tl;dr: 

• The Model provider needs to verticalize into the app layer aggressively
• The AI Industry Arch will probably stratify into [Infra-App], not [Infra-Platform-App].

OpenAI being as aggressive as they are in the app layer took people by surprise because there was a soft consensus in the industry to compare Model providers to the cloud platform layer (eg. this Sequoia graphic)

But scaling laws prevent this business from being similar to cloud platforms. 

As we enter a data availability bottleneck, the Model provider needs to consume entire app verticals because that seems to be the surest way to not just vendor lock demand, but more importantly, increase the supply of new and useful human data. The more, the merrier.

Plus, an app ecosystem powered by the same subscription plan reduces the swapability of the underlying model provider. 

If open source AI and compute price falls take off and models become commoditised in the future, the Provider cannot depend on pure App players like Cursor, who will either build their own models or improve their "auto" feature to always select the best globally available model for any task. The Model provider's app ecosystem will then act as a distribution moat.

There is a lot of incentive for the Model provider to consume the Entire app layer, if they can afford to do that. (By tying the fortunes of Nvidia, Oracle, AMD, Broadcom, Microsoft, Softbank, and the US Government to its own success, Sam seems to be making sure that Infra players will have a business incentive to favour OpenAI over other providers in the verticalisation wars.)

But is the verticalization really absolute? Even OpenAI is partnering with 3rd party cloud era apps like Figma, etc., by providing them the means to bring their utilities on OpenAI. Wouldn't there remain some domains like this where the provider will not eat the wrapper and maybe partner with it instead? 

Yes. 

But I think eventually these apps will also be abstracted away. Both Figma and Canva will hide behind a "design" tool on ChatGPT. Based on the task, ChatGPT will choose the best vendor. The User will not need to worry about the brand.

There seems to be no reason for "AI-First" apps to exist. Models will eat them all up eventually. FDE programs might ensure that this happens at the Enterprise level as well, although it will be much slower. 

If AI solutions are a good fit for a problem space, there is no reason for the Model provider to refrain from providing a first-party app to the user eventually. The cost is minuscule.

The AI Industry Arch will probably look like [Infra-App], not [Infra-Platform-App].

The Model layer is the App layer.

If you are in the software business, this is an existential predicament. Your code is worthless as an IP. You need to compete with models of your own. But you can't afford that even if you have billions in cash.

The boring implication is that we are heading towards an oligopoly. The exciting one is that the collective incentive to commoditize the Model layer is rising. 

A future where Everyone is a Model Provider might be closer than we think.