ON-PREM & EU
Data stays here
Not every company can send operations data to an arbitrary cloud. Then AI belongs where the data already is — on your network or in an EU data centre.
Who this is for
IT, leadership, and anyone who has to take data protection seriously. Typical: machine data, recipes, anything with people in it — things that should not 'just' go abroad.
What on-prem and EU mean
On-prem: the path runs on your machines, behind your door. EU: a data centre in the European Union, with a clear contract. Both are possible. A US cloud only if you choose it — not as the default.
PRACTICE · ANONYMISED
Searching and comparing documents — without them leaving the building
- Situation
Technical documents — test reports, supplier files, scanned drawings — were spread across folders. Anyone searching for something or comparing two versions needed someone who knew where to look.
- Built
Scans made readable via OCR, all documents indexed. A language model answers questions about them and sets out differences between versions — with a reference to the passage in the document.
- Result
An answer instead of a folder search. This approach runs on your own network or in an EU data centre — the documents don't have to go anywhere.
GDPR, without legal fog
Who processes which data, where it lives, for how long, who can see it. I write that so your privacy and IT people can work with it. Software does not issue a ruling — but without this clarity every AI project stays fragile.
How we start
First: may the data leave the building — yes, no, EU only. Then the smallest path that keeps that rule. No programme as the first step.
Questions plant managers ask
Is AI without a US cloud good enough?
For the typical tasks in a plant — searching documents, generating reports, analysing controller exports — yes. Open language models run on your own server or with an EU provider. For these tasks the gap to the large US models rarely decides anything; the data staying in-house often does.
What hardware do I need?
For document search and reports a single server with a GPU is often enough; for analyses without a language model, less. I size it for the concrete case — no bigger than necessary, and only after the pilot.
Where does the data really sit in an EU setup?
In a data centre in the EU, with a provider under a European contract and a data-processing agreement. Who processes what, where, for how long, who has access — that ends up in writing so your data-protection people can work with it.
Can I still move to the cloud later?
Yes. The path is built so that model and storage location are interchangeable. On-prem is a decision, not a one-way street.