DeployCo

In the AI industry there are two very different types of companies. One type builds the models themselves: they gather enormous amounts of data and let data centers run for weeks on end until a system emerges that can generate text or images. The other type develops no system of its own at all. They take a finished model from outside and build something out of it that a particular customer can actually use. For this second group, the coined term DeployCo has become established in business writing. It combines the English “to deploy” with the abbreviation “Co” for Company, i.e. firm.

Why the distinction between model builders and users matters

Training a model costs an extreme amount of money. For the largest systems, figures in the hundreds of millions are mentioned, sometimes more. Only a handful of corporations can keep up with this race. A DeployCo deliberately stays out of this race and thereby saves itself the most expensive part.

For investors the distinction matters because both types produce completely different numbers. A model maker first burns through enormous sums and hopes for later returns. A DeployCo can earn money early on, because it bills for projects. In return, its growth is often slower, since every new customer means new work.

There is also a risk that keeps coming up in discussions: dependency. Anyone building on someone else’s model is dependent on that provider’s prices and rules. If the provider raises fees or changes the terms, this hits the DeployCo immediately. Some companies therefore hedge against this by being able to use several models in parallel.

What a DeployCo actually builds

A raw language model is unusable for most businesses. It knows neither the internal processes nor the technical jargon of a company. A DeployCo closes exactly this gap. It connects the model to the customer’s databases, builds a user interface, and determines which responses are permitted.

One example: an insurance company wants claims reports to be automatically pre-sorted. The DeployCo first checks which tasks are even worth pursuing. Then it tests how reliably the model performs on real cases. Only after that is the system connected to the existing software and fine-tuned over the course of months.

On top of that comes a part that is not very glamorous but eats up a lot of time: data protection, logging, and training staff. This effort is precisely the real reason customers pay for something like this. You can picture a DeployCo as an interior fitter. Someone else supplies the shell of the building; it is only through them that the house becomes livable.

DeployCos in stock market reports and everyday work

The term appears mainly in analyses and stock market reports. When observers ask who is actually profiting from the AI boom, they like to split the market into three layers: the chipmakers, the model builders, and the DeployCos. Traditional consulting firms, too, are now frequently placed in this third group.

In everyday life, one encounters their work without knowing the name. The chatbot of a health insurer or the search field in a company intranet often comes from such service providers. A common misconception is that a separate model lies behind every AI product. Usually it is the same well-known model, merely packaged and connected differently.