Microsoft wiring AI into Excel to power future autonomous agents: Satya Nadella
Nadella described a bold vision: embed AI models directly inside Excel so the spreadsheet becomes a core platform for future autonomous AI agents. In a podcast with Dylan Patel, CEO of SemiAnalysis, and host Dwarkesh Patel, he explained that Excel is evolving from a conventional productivity tool into a fundamental system that AI can use to perform complex tasks without constant human input.
He also offered a first look at Microsoft’s Fairwater 2 data center, signaling a strategic move to build the computing power required for advanced AI across the company’s tech stack. Nadella outlined plans for multiple Fairwater facilities, each housing hundreds of thousands of GB200s and GB300s. When networked, these sites would deliver more than 2 gigawatts of capacity, and even a single Fairwater building would outperform many existing AI data centers.
The conversation touched on what drives success in the AI race. Patel highlighted two schools of thought: one that tools and scaffolding around models determine winners, and another that the models themselves represent the primary value. He pointed to OpenAI’s revenue surge after introducing a competitive code model and Anthropic’s jump in inference margins—from well under 40% to over 60%—even with more competitors entering the field.
Nadella pushed back on the idea that firms can simply “wrap a model” and turn a quick profit, arguing that modern models are far more capable. He cited Excel Agent as an example: rather than placing AI solely at the user interface, Microsoft is embedding AI in the mid-layer of Office 365. Using GPT-family technology, the system learns Excel’s internal structure, identifies formulas, corrects errors, and works directly with Excel’s built-in tools. He likened this to equipping AI with a detailed playbook for how sophisticated Excel users think and operate. In this arrangement, Excel effectively ships with an AI analyst built in.
Microsoft’s tools as the foundation for autonomous agents
Nadella noted that fierce competition among model providers will empower developers to switch models if prices rise. He drew a parallel to the competition among large cloud providers, with open-source models acting as a balancing force. Microsoft intends to stay in the hyperscale arena, continuing to support multiple models. It will retain access to OpenAI models for seven more years, pursue its own MAI models, and build model-focused application layers for security, knowledge work, coding, and science. These applications will feature deep, integrated models rather than mere add-ons.
Microsoft’s AI analyst concept
Dwarkesh Patel asked whether future AI models might eventually run computers as well as human workers, potentially removing the need for deep integration. Nadella responded that Microsoft is already building an AI analyst capable of using the tools people rely on. The company envisions two futures: one where humans still guide tools like Excel, and another where organizations allocate computing resources to fully autonomous AI agents that have built-in access to the same tools and operate more efficiently.
Nadella concluded by noting that Microsoft’s current business of providing tools for human users will gradually shift toward an infrastructure layer for autonomous agents that perform work independently. Core Microsoft 365 functions—storage, archiving, search, and management—will remain essential even when AI becomes the primary user. In essence, this signals the emergence of a new kind of infrastructure to support a future driven by autonomous AI.
)
