example of community-wide benefit. · Culture: moving beyond “growth mindset” toward “becoming a learner again,” while doing three things at once: running critical systems well, continuously evolving, and inventing the next generation of products. Organizationally, the company operates around the work itself rather than the org chart, establishing Microsoft Frontier Co. and deploying 6,000 experts on-site to co-create with customers. The letter closes with a line from Horace: “Carpe diem, seize the day.”

My annual letter: Building the frontier ecosystem

Source: https://www.linkedin.com/pulse/my-annual-letter-building-frontier-ecosystem-satya-nadella-iujec/ Below is my annual letter, published today in our Annual Report 2026 (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fwww%2Emicrosoft%2Ecom%2Finvestor%2Freports%2Far26%2Findex%2Ehtml&urlhash=B18m&trk=article-ssr-frontend-pulse_little-text-block):

Dear shareholders, colleagues, customers, and partners:

For Microsoft, the central question as we enter this Super Intelligence era is not just how big the opportunity becomes, but how we deliver the real-world outcomes that every country, community, and company demands and deserves. That means aligning everything we do to convert technological progress into economic benefit for everyone, everywhere.

Microsoft is uniquely positioned to do this. We have a broad tech stack, diverse businesses and business models, and a platform- and partner-first approach in our DNA. That’s why we are building a frontier ecosystem (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2Fsnscratchpad%2Ecom%2Fposts%2Ffrontier-ecosystem%2F___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6NzozZThiOjNkMzE3NjgzYjFmZjVlNWE0OGE3NmY1NmE2MzcwYWQ0N2RlZmJkNWJiNTBmMmUxOTdkNzAwZTIwZTA2MTQxNGE6cDpUOkY&urlhash=vuJZ&trk=article-ssr-frontend-pulse_little-text-block) that enables others to build more value on top of and through the ecosystem than is captured inside. This is how we create broad economic surplus and deliver durable returns for investors.

After all, Super Intelligence is changing every organization’s inputs (how they work) as well as their outputs (the products and services they sell). At the same time, inference is making marginal cost an important factor in software economics as usage scales. As a result, customers will increasingly measure the cost of tokens against the outcomes they produce. We call this the “cost-to-outcome” frontier, and we are focused on optimizing it.

This shift expands Microsoft’s opportunity across every layer of the stack, from apps and agents that compress and complete tasks, workflows, and processes; to models that power agent reasoning and actions; to infrastructure that trains and serves those models and agent loops in the most efficient way. Any high-value activity or process that can be made faster, less expensive, or more effective through intelligence becomes part of that opportunity.

Based on the demand we see and the efficiencies we are driving, I am confident in the long-term returns on our investments. That confidence starts with being disciplined about what we build and who we build it for. We are investing to serve a diverse set of customers, while also creating the infrastructure to power our own differentiated and high-value first-party products like Copilot. We are building against a growing book of business tied to this secular shift.

We are bringing capacity into service faster, increasing utilization, and continuously improving performance and cost at every layer of the stack and across the stack as a whole. At the same time, we are evolving our business models to reflect the value and usage of our services. Together, these efforts help our customers apply technology to innovate and grow, while strengthening our ability to deliver long-term value to shareholders.

This is already translating into strong financial performance (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fwww%2Emicrosoft%2Ecom%2Fen-us%2FInvestor%2Fearnings%2Ffy-2026-q4%2Fpress-release-webcast___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6Nzo2ZWU4OjNhMTE2MThkM2E4YWZmYzdhNzMyNGIxZjVkNmVhMTY2YWIyNmRkNDA1OTUyZWIwYTg4ODVjMjY2NThjMTQ1ZTA6cDpUOkY&urlhash=JKtR&trk=article-ssr-frontend-pulse_little-text-block). Fiscal year 2026 was another record year for the company. Revenue was $331.8 billion, up 18 percent year-over-year. Microsoft Cloud revenue was $214.4 billion, up 27 percent year-over-year. Azure grew 41 percent year-over-year and surpassed $100 billion for the first time, while Microsoft 365 Copilot reached over 30 million paid seats. And operating income was $155.2 billion, up 21 percent year-over-year.

As we look ahead, we remain grounded in three things: the why of our mission, the what of our strategic priorities, and the how of our culture.

OUR WHY

Our success depends on our customers’ success. We want every organization to own its own continuous learning system so it retains its core IP. And we want to empower every person, amplifying their agency and ambition.

