Satya Nadella – How Microsoft is preparing for AGI

Satya Nadella discusses Microsoft's massive AI infrastructure build-out, the transition to AI-driven business models, and the company's multi-pronged strategy for navigating competition, in-house model development, and the geopolitical landscape of AGI.

November 12, 2025 5 min read

Satya Nadella – How Microsoft is preparing for AGI — source: https://www.youtube.com/watch?v=8-boBsWcr5A

Satya Nadella outlines Microsoft’s comprehensive strategy for the era of AI, positioning the company not just as a software provider but as a foundational, capital-intensive industrial entity powering the next technological revolution. The approach spans the entire stack, from building the world’s most powerful datacenters to developing a multi-faceted model and application strategy, all while navigating a complex geopolitical landscape.

Building the Future of AI: The Fairwater Datacenter

Microsoft is undertaking a colossal infrastructure expansion, exemplified by its new Fairwater series of datacenters. These facilities are designed for unprecedented scale and flexibility.

Massive Scale: Each Fairwater building is more powerful than any existing AI datacenter, with plans for multiple interconnected sites creating over 2 GW of total capacity. The goal is to achieve a 10x increase in training capacity every 18-24 months.
Interconnected Architecture: The facilities are linked by a one-petabit network, enabling model and data parallelism across buildings and even across entire regions (e.g., connecting to other Fairwater sites in Wisconsin via an “AI WAN”).
Workload Fungibility: This infrastructure is not purpose-built for a single task. It’s designed to be a fungible resource, capable of aggregating massive compute for a single large training job and then dynamically reallocating those resources for data generation, inference, and other workloads. This flexibility is crucial for maximizing the return on capital.

The Transition to AI-Driven Business Models

The shift to AI introduces a new economic reality where the near-zero marginal cost of traditional Software-as-a-Service (SaaS) is replaced by the high Cost of Goods Sold (COGS) of token generation.

Market Expansion Over Margin Compression: Nadella draws a parallel to the server-to-cloud transition. While the cloud introduced COGS for services that were previously pure software licenses, it massively expanded the total addressable market. He anticipates AI will have a similar, if not greater, market-expanding effect, making services like coding assistants a category larger than traditional knowledge work tools.
Diversified Monetization: The core business model levers will remain—subscriptions, consumption, ads, transactions—but they will be adapted. Subscriptions will evolve to become entitlements to a certain amount of consumption, with tiered pricing structures reflecting usage. Microsoft’s portfolio, which already spans all these meters, is well-positioned for this shift.

Competing in the Application Layer: Models vs. Scaffolding

A central theme is whether the ultimate value in the AI stack will accrue to the frontier models or to the application “scaffolding” that integrates them into workflows.

The Scaffolding Advantage: Nadella argues that the platform with data liquidity and deep workflow integration will hold the advantage. The rise of powerful open-source models means that the core intelligence layer risks commoditization. A company that controls the scaffolding can take a capable checkpoint, fine-tune it with proprietary data, and achieve state-of-the-art performance for specific tasks.
GitHub as a “Mission Control”: The strategy for GitHub Copilot illustrates this. Despite intense competition, GitHub’s core platform (repos, issues, actions) remains central. Microsoft is building an Agent HQ or Mission Control on top of GitHub, creating a “cable TV of AI agents” where users can deploy, monitor, and steer agents from Microsoft, Anthropic, Cognition, and others. This makes Microsoft a beneficiary of the entire ecosystem’s growth, not just its own agent.
Deep Integration over UI Wrappers: The value is not in simply wrapping a model with a new UI. It’s in deeply integrating the model into the application’s middle tier. The “Excel Agent,” for example, is not a pixel-level tool but a model taught the native artifacts and logic of Excel, allowing it to reason about and manipulate formulas and data structures directly.

The Future is Agentic: From End-User Tools to Agent Infrastructure

The long-term vision extends beyond building better tools for humans to building the fundamental infrastructure for autonomous AI agents.

A “Per-Agent” Business Model: As AI agents become more autonomous, they will need their own computing resources. Microsoft’s business will evolve from selling “per-user” licenses to provisioning “per-agent” infrastructure.
Infrastructure for Autonomous Work: An autonomous agent will require a provisioned computer (like Windows 365), storage, an identity, security services, and observability. The entire substrate built for Microsoft 365 becomes the bootstrap infrastructure for this new class of non-human workers, dramatically expanding the market for Microsoft’s core services.

Microsoft’s Multi-Pronged Model Strategy

Microsoft is pursuing a dual strategy for model development, leveraging its unique partnership with OpenAI while building its own world-class capabilities.

OpenAI Partnership: The partnership provides Microsoft with access to frontier-class models for the next seven years, which it will continue to integrate deeply into its products and enhance with proprietary data and fine-tuning. Microsoft also gains access to all of OpenAI’s systems-level IP.
Microsoft AI (MAI): The company is building a “superintelligence team” under MAI. The goal is not to duplicate the work being done with GPT but to focus on complementary research and product-optimized models. This includes specialized models for image and audio, as well as cost- and latency-friendly models for specific product features. The next step is to build a first-class omni-model.

Navigating Hyperscale Economics and a Multipolar World

The enormous capital expenditure required for AI necessitates a disciplined, strategic approach to infrastructure deployment.

Strategic Pacing: Microsoft’s “pause” in datacenter expansion was a strategic course correction. Nadella emphasizes the need to avoid getting locked into one generation of hardware (e.g., H100s) when the next (e.g., GB200, Vera Rubin) will have vastly different power and cooling requirements. The strategy is to build in sync with the technology curve (“speed-of-light execution”) rather than building massive, monolithic capacity that quickly becomes outdated.
Sovereign AI and Global Trust: In a world where every nation is concerned with AI, the ability to be a trusted global partner is paramount. Microsoft’s strategy is to meet countries’ demands for sovereignty by making direct foreign investments (building AI factories globally) and providing the technical and policy frameworks (e.g., EU Data Boundary, confidential computing) that give them agency over their data and AI usage. This builds trust in the American tech stack, which Nadella sees as a critical competitive advantage.