I'm not sure about that to be honest.

1. Financial perspective. People are too focused on what "the best" is, rather than what is the most financially viable at scale, which is what really matters. At the peak of the Ethereum mining craze in 2016/2017 the GTX Titan X was the best performing GPU. But buying 200$ AMD Polaris GPUs was what gave you the most performance per $ and per watt.

2. Open source models keep being impressive and lagging only so much behind the closed source ones. It's hard to predict the future, but few years from now the most viable application might be to internally fine tune and deploy on whatever cloud or internal infra open source models. I have already many use cases in prod where Gemini Flash 2.0 did a great job, and that's an old model by today's standards (summarizing news/translation). Now I have in production a service that reviews pull requests and updates documentation/JIRA accordingly when they are merged. That requires quite more plumbing, agentic approach and thinking, but yet again open source models can do a terrific job already there.

3. At the end of the day, the lion share is going to be eaten by whoever provides the best applications, not models, but conversely we're also living in a space where more and more you can just build roughly-the-same-feature with few $ worth of APIs.

4. Even more, the biggest benefit will lie among those who will leverage AI in the best way. Companies and individuals able to really delegate successfully complex tasks making crazy savings. Who knows who's really gonna take the biggest advantage. Maybe US companies, maybe not.

Thus, in essence, I envy your certainties around the future, I personally have lots and lots of doubts and have no clue who's gonna eat whoever's lunch.