Own ChatGPT for the company: Costs

A separate ChatGPT for the company – i.e. a self-hosted chat interface such as Open WebUI with a freely chosen language model – does not cost a license per user. The costs consist of server, model usage, setup, and operation. For models from German data centers, prices range from €0.10 to €3.60 per million tokens, depending on the model (IONOS AI Model Hub, net, as of October 2026).

Many companies are noticing that employees have long been using private AI accounts for work – with company data. Having your own AI chat is the controlled alternative. But what does it really cost? This article shows the cost drivers.

The four cost blocks

Cost block	What is involved	What it depends on
Surface	Open WebUI on a server	small server is usually sufficient
Language model	API per token or own GPU	Usage, model size
Setup	Installation, SSO, Knowledge Collections	Scope, Connections
The establishment	Updates, Backups, Help	Number of users, scope of services

Model costs: API or own hardware?

The language model is the variable part. Two ways:

Model via API from Germany: Billing per token (one token roughly corresponds to one syllable). For the IONOS AI Model Hub, for example, Mistral Small 24B costs €0.10 input and €0.30 output per million tokens, Llama 3.3 70B costs €0.65 per million tokens (net, as of October 2026).
Own hardware: a server with a powerful graphics card running an open model. High one-time costs, but no costs per request. What is necessary is explained in the article Hardware for your own AI.

Calculation example model costs

Example: 30 employees ask an average of 15 questions per working day. With context, each request consumes around 2,000 tokens input and 500 token output. With 20 working days, this is about 18 million input and 4.5 million output tokens per month. With Mistral Small 24B, this would be around €1.80 plus €1.35, i.e. a good €3 net per month; with Llama 3.3 70B, around €15. Knowledge collections increase the input tokens because found text passages are sent along.

The example shows that with models from the API, the running model costs for typical office applications are rarely the big item. It is more expensive for very large models, long documents or automated mass processing.

Setup and operation

The larger part is usually the one-time setup: server, registration via the company account, model selection, knowledge collections, rights and training. After that, there are updates, backups and contact persons for questions. We will give concrete prices after a short conversation, because the scope and connections vary greatly.

Comparison with licenses per user

Commercial AI chats for companies usually charge per user per month. With a self-hosted solution, on the other hand, you pay fixed operating costs plus consumption. The more employees use the chat, the stronger the argument for your own solution. What's more, data and knowledge remain on your infrastructure, and you can change the model at any time.

How to get off to an economical start

start with a medium-sized model via API from Germany
Observe usage for two months
only then decide on large models or your own hardware

We set up Open WebUI, choose the right model with you and keep an eye on the costs. More about open models on the Open Source Language Models page.

Counselling and accompanied implementation is eligible for funding: Through INQA coaching, the federal government covers 80% of the consulting costs of up to €1,200 per day for up to 12 days, i.e. up to €11,520 subsidy. In Mecklenburg-Western Pomerania and the other new federal states, BAFA consulting funding is also eligible at 80% until 31.12.2026 (50% in the old federal states). Funding is provided for consulting, not software. The funding check shows which program fits.

Frequently Asked Questions

How much does your own ChatGPT cost per month?

This depends mainly on usage and model. The model costs via API from Germany are often in the single- to low double-digit euro range per month for typical office use; in addition, there are servers and support.

Is having your own ChatGPT cheaper than ChatGPT Team or Enterprise?

Often from a medium number of users, because there is no license per capita. If there are very few users, a ready-made service can be cheaper.

Do I need my own graphics card?

No. Models can be used via API from German data centers. Own hardware is only worthwhile if there is high, even use or strict data protection requirements.

What is a token?

A token is a text module, roughly a syllable or a short word. AI services charge according to the number of tokens processed, separated by input and output.

Would you like to estimate the costs for your case? Get in touch via the contact page.