Why Dutch AI users are switching to European alternatives

For years, choosing an AI service was only about quality. Which model writes best, which model reasons most sharply, which model can handle an eighty-page PDF. Those are still good questions, but a second question has been added that you rarely heard a few years ago: where does this actually go when I hit send.

There are no hard Dutch figures on how many people are actually switching, and we're not going to make them up here. What does exist are three arguments that come back in every conversation about this topic. This article covers those three, and is then just as honest about what you currently give up if you choose the European route.

What exactly is shifting

This isn't a mass exodus. It's a shift in how people justify their choice. Where the question used to be "which AI is the best", the question is now more often "which AI is good enough and acceptable for this task".

That distinction makes most people not ideological but practical. Someone writing a poem doesn't care about processing regions. That same person, wanting to have a customer file summarized two hours later, suddenly does. The result is that people rarely switch completely; they split their usage. The sensitive things go to a European service, the rest stays where it was.

That split is also why this topic is more practical than it sounds. You don't have to choose between principles and the best tool. You just need to know which task belongs where.

Reason 1: you want to know where your text ends up

This is the weightiest reason, and the only one that's truly decisive for business users. As soon as your prompts contain customer names, medical information, personnel files or contracts, the processing location is no longer a detail.

What you can say about this with certainty is limited but useful. Mistral processes in the EU, in France and Sweden. Microsoft has an EU Data Boundary for business customers. And DeepSeek runs in China without an adequacy decision, which makes it unsuitable for personal or customer data, however good the models are.

For everything else: check it with the provider itself and don't rely on a summary, not even ours. The practical comparison per service, including what happens to your conversations by default, is in AI and privacy in the Netherlands. This article is about why people are moving; that article is about what you need to arrange.

Reason 2: the bill in euros

The second reason is less principled and therefore perhaps even more influential: money you can't predict is annoying money.

American providers invoice in dollars and exclude VAT. OpenAI, Anthropic, Perplexity and xAI all do this. At an exchange rate of $1 = €0.8687, a $20 subscription becomes €17.37 excluding VAT and €21.02 including VAT. If the dollar moves, your monthly cost moves with it, and across three subscriptions you notice that.

Google and Microsoft do price their Dutch consumer subscriptions in euros including VAT. Google AI Pro costs €21.99 and stays €21.99. Mistral invoices in dollars, but Le Chat Pro sits at $14.99, which converts to €15.76, and is still the cheapest serious Pro subscription.

If you want to see the full price breakdown, including annual totals and the effect of the VAT reverse charge, you'll find it in the complete guide to AI subscriptions.

Reason 3: language and context

The third reason is often mentioned first and turns out to be the least solid in practice. The idea is that a European provider writes better Dutch because it's closer by. That isn't true.

In our test, Le Chat scores 4 out of 5, exactly the same as ChatGPT and Gemini, and below Claude, the only one to reach a 5. Le Chat writes neutrally and tidily and stumbles over colloquial language. So European origin doesn't make the Dutch output better, but it doesn't make it worse either, and that's news for both camps. The full verdict per provider is in which AI speaks the best Dutch.

What does play a role is everything around the model: an interface in your own language, support that answers in your time zone, an invoice your accountant recognizes. Those aren't model capabilities, but they do determine how a service feels in daily use.

Where Europe still lags behind

An honest article on this topic also names what you give up, because it isn't nothing.

The context window is the clearest difference. Mistral Medium 3.5 has 256K context, while GPT-5.6 and Gemini 3.5 Flash are at 1.05M and Claude at 1M. In a chat you hardly notice; when plowing through a stack of reports in one session, you do.

Images and video are the second difference. GPT Image 2 and Sora 2 are at OpenAI, Nano Banana Pro and Veo 3.1 at Google. There's no European equivalent, so if you create images or video, you can't make that switch.

The third is depth of features. Claude Pro includes Claude Code and Claude Cowork, Google AI Pro includes NotebookLM, Deep Research and 2 TB of storage. Le Chat Pro is a strong, cheap chat with a no-telemetry mode, but it's a smaller package.

Is switching sensible for you?

Sober advice, per situation.

You use AI privately for texts, ideas and questions without sensitive content. Then this isn't an urgent matter. Choose on quality and price; if you like Le Chat, it also saves you more than €5 per month compared with the American plans.

You're a Dutch freelancer (zzp'er) or run a small business and customer data ends up in your prompts. Then it is urgent. Set up a service whose processing you can justify, and request the data processing agreement before you start. More on that trade-off is in AI for freelancers and small businesses.

You work in healthcare, education or government. Then your organization usually decides this for you already, and the question is mainly which service works best within that framework.

You want both. That's the most common answer: a European service for sensitive work and an American one for the rest. That normally means two subscriptions, and that's exactly the kind of stacking that drives up costs.

One subscription that includes multiple models solves that for your wallet. What it doesn't solve is the question of where each underlying model runs: that stays the same, regardless of which interface you use it through. That's exactly why the next article in this series is about the practical side.

If you're looking for a concrete list of alternatives, you'll find them on our page with ChatGPT alternatives and in the article on European ChatGPT alternatives and the GDPR.

Conclusion