If you think there's no Manhattan Project for AI running in parallel, you're naive.

The question isn't whether OpenAI, Anthropic, xAI, or anyone else wins. The question is whether a serious government would allow its most strategic capability to depend on a company, founder, board, cloud provider, or vendor it doesn't fully control.
Of course it won't.
You don't outsource nuclear command systems. You don't outsource intelligence collection. You don't outsource the technologies that determine national power.
The game theory is straightforward. If AI capabilities diffuse quickly, building sovereign models becomes cheaper over time. If capabilities compound and create durable advantages, dependence becomes existential.
In both worlds, the answer is the same: build your own.
The commercial labs push the frontier. The state learns from them, recruits from them, partners with them, and funds them.
But I'd be shocked if it wasn't quietly building its own stack underneath the entire industry.
There are only 2 serious players: US and China, everyone else is irrelevant. Someone was going to make a move under the guise of national security.