Satya Nadella on Microsoft's Strategy in the AGI Era
In 2025, Microsoft CEO Satya Nadella clearly articulated the company's structural transformation in the AI era.
The setting was the new 'Fairwater 2' data center in Atlanta—one of the world's most powerful AI training facilities. The interviewers were Dylan Patel, founder of SemiAnalysis, and Dwarkesh Patel of the economics podcast.
This article dissects Microsoft's strategic thinking regarding 'AGI (Artificial General Intelligence)' as it emerges from Nadella's remarks.
1. Infrastructure Revolution: AI Data Centers Evolving at 10x Speed
Nadella first stated, 'We are increasing our training capacity by 10x every 18 to 24 months.'
The fiber optic connectivity at Fairwater 2 is equivalent to the entire Azure network from two and a half years ago. A petabit-scale network spanning multiple regions comprehensively supports AI model training, inference, and data generation.
1-1. 'Design for Scale'
Nadella notes, 'To accommodate future massive models, we are designing them to be trainable by bundling multiple regions.'
Furthermore, Fairwater 4, currently under construction, will also be connected on the same network, linking with the Wisconsin facility via an AI WAN (Wide Area AI Network).
This structure makes 'model parallelism' and 'data parallelism' a reality.
1-2. Flexibility to Optimize for Each Era
Next-generation chips (e.g., Nvidia GB200, Vera Rubin Ultra) have different power density and cooling requirements.
'That is why you shouldn't build it out with a single specification. You need a flexible scale that can expand with the passage of time,' Nadella emphasizes.
Microsoft's infrastructure strategy is not 'single-generation optimized,' but 'continuous-update optimized.'
2. The Economic Model of AGI: A Tool or an 'Intelligent Agent'?
Nadella's view on AGI stands apart from the typical 'AI panacea' theory.
He says, 'AI is a "guardian angel" and a "cognitive amplifier."'
In other words, he positions it as a tool to complement and expand human cognitive abilities.
2-1. The Metaphor of 'Satya Tokens'
During the conversation, when the host joked, 'In the future, models will generate "Satya tokens," and they will have value,' Nadella smiled and replied:
'If I were operating as an API, you might not be able to afford the cost.'
This exchange hints at the economic value of AI (token unit price) and the scarcity of human knowledge.
2-2. Redefining the SaaS Model
SaaS (Software as a Service) was traditionally a model where 'costs do not increase even if the number of users grows.'
However, in the AI era, inference costs (COGS) are skyrocketing, causing simple scaling to fail.
Nadella stated, 'Business models in the AI era will shift to "consumption-based subscriptions,"' defining tiered plans like Copilot ($20/month) as 'consumption rights.'
3. The Future of the 'AI Software Factory' as Shown by GitHub Copilot
Nadella positions GitHub not merely as a development platform, but as an 'AI software factory.'
In 2024, GitHub Copilot reached 26 million subscribers.
Nadella says, 'One developer joins GitHub every second.'
3-1. The Emergence of Competition is 'Healthy'
When asked about the rise of competitors like Claude Code, Cursor, Cognition, and Replit, Nadella asserts:
'It is a good sign that new competitors are appearing. It is proof that the market is expanding.'
The AI coding market has expanded to a $50 billion scale in just one year.
Nadella says, "If our market share drops, I don't mind as long as the entire market grows tenfold."
3-2. The Next Battlefield: "Agent HQ" and Multi-Agent Control
Microsoft is envisioning "Agent HQ" on GitHub.
It is a "mission control" mechanism that can collectively control and monitor multiple AI agents (such as Claude, Grok, and Cognition).
Nadella describes this as a "cable TV-like subscription," envisioning a new OS-like ecosystem consisting of VS Code + GitHub + an AI control layer.
4. Model Competition and the "Scaffolding" Strategy
The battle for supremacy in the AI model layer is intensifying. Anthropic's inference margin exceeds 60%, and OpenAI and Google are also growing rapidly.
However, Nadella believes that "the value lies not in the model itself, but in the 'scaffolding' that wraps it."
4-1. The Symbolism of the "Excel Agent"
The Excel Agent is not just a UI wrapper.
It is an "intermediate layer model that can understand cell formulas and self-correct errors," a structure that integrates AI into the core of Office.
"The Excel of the future will be a tool that comes with an analyst included."
In other words, AI is becoming intelligent by being embedded into the internal structure of applications, rather than existing outside of them.
4-2. The Agent Economy and the "One Person + One Agent Era"
Nadella predicts that "business per user will shift to business per agent."
In the future, companies will have a structure where they manage an equal number of human employees and AI employees (agents).
Microsoft 365 and Windows 365 will become the "infrastructure for those AI employees."
5. MAI Plan: The "Second Intelligence Team" Running Alongside OpenAI
While maintaining its seven-year exclusive contract with OpenAI, Nadella has established an independent research organization, "Microsoft AI (MAI)."
He recruited Mustafa Suleyman (founder of DeepMind) and others to form a "team aiming for superintelligence."
"While making the most of OpenAI models, we will generate breakthroughs through our own research."
MAI is already ranked 9th in the world for image models and has implemented voice models into Copilot.
The next step is to develop an "omni-model" that integrates text, images, and audio.
6. Transformation into a Hyper-Capital-Intensive Company: Microsoft as a Factory
"We are no longer just a software company. We are a knowledge-intensive and capital-intensive company," says Nadella.
Microsoft's CAPEX for AI has tripled over the past two years.
Even so, Nadella asserts, "We will increase capital efficiency through knowledge."
"Our job is to optimize cost efficiency per token (tokens per dollar per watt)."
The company is deploying its own chips (Maia and Cobalt) and aims to optimize TCO through collaboration with Nvidia and IP sharing with OpenAI.
The cloud is now a "data factory," and Microsoft is evolving into an "industrial company that generates revenue using knowledge as raw material."
7. The Political Economy of Global Trust: Responding to the Era of AI Sovereignty
Finally, Nadella states, "Trust is more important than technology."
"As long as the world trusts American technology, American companies will continue to win."
In an era where AI is directly linked to national security, Microsoft is deploying "Sovereign Cloud" solutions that align with each country's data sovereignty and legal regulations.
In Europe, it has adopted the "EU Data Boundary," and in Japan and India, it is employing a model of local data centers combined with government partnerships.
The geopolitical position of being a "trusted American company" is its greatest competitive advantage.
Conclusion: AGI is not the "final revolution," but a continuous industrial process.
Nadella views the AI revolution not as an "end point," but as a "continuation of industrial evolution."
Railroads, electricity, the internet, the cloud—and now AI.
Just as the revolutions of each era have accelerated economic growth, AGI will also become an "augmentation tool" that redefines productivity and creativity.
A future where "AI does not replace humans," but rather,
"AI amplifies humans."
That is the core of the "Microsoft in the age of AGI" that Nadella envisions.
