Updated 7 min readBy The VibeDeploy team

Data Residency for AI Websites in the EU: What Sits Where

Residency is a physical question: which bytes live on which machine. A follow-the-data map of an AI-built site and how to verify where everything sits.

Short answer

Data residency is the physical question of which machine holds which bytes. For an AI-built site the answer has up to eight parts: served files, account and deploy data, builds, snapshots and backups, logs, form submissions, analytics and any backend database. Follow one page view, list every external hostname in the live HTML, and ask your host in writing where each part runs. Forms and analytics are where AI defaults most often send data abroad.

Most of the map in one country

Files, deploy history, forms and analytics on servers in Belgium, with the exceptions named on a public sub-processor list. 14 days free, no card.

"Where does the data live?" sounds like one question and is really two. One is legal: which country's laws govern the data. The other is physical: which machine, in which building, holds the actual bytes. The first is a matter of contracts and corporate jurisdiction. The second is a matter of geography you can literally point at on a map. This post is about the second one, residency in the concrete sense, applied to a site an AI built for you. Not the abstract layers of governance, which are covered separately in data sovereignty for AI-built websites, but the plain inventory: for each moving part of the site, where does the byte physically sit.

Follow one page load

The clearest way to find out what sits where is to trace a single visit. A person in Antwerp opens your AI-built site, and here is the chain of bytes that fires.

First, their browser asks DNS where your domain points. That query hits a nameserver, and the nameserver's location is the first residency fact, usually overlooked because DNS feels invisible. Then the browser connects to the host that serves your site and pulls the HTML, the stylesheet, the images, the scripts. Those files sit on a specific server in a specific data centre: residency fact two, and the one most hosts advertise. If the page loads a font, a map or a video from somewhere else, the browser fetches those too, from wherever those providers keep them: residency fact three, and the one AI tools quietly introduce. Finally, if the visitor submits the contact form, their message travels to whatever endpoint the form points at, and lands in whatever storage receives it: residency facts four and five.

A single page view has already touched five physically distinct locations. "The site is hosted in the EU" describes at most one of them.

The inventory, part by part

Broken out, an AI-generated site has these physical residency points, and each deserves a straight answer rather than a shrug:

  • The served files. The HTML and assets a visitor downloads. This is the layer everyone checks. It is also the easiest to get right and the least likely to be the problem.
  • The account and deploy data. Your login, your site's configuration, the record of every deploy. This lives in a database, and that database sits on a machine somewhere. On several large platforms the files sit on EU edge nodes while this database runs on another continent.
  • The build pipeline. If your site is built from source rather than served as-is, the build runs on a machine that briefly holds all your code. Where that build machine lives is a residency point people forget entirely because it is transient.
  • The snapshots and backups. Every previous version of the site, captured and stored. Backups are data too, and a backup restored from another jurisdiction is data that left and came back.
  • The logs. Access logs and error logs contain visitor IP addresses, which are personal data. Where logs are stored and for how long is a real residency question, not an operational footnote.
  • Form submissions. The actual personal data visitors type. AI tools routinely wire generated forms to a handler the model knew about, which is frequently not in the EU.
  • Analytics. Same story: a tracking snippet included by default puts every visitor IP into a provider's storage, wherever that provider keeps it.
  • The backend and its database. More and more AI tools add a login, an API or a database to the site. That server, and the database behind it, hold the most personal data of all, and they often run with whichever backend provider the tool prefers.

A residency review that stops at the first bullet has checked the one point least likely to leak and skipped the seven where AI defaults most often send data elsewhere.

The one flow that legitimately leaves

Here is the honest exception, and it matters because pretending otherwise would be dishonest. For an AI-generated site, the generation step itself usually runs on a model provider's infrastructure, and the largest ones are outside the EU. When you describe your site to an assistant, that prompt goes to wherever the model runs.

This is fine, and it is worth being clear about why. What travels during generation is your instructions and the resulting code, not your visitors' personal data. No customer email, no visitor IP, nothing a person entrusted to your site is involved at build time, because there are no visitors yet. Residency is about where your operational data lives once the site is running and real people are using it. The build-time trip to a model provider is a separate event that touches none of that. So the sensible posture is precise rather than absolute: the generation can happen anywhere, and everything the running site does with real people's data stays in the EU. Conflating the two leads people to either fret about the wrong thing or wave the whole question away.

How to verify without taking anyone's word

Residency claims are checkable, and vague answers should be treated as failed checks.

  1. Open your live site's source and list every external hostname. Each script, font, image or form action pointing at a domain that is not yours is a byte flow to account for. This is a ten-minute job and it finds the analytics and form leaks immediately.
  2. Ask your host, in writing, where each part runs, not just the files. Files, database, build machine, backups and logs, itemised. A host that can answer per component is a host that thought about it. A host that answers only "our servers are in Europe" has answered one bullet of eight.
  3. Check for a Data Processing Agreement and a subprocessor list. Any layer touching personal data should be covered by a DPA, and the subprocessors should be named. If you cannot get the list, you cannot know where the data goes.

Whatever the audit finds, fix the forms and analytics first. They are the two points where real visitor data flows today, and they are the two AI tooling most often sends abroad by default.

What actually sits where on VibeDeploy

The operating company is Tnet Comm.V, a Belgian company, and most rows of the inventory point at one place: its own Kubernetes cluster in Belgium. The served files, the deploy history, the account database and GitHub builds all run there, and the restore point taken before each deploy stays on that cluster rather than going off-site. If the site has a backend, its containers (your own server, Postgres, MySQL, Redis or workers) run on the same cluster; container data is not yet included in snapshots. Contact forms go through a built-in relay that emails submissions from a mail server in Belgium, so no external form handler enters the map, and the analytics are built into the platform, so there is no third-party analytics store to place.

The exceptions, because the point of an inventory is that it survives scrutiny. The generation step is your choice of AI tool and never sees a visitor's data. Backblaze, a US company with an EU region, is earmarked for an off-site copy of the platform database but receives nothing yet. That database holds accounts, billing, settings and the built-in visitor statistics (IP address, browser and referrer per page view) but no site files, and its nightly backups are written to a separate volume inside the Belgian cluster. Domain verification looks up public DNS records through Cloudflare and Google Public DNS, and the optional AI site-builder features run through OpenRouter and Google Stitch, both in the US. Customer site content does not pass a US processor unless you use those builder features. All eight sub-processors are named on the subprocessor list rather than buried, and the DPA applies on every paid plan, so the verification steps above come down to reading two pages rather than chasing a sales rep.

Pricing is flat rather than metered, starting at 4 euro per month with a 14-day trial and no card up front; prices include VAT where it applies, on the pricing page. If your project is still on localhost, the path onto this infrastructure is in deploy localhost to production, and once it is live, the reversibility layer that keeps every version inside the same jurisdiction is in staging, snapshots and rollbacks for AI-built sites. Choosing between providers on these rows is covered in choosing an EU hosting provider for AI websites, with a sourced comparison in European alternatives to Vercel and Netlify. Residency for an AI-built site is not a mystery. It is an inventory with a fixed number of rows, and every row has a place you can name.

From the VibeDeploy team

Ship your AI-built site in minutes

VibeDeploy hosts your AI-built websites in the EU with custom domains, automatic SSL, and a 14-day free trial that gets you online today.

Related reading