Satya Nadella says Microsoft will shift from ‘per user’ to ‘per agent’ pricing as AI takes over More work

Microsoft

Microsoft is preparing a major shift in how it prices its software business. CEO Satya Nadella has confirmed that the company is moving away from traditional “per user” licensing and towards a “per agent” model as artificial intelligence systems take on more autonomous tasks inside workplaces. Speaking on the Dwarkesh Podcast, Nadella said Microsoft is evolving from an end-user tools company into an infrastructure provider that supports AI agents capable of completing work and making decisions on their own.

Nadella explained that the logic behind Microsoft’s existing business, including Microsoft 365, is being redefined. The suite already generates around Rs. 8 lakh crore in annual revenue, and Nadella believes it will become the primary environment where AI agents operate. These agents will need secure storage, archival systems, identity management, discovery tools and monitoring layers, all of which sit beneath Microsoft’s existing products. According to Nadella, this new foundation will grow faster than the number of human users because each AI agent will require its own compute resources and security controls.

Microsoft has already begun laying the groundwork. Earlier this year, it introduced a pay-as-you-go model for AI agents, building it on top of the free Copilot chat experience for Microsoft 365 customers. Instead of paying for seats or logins, companies can now pay for the specific work their AI agents perform. Other companies in the industry, including Anthropic and Google, also rely on usage-based billing for their models, but Nadella’s comments suggest Microsoft sees “per agent” pricing becoming a defining part of enterprise software economics.

Nadella urged companies to think beyond tools and focus on the underlying systems that support both human users and AI agents. He said the real shift is understanding that each agent requires dedicated infrastructure such as compute, security controls, identity layers and observability systems. All of this, he said, will be bundled into a new kind of end-user computing infrastructure business, one that continues to expand even if the number of human workers does not.

He also discussed how AI agents are being integrated into Microsoft’s tools, offering Excel Agent as an example. Nadella said the system goes beyond UI-level automation by embedding intelligence directly into the middle layer of Office. The agent learns from markdown-based teaching, allowing it to understand formulas, correct errors and behave like a sophisticated Excel analyst. He noted that competition between model providers ensures that Microsoft can always switch to the most cost-effective model when building such features.

Nadella emphasised that the goal is to build AI agents that act as skilled professionals within Microsoft’s tools. Excel was created for analysts, he said, and the AI agent is being designed as an analyst with built-in knowledge of how to use those tools. This is the direction Microsoft expects productivity software to evolve in, with autonomous agents becoming active participants in everyday workflows.