This year, we made meaningful progress in defining this vision and building the foundation to realize it, and have seen many successful examples of intelligence diffusing across the world:

• In North America, Kraft Heinz (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fblogs%2Emicrosoft%2Ecom%2Fblog%2F2025%2F10%2F28%2Fbecoming-frontier-how-human-ambition-and-ai-first-differentiation-are-helping-microsoft-customers-go-further-with-ai%2F&urlhash=8q4a&trk=article-ssr-frontend-pulse_little-text-block) put an intelligence platform on the factory floor across its North American sites, helping drive a 40 percent decrease in supply-chain waste, a 20 percent increase in sales forecast accuracy, and a 6 percent improvement in product yield.

• In Central America, Costa Rican dairy cooperative Dos Pinos (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fnews%2Emicrosoft%2Ecom%2Fsource%2Flatam%2Ffeatures%2Fai%2Fcosta-rica-dos-pinos-ai-agents-microsoft-copilot-en%2F%3Flang%3Den&urlhash=VA73&trk=article-ssr-frontend-pulse_little-text-block) is using a fleet of agents to decrease packaging-review errors by more than half, cutting time to market by roughly 10 days while enabling employees to focus on higher-value work.

• Across Europe, WWF divers are hunting abandoned fishing nets with GhostNetZero.ai (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Funlocked%2Emicrosoft%2Ecom%2Fghostnetzeroai%2F&urlhash=8Qh_&trk=article-ssr-frontend-pulse_little-text-block), built with our AI for Good Lab, to read sonar with 94 percent accuracy and turn hours of review into minutes.

• In Asia, Air India (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fwww%2Emicrosoft%2Ecom%2Fen%2Fcustomers%2Fstory%2F26047-air-india-azure-openai-in-foundry-models&urlhash=rvc6&trk=article-ssr-frontend-pulse_little-text-block) has resolved more than 13 million customer conversations with an agent at a 97 percent success rate.

• Engineers at Beca (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fnews%2Emicrosoft%2Ecom%2Fsource%2Fasia%2Ffeatures%2Fpairing-geotechnical-data-with-ai-helps-new-zealand-to-build-better%2F&urlhash=9Lju&trk=article-ssr-frontend-pulse_little-text-block) in New Zealand are using Azure, Foundry, and their BEYON platform to make the National Geotechnical Database more accessible, so today’s buildings can be designed with the lessons learned from the 2011 Christchurch earthquake.

• And in Africa, route planners at Coca-Cola Beverages Africa (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fcommunity%2Edynamics%2Ecom%2Fgalleries%2Fgallery-posts%2F%3Fpostid%3Db3d3f6c0-6c03-f111-8407-7ced8dde4396&urlhash=1sA_&trk=article-ssr-frontend-pulse_little-text-block) are getting up to 90 minutes back every day as route-planning agents run across 15 markets and 10,000 users.

OUR WHAT

Our opportunity

Realizing this opportunity at scale requires us to build differently across every layer of the tech stack. That starts with making strategic choices about where we invest, what we build, and how each part of our portfolio comes together.

The tech stack we are building is model-forward and agent-first. Models will continue to become more capable, while agents will take increasingly complex workflows through to completion. As they do, our ambition is to deliver control, capability, choice, and cost advantages that compound over time for our customers and for us. And through our support for business models that span subscriptions, consumption, advertising, and hardware, we can help make Super Intelligence affordable and accessible so more people and organizations around the world can use it in the way that’s best for them.

Let me walk through how we are approaching this across our two new reporting segments: Agents and Infra and Devices and Consumer.

Agents and Infra

Computing has historically been interactive. A user moved from app to app or tool to tool to get something done. Today, it is becoming more adaptive and proactive. With agents, we can now macro delegate and micro steer. In other words, we define a goal and the system works on our behalf, returning for direction and judgment when needed.

It’s why we evolved Microsoft 365 Copilot this year, from synchronous chat assistants to asynchronous agents that can complete complex, long-running tasks. We surpassed 30 million paid seats and saw strong early adoption of our new E7 suite (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fblogs%2Emicrosoft%2Ecom%2Fblog%2F2026%2F03%2F09%2Fintroducing-the-first-frontier-suite-built-on-intelligence-trust%2F___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6NzphMGI1OjJlMjhiNjcyNDE4NWVhMmRkNmY1NTFmYTEyYjc0ZTMxMDdhODBkNWU5OGJlYTE4NGY0N2Q4ODY2ODlmZGQxYmE6cDpUOkY&urlhash=pV8b&trk=article-ssr-frontend-pulse_little-text-block).

