One of the first questions leadership teams ask is whether their data will leave the building. The honest answer is that it depends on which kind of model you use, and each kind has trade-offs.

Hosted models through an API

The most capable models are usually reached over the internet from a large provider. Business agreements typically commit the provider not to train on your data, and many offer regional hosting. This is the fastest and often the cheapest way to start.

Private deployments in your cloud

The same or similar models can run inside your own cloud account, so data stays within an environment you already govern. Costs are more predictable at volume, and it is often the easiest route through a security review.

Open models you run yourself

Open models can run entirely on your infrastructure, even without an internet connection. You get full control and no per-use fees, but you take on the work of hosting, updating and monitoring them, and the most capable open models still need serious hardware.

How to choose

Start with the sensitivity of the data and the regulation that applies to it. Then look at volume: low volume favours hosted models, high and steady volume can favour private or self-hosted. Finally, be honest about who will run it.

Most organisations end up with a mix: hosted models for general work, private deployments for sensitive data. The important thing is to decide deliberately, write it down and review it as the options change.