AI Is Being Democratized at the Surface and Concentrated at the Foundation
February 21, 2026 by Asif Waliuddin
Microsoft just made agentic AI free. Excel agents. PowerPoint agents. Word agents. Available on the free Copilot tier. Anyone with a Microsoft account gets access.
The same week: Meta committed to a trillion-dollar data center buildout. TSMC raised capex 37% to $56B. OpenAI signed a $10B compute deal for 750 megawatts.
Both sides of the AI debate are using these facts to prove their point. The democratization crowd points to free Office agents. The concentration crowd points to trillion-dollar infrastructure. They are both right. And that is the problem.
Here's what most people are missing:
-- The AI stack has two layers with opposite dynamics. The application layer is commoditizing (Microsoft gives it away to lock in Office users). The infrastructure layer is concentrating (fewer companies control more of the compute, foundry access, and energy).
-- "AI democratization" is technically accurate at the surface. Anyone can use an AI agent in Excel. But you cannot run it on your own infrastructure. You cannot modify the model. You cannot move it to a different provider without rewriting your workflow. You are a tenant, not an owner.
-- This is not a new pattern. Microsoft democratized word processing in the 1990s while concentrating operating system control. The application layer was free or cheap. The platform layer was a monopoly. Same architecture, different technology.
The uncomfortable framing: the more "democratized" AI applications become, the more dependent users become on the four companies that control the infrastructure underneath them. Free applications are not generosity. They are distribution strategy for platform lock-in.
Microsoft is not democratizing AI. They are democratizing the use of AI while centralizing the ownership of it. Those are very different things.
The question for technical leaders: when your company's agentic workflows run on someone else's infrastructure, someone else's models, and someone else's platform -- what exactly do you own?
Follow for more AI Hype vs Reality takes.