Who will win the AI race? Chip Makers, US AI Labs, Open AI Labs

Here's my overall thesis working in the tech/AI industry for last 15 years. And now working closely with large AI systems in enterprises.

Based on my observations, the ultimate winner of this AI race is going to be either the chip sellers (NVDA, TSMC, MU, etc) or the first AI Lab to AGI. If not AGI, LLMs will just be commoditized.

I don't think we are close to AGI. The models are improving for sure, but I don't quite think it has all the ingredients required for AGI yet.

Here's what I think will happen in the next 3 years.
– The AI Data Center infrastructure spend will keep on increasing. Once the US gets saturated, the hyperscalers will explore other geographies.
– A large % of this infrastructure will initially be used for training which will slowly move towards more and more infer.
– The AI Labs will keep releasing new improved models every 6 months or so, until the improvements become negligible.
– The open (mostly Chinese) AI Labs will keep distilling SOTA models and releasing them 4-6 months after SOTA model is released from one of the leading US based AI Lab.
– The US AI Labs need to charge premium for R&D amortization, safety research, RLHF alignment work, and profit margins which open AI Labs do not. Due to this US AI Labs models will be significantly more expensive than open models.
– For example, Kimi K2.6 leads on agentic and coding benchmarks – it tops SWE-Bench Pro at 58.6% (vs GPT-5.4's 57.7%). K2.6 is approximately 17x cheaper on input and 12x cheaper on output than GPT-5.4. For a team processing 100M tokens/month, that's roughly $100 (Kimi) vs $1,500 (GPT-5.4). 
– Ultimately, marginal intelligence improvements will not be worth paying premium for (e.g. new iPhones are marginally better). So, bulk of the inference usage will be driven on open models. This is already happening. For example, Cursor shipped a an open model as their own likely with some fine tuning.
– Infrastructure companies will continue to win since they are still needed for inference but the US AI Labs may not.

Of course, all 3 will "win". The question is as investors, where can we get highest ROU from. If you agree with this thesis, where should we be investing?