Application Layer

Large computer systems are built in layers, much like a house is built in stories. At the very bottom lie cables, machines, and power. Above that come programs that shuffle data back and forth. And right at the top sits the application layer: everything a human directly interacts with. A weather app, an online shop, a chat program – these are applications in this sense. The term therefore doesn’t say what a program can do, but rather where it stands in this stacking order.

Why the money is made at the top

The layer concept is more than just an organizing aid. It enables division of labor. Whoever builds an app doesn’t need to know how data physically travels through a fiber-optic cable. They rely on the layers below to do their job. This allows many companies to work simultaneously on different levels without having to coordinate with each other.

In the AI economy, this has become a very concrete question. At the very bottom stand chip manufacturers like Nvidia. Above that sit the data centers, then the large language models from OpenAI, Google, or Anthropic. A language model, in this context, is a program that has learned from vast amounts of text to formulate meaningful answers. And at the very top, in the application layer, sit the companies that turn all of this into concrete products.

Investors argue over which layer will yield the greatest profits in the long run. One argument holds that the models below become interchangeable and thus cheap, while direct customer contact at the top remains valuable. The counterargument holds that anyone who merely resells someone else’s models has little of their own and can easily be displaced.

What an application provider builds itself

A typical AI application doesn’t have its own model. It sends the user’s request via an interface to a third-party model and receives a response back. An interface, in this context, is nothing more than a fixed agreement on the format in which requests and responses are exchanged. For every request, the provider pays the model operator a small fee.

The actual work lies in everything surrounding this. The provider formulates the instruction to the model, checks the response, attaches company-specific documents, and stores results. It takes care of registration, billing, data protection, and making sure the interface is understandable. It’s precisely these inconspicuous parts that often determine whether a product is usable.

A common misconception is that the application layer is merely a thin shell. For some products this is true, and people disparagingly speak of a wrapper, meaning pure packaging. For others, years of detailed work go into it, for instance in connecting to hospital software or accounting systems. The difference lies in how hard the product is to copy.

These are programs you use every day

Almost every program on your phone belongs to the application layer. The browser is one example, the music app another. Even a chatbot in a web browser counts, even if a massive model is computing behind it. What you see and click on is always the topmost layer.

In business news, the term usually appears in connection with startups. When it’s said that a company operates “in the application layer,” it means: it doesn’t train its own models, but builds software for a particular industry. Examples include programs that review contracts, summarize medical reports, or generate advertising copy.

Beware of a mix-up: in network engineering, application layer also refers to the topmost of seven precisely defined levels of a standard model for data transmission. There, it’s about technical rules, not business models. However, the context almost always makes clear which meaning is intended.