GDPR and the EU AI Act on dedicated hardware
Moving an LLM from a shared multi-tenant API to a dedicated DGX Spark changes one GDPR answer: who else can reach your data while the model runs. It does not change your EU AI Act risk tier or your role as provider or deployer. This page sets out what GPUwerk can document for your data protection officer and what stays your own analysis. It is not legal advice.
Two separate frameworks, two separate questions
GDPR and the EU AI Act get lumped together in conversation because both matter to a company deploying AI in Europe, but they answer different questions. GDPR is about personal data: what you may process, on what lawful basis, and who else touches it along the way. The AI Act is about the AI system itself: which risk tier your use case falls into, and whether you're acting as a provider (roughly, the entity that develops or places the system on the market) or a deployer (an entity using it under its own authority). Neither framework is decided by where your compute physically sits. Hosting location is a GDPR and data-residency question; risk tier and role are an AI Act question. Keep them separate. Vendor marketing that blurs the two, including any hosting provider's, is where most of the confusion comes from.
What choosing dedicated hardware actually changes
Running your model on a dedicated DGX Spark instead of calling a shared multi-tenant API does not change your AI Act risk tier or your provider/deployer status; those turn on what the system does and who's responsible for putting it into service, not on the hardware underneath it. What it does change is the GDPR-relevant question of who else can access your data while the model runs. On a shared API, that answer is a description of the vendor's multi-tenant architecture and contractual isolation controls. On a whole dedicated machine, where you have root inside your own container, the answer for other customers is nobody: no other customer's workload runs on that machine while you rent it. GPUwerk keeps infrastructure administrator access, as any host does. That is a simpler fact to hand a data protection officer or an auditor. It is not a bigger legal shield by itself.
The GDPR evidence GPUwerk provides
- Location. GPUwerk's own tiers run on hardware in Hradec Králové, Czech Republic, operated by PRINT IT! SE. The Partner tier, which you can choose as a location in the console whenever a partner node is free, runs on an independent partner's machine at the partner's premises. That machine is in the EEA unless the renter chooses any location at deploy, which can place it outside the EEA as a third-country transfer, under the European Commission's standard contractual clauses. Renters may use any data on the tier and choose its location.
- A published Article 28 DPA. A standard GDPR data processing agreement is available at no charge at gpuwerk.com/legal/dpa.
- A short sub-processor list. Published at gpuwerk.com/legal/sub-processors; there are none for instance workloads on GPUwerk's own tiers. On the Partner tier, the partner hosting the machine is the sub-processor for that instance.
- No training on your content. GPUwerk does not access, read, copy, index or analyse the content of your instance. Infrastructure administrator access exists, as on any hosted service, and is not used on your content except at your request for support or where a legal obligation requires it.
- A stated deletion path. Terminating an instance deletes its container and workspace from the node; nodes are sanitised before reallocation to another customer. GPUwerk keeps no copy of a workspace anywhere else, so nothing is kept after a terminate.
You remain the controller of your own processing, so the compliance of what you do with the model is yours to assert; this is the infrastructure evidence that supports that assertion. For the full detail behind each of these points, see private LLM hosting and the GDPR checklist for AI tools.
The AI Act question, briefly
Most companies that self-host an existing open-weight model for internal use, without substantially modifying it or offering it to others as a product, are deployers rather than providers. Deployer obligations are narrower. The classification depends on what you do with the model and whom it affects, not on whether it runs on a dedicated Spark or a shared API.
Dedicated hardware helps after that classification. Deployer obligations include human oversight and, for higher-risk uses, monitoring and record-keeping. A clear answer to "who has access to the data going into the system" is a good starting point for that oversight. It does not replace the classification itself. The full reasoning, including when you can end up with provider-like obligations, is in the EU AI Act and self-hosted LLMs explainer.
What this page does not claim
Dedicated hardware does not make you AI Act compliant. A DPA alone does not satisfy GDPR for your specific processing. Neither replaces review by counsel who knows your model, your data and your use case. GDPR and the EU AI Act are the two frameworks buyers ask about, so this one page covers both; a hosting provider cannot responsibly offer more legal depth than that.
What a dedicated DGX Spark in EU-Central gives you is a short, concrete answer on data access, while counsel handles the classification and the processing analysis. See private LLM hosting for how it is set up, or send your questions directly.
Frequently asked questions
Does running our LLM on dedicated hardware instead of a shared API change our EU AI Act obligations?
No. The AI Act assigns obligations by risk category and by role, provider or deployer, not by where or how the compute is hosted. Most companies self-hosting an existing open-weight model for internal use are deployers rather than providers, whether that model runs on a shared API or a dedicated machine. What dedicated hardware changes is a separate question: who else can access your data while the model runs. The longer explainer on the EU AI Act and self-hosted LLMs covers the full provider/deployer distinction. Not legal advice.
Is GPUwerk GDPR compliant?
GPUwerk provides the evidence a data protection officer asks for: hardware and operations in EU-Central (Hradec Králové, Czech Republic), an Article 28 data processing agreement published at gpuwerk.com/legal/dpa at no charge, a published sub-processor list (for instance traffic on GPUwerk's own tiers: Hetzner as the network gateway in Helsinki, and Cloudflare only for a model endpoint terminated on that gateway), and single-tenant machines whose content GPUwerk does not access, read, copy, index or analyse. The Partner tier is different: an independent partner hosts the machine, possibly outside the EU. You may use any data on it and you choose the location: if your data must stay in the EEA, choose a location there. You remain the controller, so the compliance of your own processing is yours to assert; GPUwerk supplies the infrastructure evidence behind it.
Why does a dedicated machine make the GDPR answer simpler than a shared API?
On a shared multi-tenant API, the honest answer to "who else's traffic touches this model or this hardware" is usually a list of the provider's other customers and infrastructure, even when contractually isolated at the data layer. On a whole dedicated DGX Spark, the honest answer for other customers is nobody: it is your container and your session, and no other customer's workload runs on that machine while you rent it. GPUwerk keeps infrastructure administrator access, as any host does, and under the DPA does not use it on your content except at your request or where the law requires it. That is a shorter, more concrete answer to give counsel or an auditor, not a stronger legal guarantee by itself.
Is this page legal advice?
No. It describes GPUwerk's infrastructure and the general shape of GDPR and the EU AI Act at a level I'm confident is accurate, but it is not legal advice and is not a substitute for review by qualified counsel familiar with your specific processing, model, and deployment.