The barriers to entry for LLM are obvious: as you pointed, the field is extremely capital intensive. The only reason there are seemingly multiple players is because the amount of capital thrown at it at the moment is tremendous but that's unlikely to last forever.

From my admittely poorly informed point of view, strategy-wise, it's hard to tell how wise it is investing in foundational work at the moment. As long as some players release competitive open weight models, the competitive advantage of being a leader in R&D will be limited.

Amazon already has the compute power to place itself as a reseller without investing or having to share the revenue generated. Sure, they won't be at the forefront but they can still get their slice of the pie without exposing themselves too much to an eventual downturn.