In software development, GitHub Copilot now has 50 million users, and we worked to scale the platform to meet unprecedented demand.

In security, we introduced Project Perception (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fblogs%2Emicrosoft%2Ecom%2Fblog%2F2026%2F07%2F27%2Frethinking-security-for-the-age-of-ai%2F___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6NzpkMTk2OmM1YTU4MzNiMDFiYTc2ZjVjMGE5MGRiY2VmZTE4MmU4MGU5MzA1NTc2MzRiNjUyYzM3NGI2MTc2NWVjYTA2Yjc6cDpUOkY&urlhash=Czzu&trk=article-ssr-frontend-pulse_little-text-block), a multimodel agentic security system that uses agents to simulate attacks, identify threats, and implement fixes.

And in healthcare, we are on pace this calendar year to document more than 100 million patient encounters with Dragon Copilot (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fwww%2Emicrosoft%2Ecom%2Fen-us%2Fhealth-solutions%2Fclinical-workflow%2Fdragon-copilot___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6Nzo3MTgyOjY3MzUwMDYyMzM0YzBkYjQ0MzhhMjJjYjRkMDkyZmJiMTdmNjE2ZDI5NTg2ODkxNGI1N2E5NjkzYzRmZGViNjI6cDpUOkY&urlhash=FX8P&trk=article-ssr-frontend-pulse_little-text-block), helping clinicians automate workflows so they can spend more time with patients, not paperwork.

We increasingly think of all our apps together as a new operating system for work that transcends traditional software categories to understand user intent and act on it. As our apps move toward agent-first capabilities that can be called and orchestrated in the flow of work, the system becomes more valuable and more deeply embedded. Just last month, we announced the first version of this, a new Copilot app (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fblogs%2Emicrosoft%2Ecom%2Fblog%2F2026%2F09%2F25%2Fintroducing-the-new-copilot-with-home-code-and-autopilot%2F___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6Nzo2M2Q4OjJkNzAwNDYwNWMwNDQ1MTI0ZWQ5YTYyMGUwYjQ3M2Q4NWY5MThiM2RhNTRkNTRkZDM3ZGY2NzFkYjUwN2ViY2E6cDpUOkY&urlhash=PRKj&trk=article-ssr-frontend-pulse_little-text-block) that brings chat, cowork, code, and autopilots into one unified experience, bringing this intelligence into the flow of human ambition at home and at work.

Underlying all this is a new multimodel system where the harness — including context, memory, and action space — is separate from the model. This architecture optimizes the cost-to-outcome frontier by allowing us to use the right model for the right job. And it strengthens business continuity and resilience because models become interchangeable.

That’s why we offer the broadest model catalog in the cloud, with more than 11,000 models from OpenAI, Anthropic, Mistral, SpaceXSI and others, as well as our own MAI family (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fmicrosoft%2Eai%2Fmodels%2F___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6Nzo2NjIwOmZkMzkxNTNjYjYwYjg3NmVjYWI4ZWI2NGM0NWUyNTVhNjgwNzY0NDBlNjFmNGZmOTRhNmU1ZTNmMjBhMWU3NTQ6cDpUOkY&urlhash=B8nV&trk=article-ssr-frontend-pulse_little-text-block), which we expanded with more than a dozen new models this year.

Microsoft Foundry brings these models together with knowledge, tools, and enterprise controls so organizations can build their own learning loops that know their business context and take action safely at scale. This year Foundry surpassed 100,000 customers.

The world’s most valuable data resides inside customers’ systems. Every customer should be able to use that data to build agentic systems that improve continuously while retaining control of their IP (https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fprotect%2Echeckpoint%2Ecom%2Fv2%2Fr01%2F___https%3A%2F%2Fsnscratchpad%2Ecom%2Fposts%2Freverse-information-paradox%2F___%2EYzJ1OndlY29tbXVuaWNhdGlvbnM6YzpvZmZpY2UzNjVfZW1haWxzX2F0dGFjaG1lbnQ6MzRlODVhOWMxZjY1MTBlMjQ0MDkzMjNkZWMxNzYzMDQ6NzplZmIwOjc1OGRhOGMzNDFhZDY0OTk1ZTMwNDg4MDhiZjZkNDBmNmQ2MzM3ZjdlYWMwNTQ0ZTc3ODNiNDg2ZDFkOTk5ZmM6cDpUOkY&urlhash=WqEI&trk=article-ssr-frontend-pulse_little-text-block). And they should be able to switch out a “generalist” model