Infrastructure data doesn't belong
on someone else's AI cloud.
Private LLM hosting on a dedicated machine in EU-Central, for water utilities, municipal operators, and other critical infrastructure teams that need AI tools without sending operational data to a third-party model provider.
Not a security or compliance certification. This page describes GPUwerk's infrastructure and contractual commitments. Whether this setup satisfies your regulator, insurer, or internal security policy for critical infrastructure work is a decision for your own security and compliance team.
Why water and utility operators ask for this
Asset maps, SCADA point lists, maintenance histories, and incident reports describe how a public water system actually works. That's not information most operators want sitting in a shared AI provider's logs, even indirectly.
"A public model provider now has our facility layout"
Pasting a network diagram or a pump station schematic into a consumer chatbot puts that document on infrastructure you don't control, under terms you likely haven't read closely. A dedicated instance keeps the same kind of AI assistance on a machine assigned only to you.
"We can't tell where our data actually goes"
The hardware is physically in EU-Central, operated by PRINT IT! SE, a Societas Europaea registered in Prague. There is no US parent and no other region in the chain. Under the data processing agreement, there are no sub-processors for instance content, and GPUwerk does not access your workspace's content except at your request for support or where a legal obligation requires it.
"Engineers use public AI tools informally anyway"
Staff drafting incident reports or summarizing inspection logs in a public tool is a real exposure, not a hypothetical one. A sanctioned instance with the same chat interface keeps that workflow off third-party infrastructure. See the private ChatGPT setup →
What this is, and what it isn't
Keep the boundary clear between the IT-side AI workload described here and any OT or control-system network.
| Question | Answer |
|---|---|
| Is this an OT or SCADA product? | No. GPUwerk hosts general-purpose LLM infrastructure. It is not designed, tested, or certified for direct integration with control systems, and we don't position it that way. |
| What is it used for in practice? | IT-side tasks: drafting incident and compliance reports, summarizing maintenance logs, answering questions against internal documentation, and similar text work that currently might go through a public AI tool. |
| Where is data physically processed? | EU-Central, on a dedicated single-tenant machine assigned to you |
| Who can access instance content? | Through the instance itself, only holders of your SSH keys; password login is disabled fleet-wide. GPUwerk keeps infrastructure administrator access, as on any hosted service, and does not use it on your content except at your request for support or where a legal obligation requires it. |
| Sub-processors for the workload? | None, listed at /legal/sub-processors |
| DPA (GDPR Art. 28)? | Published at /legal/dpa, no charge |
| Certifications held? | None claimed on this page or elsewhere on the site, including for critical infrastructure. Verify directly with us before relying on this page if your procurement requires one. |
Pricing
On-demand billing starts at $0.79/hour for a spark-1x instance and $1.79/hour for spark-2x. A held (reserved) allocation runs at 75% of the on-demand rate: $0.59/hour and $1.34/hour respectively. Full details on the pricing page.
Questions we get from utility teams
Is this suitable for OT or SCADA-adjacent systems?
GPUwerk hosts the LLM workload on a dedicated machine; it is not a SCADA or OT product and holds no operational-technology certification. Most utilities that use it keep the instance on the IT side, for tasks like drafting incident reports, summarizing maintenance logs, or answering questions against internal documentation, with a network boundary between that instance and any control system decided by your own engineering and security team.
Does GPUwerk hold any critical-infrastructure security certification?
No. GPUwerk does not claim ISO 27001, IEC 62443, or any national critical-infrastructure certification. If your procurement or regulator requires one of these, verify directly with us before relying on this page, and treat the absence as a gap to plan around rather than something implied by the rest of this page.
Where is the data physically processed?
EU-Central, on a dedicated single-tenant NVIDIA DGX Spark machine assigned to your account, operated by PRINT IT! SE, a Societas Europaea registered in Prague, Czech Republic.
Can we run a pilot before committing to a contract?
Yes. On-demand billing starts at $0.79/hour for a spark-1x instance with no minimum term, so a pilot on a non-critical workload, such as document search over maintenance manuals, is a reasonable way to evaluate the setup before moving anything closer to operations.
Related pages: private AI hosting for energy, manufacturing, government, and the private LLM hosting overview.
Related pages
Bring your operational AI workload to a machine you control.
Talk to the people who run the racks, or start with a pilot and a practical rollout plan.
Talk to us Deploy an instance