> The moat is in data, training algorithms, compute capacity, and UX.
That’s a very shallow moat, then. I mentioned DeepSeek because everyone insisted the American giants had huge leads over the rest of the world until the day they got a big reminder that three of the things you mentioned are commodities and UX isn’t a moat.
There used to be a ton of chip manufacturers. Now there's only one dominant player left: TSMC. Eventually, the winner outpaces everyone and the cost to compete is so astronomically high, only 1-2 will be left.
Not everyone can give away their models for free forever - especially when the cost to train a new model is exponentially more expensive.
The LLM labs can't all keep up. Most of them will fail. Some of them will merge. In the end, there will be a few dominant players.