.....Why should there be ten major AI companies? This business has all the makings of a monopoly. Ultimately, the best algorithm wins out—just as Google prevailed in search. The prize for the winner may be gigantic—fair enough. But the others will see their value approach zero. Consequently, the overall cost is too high, because every company is valued as if it were the monopolist that survives......

The above is a discussion of the LLM model alone. That is the single layer that is absolutely a commodity and has almost no value, other than a necessary running component.

The value is in the other two layers - the application (use habbit, distribution, network effects, etc) such as WhatsApp, Google Search box, Instagram, YouTube etc. And secondly the same infrastructure/compute layer which require massive massive scale to be competitive.

You can have hundreds of monopolies running together in the same way that you can have hundreds of bridges, or lots of railroad companies, or tens of thousands of prime real estate office blocks, that each have a local monopoly. In the case of compute infrastructure, the demand next 20 years is so stupendously larger than supply today, with many firms producing at sufficient scale to offer competitive pricing, that they become similar to massive bridges with ongoing revenue streams.

- Manlobbi