There is clearly a moat. The moat is in data, training algorithms, compute capacity, and UX.

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.

> 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.