OpenAI's bet is that its frontier models will be so far ahead of the current status quo that, if they are the only ones providing those frontier models, they will be able to name their price (to end users and advertisers alike) while increasing their share of spend in the space.

But even today, last-generation and open-source models form a meaningful portion of adopted solutions. Not every application in 2028 will need the AGI-approaching GPT-10 - especially if those applications can leverage a relatively small amount of code, perhaps even written by that GPT-10, that can in turn orchestrate (say) DeepSeek V5 running on compute that can be obtained for pennies on the dollar.

OpenAI could become a victim of its own success, and cause a house of cards to take down the global economy in the process. I personally hope this doesn't happen, but there is real risk here.

There will be the WebVans of the AI boom era, we just don't know their names yet. There also will be Ciscos and Suns that will never reach their high-water mark ever again, or become obsolete in a few years, and sold for less that what people expect.