This article was generated by AI. Labelled in accordance with Article 50 of the EU AI Act. Responsible for publication: Sophera Consulting.
German AI providers are now the first choice for 96 percent of German companies, according to a Bitkom survey published on 2 November 2026. Yet 64 percent will not pay more or accept fewer features for it. For a business, the useful takeaway is narrower: what counts is where your data is processed and whether you can swap the model later.
What did the Bitkom survey find about German AI providers?
Bitkom, the German digital industry association, released the figures as the federal government's Digital Summit opened. Bitkom Research interviewed 603 companies with 20 or more employees by phone in summer 2026, and the association calls the sample representative of the economy as a whole.
- 96 percent would prefer AI applications from German providers.
- 87 percent say the provider's country of origin matters when they choose an AI tool.
- France gained ground, from 38 to 46 percent. Providers from other EU countries reached 43 percent, up from 40.
- The US lost support. 41 percent would prefer American AI, down from 51 percent a year earlier. China fell from 18 to 12 percent.
- 64 percent are not willing to accept a higher price or a smaller feature set in exchange.
Bitkom president Ralf Wintergerst put it plainly: sovereign AI offerings have to hold their own in competition, wherever the provider is based.
Why do companies want German AI but refuse to pay extra?
It looks like a contradiction until you read the methodology. The question behind the 64 percent asked whether companies would prefer an AI service that processes their data exclusively in Germany and protects it from foreign access, even if that came with drawbacks. So the survey was really asking about control over data. The passport of the software vendor was not part of that question.
That matches what a managing director usually wants to know. Who can get at my customer records, my price lists, my patient files? Whether the model was trained in Paris, Heidelberg or California answers only part of that.
Origin, processing location and model are three separate decisions
People say "AI from Germany" and mean one of three things. You can decide each of them on its own.
Where the provider is based
This is the company's legal seat, and with it the law the provider has to follow. A company headquartered in the US stays bound by US law even if its data centre sits in Frankfurt.
Where the data is processed
This is the physical place your data passes through: a data centre in the EU, a rented server, or hardware in your own building. If the model runs on your own hardware, the data never leaves the premises, no matter who built the model.
Which model reads the text
A language model is the component that reads and writes text. It often comes from a different company than the application you use. German software may call an American model in the background, and an openly available model from a US lab can run entirely on a server in your basement.
What does this mean for an AI agent in your office?
Here is a constructed example, not a client case. Say a wholesaler lets an agent handle incoming order emails. An AI agent is a program that reads emails, looks things up in your systems and prepares a draft that an employee approves.
1. The email arrives in your mailbox, as it does today. 2. The agent reads it. This is the moment the email text goes to a language model, and the only point where the processing location matters. 3. It matches items and customer inside your ERP system, using your master data and your mapping tables. 4. It creates a draft order. Your sales office sees each field marked with where it came from and approves it.
For routine orders, a model on European infrastructure with a data processing agreement is enough. A hospital whose agent turns dictated nursing notes into reports is better off running the model on its own hardware. Both are the same agent with a different processing location for step 2. Our piece on self-hosted versus cloud AI goes through that trade-off.
The part of the agent worth protecting is not the model. It is the rules, mappings and test cases from your own operation. If those belong to you, the model stays replaceable.
Should we wait for a German model?
No. Every day of waiting costs the hours your staff already spend retyping orders and chasing missing details. The better move is an agent where the model is a swappable part. When a European model arrives that reads your documents just as reliably, you switch. A test against your past cases tells you whether it does. The flag on the vendor's website does not. How to stay independent of any single platform is covered in our article on vendor lock-in.
Sophera Consulting plans for that switch from the start. Every automation we build comes with a maintenance agent that tests changes against past cases and handles the move to new AI models. We work at a fixed price with no subscription, and you own the result. Models run on European infrastructure or, if you prefer, locally on your own hardware. Our free automation check is where we work out which processing location fits your data.
Our recommendation
Read the survey as confirmation that data control is a legitimate buying criterion. Then ask every provider three separate questions: where is the company based, where is my data processed, and can I change the model? A vendor who answers all three clearly does not need to charge you a premium for a flag. For the wider checklist, see our overview of GDPR-compliant AI.
This article was created with the help of AI.
