We keep hearing about the "fierce competition" between AI labs: OpenAI vs. Anthropic, Google vs. Microsoft, Meta vs. NVIDIA. But when you look past the press releases and benchmark scores, the reality is that the industry is rapidly consolidating power, not diversifying. The key players are all locked into a complex web of dependencies that makes the idea of genuine competition an illusion.
The Circular Dependency Problem
Look at the evidence this week:
- Anthropic's $30 Billion Azure Commitment: Anthropic, a primary competitor to OpenAI, commits its future compute capacity to Microsoft, OpenAI's largest investor and partner.
- Meta's Google TPU Deal: Meta seeks to break NVIDIA's monopoly by buying chips from Google. Yet, Meta will still be a massive buyer of NVIDIA, and Google needs Anthropic and OpenAI to run on its cloud to prove its TPU viability.
- NVIDIA's Dominance: NVIDIA sells chips to everyone—Microsoft, Google, Anthropic, OpenAI, Meta—and maintains its lead by ensuring its software (CUDA) is the optimal environment for all their models.
This isn't a healthy ecosystem with independent competitors; it's a troubling concentration of power where a few hyperscalers and one chip manufacturer control all the critical infrastructure—the chips, the cloud, and the capital.
The Rising Barriers to Entry
The early chaotic phase of AI, where anyone with a good idea could compete, is over. The new moat is infrastructure:
- Compute: If you don't have billions, you can't train a frontier model. You have to beg the hyperscalers for capacity.
- Distribution: If you don't control an existing platform (Search, Android, Office), you have to acquire your way in (OpenAI buying Sky) or settle for being a niche provider.
- Regulation: New regulations create costly compliance barriers that only large companies with lobbying budgets can navigate.
The barriers to entry are rising fast, and the window for a small player to disrupt the market is closing.
My Take
The AI industry is structuring itself around a handful of mega-partnerships. Anthropic locking itself into Microsoft, Meta supplementing with Google—these are defensive moves by giants, not evidence of a diverse, thriving market.
We wanted AI to be transformative, and it is—it's transforming power structures and concentrating control into fewer hands. The core technology keeps advancing, but those advances are increasingly happening within walled gardens controlled by a shrinking number of players. The question is whether concentrated power leads to better, more efficient outcomes, or simply to a less free and less innovative future. I lean toward the latter.