Rented Models Commoditize App Layer
Jasper: the $72M ARR Google Suite of generative AI
Cheap model access shifted early generative AI advantage away from owning the model and toward owning the workflow. Jasper and Copy.ai could launch fast because OpenAI exposed GPT-3 as an API, so they did not need to train or host frontier models, they just wrapped them in marketer friendly templates, editing flows, and integrations. That same ease of entry also meant dozens of near identical apps could appear quickly, because every new startup could rent the same intelligence by the word or image.
-
The app layer was commoditizing because the hard part, model training and serving, was abstracted away. Jasper described itself as layer two on top of foundation providers, with some actions hitting OpenAI models directly and others routing through fine tuned or smaller models built for specific templates.
-
This created a usage based cost structure instead of classic software economics. Copy.ai and Jasper paid model providers each time users generated output, and Copy.ai said production API calls were far more expensive than training, with OpenAI its second biggest cost after people. That makes the business feel closer to a reseller with software margins than pure SaaS.
-
The strategic risk was not just startup copycats, but incumbents with their own models and distribution. Jasper and Copy.ai both noted that Microsoft and Google could bundle AI into products users already live in, while independent apps had to justify a separate seat through better cross app workflows, brand voice control, or stronger task specific output.
The next phase favors apps that turn rented intelligence into proprietary workflow, data, and distribution. As model access gets cheaper and more interchangeable, value moves to products that plug into daily work, learn from customer behavior, and deliver consistent results across tools, while generic prompt wrappers get compressed by both horizontal incumbents and newer apps built on the same foundation APIs.