AI is costing a lot. The big four tech companies (Meta, Microsoft, Amazon and Google) are responsible for much of this spending. So much so, that Microsoft has already started taking the first steps towards calling itself a company, primarily, on AI agents.
Company CEO Satya Nadella elaborated on this in his annual letter, explaining that Microsoft now has two different main groups in its ecosystem: AI and infrastructure agents and consumer devices. Agents are no longer a feature within Microsoft products; They are the organizational principles of the company.
Why does it matter? Microsoft will spend about $190 billion on AI in 2026, according to Statista data. Clearly, they don’t want to be left behind in this race, and the company wants to make it clear to its investors what its core businesses are and how things are being organized around AI agents. Therefore, when it comes to revenue, they will be talking about their business in two distinctly separate segments, in this case their AI business and infrastructure and consumer devices.
Nadella said he was “confident in the long-term benefits” of these investments, citing demand and efficiency improvements.
Details. Nadella argues that computing is moving from apps we manage to systems that work for us. His phrase to describe it is that people can now “delegate in the macro and lead in the micro.”
The statistics of the letter show what state the organization is in:
- FY 2026 Revenue: $331.8 billion, up 18%.
- Microsoft Cloud: $214.4 billion, up 27%.
- azure: Revenue surpassed $100 billion for the first time.
- Operating profit: $155.2 billion, up 21%.
- Microsoft 365 Copilot: Over 30 million paying customers.
- GitHub Copilot: 50 million users.
between the lines. The most interesting part is not the agent, but the underlying architecture, as Nadella describes a system where the “instructional environment” (the context, memories, and actions that an agent can perform) is separate from the AI model. In practice, a company can swap one model for another without losing what its agents have learned about its business.
It fits Microsoft’s strategy. They want to stand above both their own models and those of OpenAI, Anthropic and the rest of Frontier AI Laboratories.
Token, baby. The CEO notes that customers will increasingly measure the results they get with the price of the tokens. And to reduce that cost, Microsoft wants to rely on its own chips. Nadella took the opportunity here to talk about his new Maya 300 chip, which according to the manager offers 40% more tokens per watt than mainstream GPU-based systems. He hopes to further strengthen the company’s fleet of first-party silicon by continuing to collaborate with Nvidia, AMD and Intel.
The other half of the company. The “Devices and Consumption” segment suggests that work with AI can move between the cloud and the user’s own computer. Nadella points to Microsoft’s more than 1.6 billion monthly active Windows devices as a basis for driving a strong local model. If it works, he says, AI could effectively become “off-limits” to the user. He also sees opportunities for a new class of “agent-centric” devices.
One more detail. The card uses “Super Intelligence” instead of “AI”. It has a lot to do with President Donald Trump’s executive order on September 29, which ordered all federal agencies to adopt the term. Wow, a rebranding to tell the world they’re up. And it shows how Microsoft and the rest of the company are aligning themselves with Washington’s language.
Picture | Wikipedia
In Xataka | Mark Zuckerberg threatens to leave California over high taxes Meta himself had to negotiate to stay with her
