What does on-prem AI mean?

It means the AI model runs on a server in the company’s own building, not at a cloud provider.

What is the difference between a dedicated server and a shared cloud?

A dedicated server belongs to one company alone; the model and data are not shared with other companies. In a shared cloud service resources are shared.

Which is more secure?

There is no single right answer; security depends on access control, backup and contract. On-prem suits those who want physical control, a dedicated server those who want no operating burden.

Can the choice be changed later?

Yes if the architecture is designed for it; this is a topic to agree at the contract stage.

On-prem GPU or a dedicated server for the company?

Short answer An on-prem GPU means the AI model runs on a server in the company’s own room; a dedicated server means it runs on a server hosted by a provider but reserved for that one company. The goal is the same in both: the company’s data stays with the company.

Last updated: 2026-11-10 · Published by: Xontainer

Why does this question come up?

Shipping documents contain trade secrets and often carry personal data. Sending these documents to a shared, general AI service is unacceptable for some companies. The answer is to run the model on a server that belongs to the company or is reserved for it. There are two ways to do that.

Option A: on-prem

The server sits in the company’s own room. Data does not leave the company’s own network.

Pro: data stays physically with the company; network and access are under its control.
Con: hardware purchase, power, cooling, backup and fault response fall to the company or to whoever sets it up.
Fits: companies with strict internal requirements and their own IT team.

Option B: hosted, dedicated to the company

The provider hosts the server, but it is not shared with another company. The model and the data run only in the environment reserved for that company.

Pro: no hardware or operations burden; faster to set up.
Con: the data is not physically in the company’s room; trust in the provider and the contract is needed.
Fits: companies that do not want to run their own server room.

What criteria to look at

Data classification: which documents fall in which category? How much is personal data and how much trade secret?
Legal requirement: what do contracts and GDPR or KVKK obligations say about where data sits?
Operating capacity: who will monitor, back up and update the server?
Load: how many documents a day? Real-time voice and translation, or batch documents?
Flexibility: is it possible to move later from B to A or the other way?

Security principles common to both

Whichever option is chosen: companies are isolated from each other and one cannot see another’s records; every AI tool call is checked for company and role; AI does not write to pricing or customer master data. These are architectural principles that do not depend on where the server sits. For detail see the data security guide.

Common mistakes

Ignoring the operating load of the on-prem option: backup, updates and fault response are continuous work.
Not asking, in the hosted option, whether it is reserved for the company alone; a shared environment is something else.
Deciding before classifying which data will be processed.
Treating the choice as irreversible; it can change if the architecture is designed for it.

Sources

Frequently asked questions

What does on-prem AI mean?: It means the AI model runs on a server in the company’s own building, not at a cloud provider.
What is the difference between a dedicated server and a shared cloud?: A dedicated server belongs to one company alone; the model and data are not shared with other companies. In a shared cloud service resources are shared.
Which is more secure?: There is no single right answer; security depends on access control, backup and contract. On-prem suits those who want physical control, a dedicated server those who want no operating burden.
Can the choice be changed later?: Yes if the architecture is designed for it; this is a topic to agree at the contract stage.

Shall we talk about your own operation?

Leave a short note in the early-access form with your role and operation. The form is a prototype for now and sends nothing.

Early-access form