AI Hosting in Switzerland: Global Cloud, Swiss Provider or Own Infrastructure?

There is no single right hosting variant for AI in a Swiss financial services firm. The choice depends on three questions: how sensitive is your data, what is your budget, and what model quality do you need? This article compares the three variants, a global cloud with a Swiss region, Swiss providers, and your own infrastructure, with their strengths, their limits, and the criteria for your decision.

Why is "where does my data sit?" the first question?

In almost every first conversation with a small financial firm, "where does my data sit?" is the first question, before features and before pricing. That is no coincidence. If you manage client wealth or policies, confidentiality is embedded not just contractually but culturally. The question is also justified, because the storage location helps determine which duties are triggered, for example the cross-border disclosure of personal data under the revised Swiss Data Protection Act. The ten most important of these duties are summarized in the revFADP checklist for financial firms. The good news: today there is a fitting technical variant for every answer to the location question.

Variant 1: Global cloud with a Swiss region

The large cloud providers operate data centers in Switzerland. Microsoft Azure runs in the Zurich region (Switzerland North) and the Geneva region (Switzerland West) (Microsoft, press release of 28 August 2019); AWS and Google Cloud run their own Swiss regions. Your data physically stays in the country, but you use the platform of an international corporation.

The advantages: you get current models as a ready-made service, high availability, extensive certifications, and business contracts with commissioned processing and a no-training clause. The limits: the parent company sits abroad, so the question of possible access from third countries belongs on the table. And pricing is variable; it needs some monitoring in operation.

From my practice: Our own solution runs in a protected Swiss cloud on Microsoft Azure in Zurich North. We chose this variant because it combines a Swiss data location with the quality of current models, including a business contract with a no-training clause. For most of the small financial firms we work with, this is the best mix of protection, cost and capability.

Variant 2: Swiss providers

Providers such as Infomaniak, Exoscale, Green or Swisscom operate their own data centers in Switzerland, with a Swiss legal entity under Swiss law. Infomaniak advertises its public cloud with entry prices from around CHF 16 per month (Infomaniak Public Cloud, as of October 2026); Exoscale runs zones in Geneva and Zurich (Exoscale data centers, as of October 2026). All provider information changes frequently and should be verified before a decision.

The advantages: Swiss law and Swiss contacts throughout, short distances, often attractive infrastructure pricing. The limits: the latest language models rarely run there as a finished service. If you want them, you operate them yourself or additionally tap into model APIs from the large providers. Model choice and the pace of innovation are smaller than on the big platforms.

Variant 3: Your own infrastructure

Servers in your own building or a rented data center, your own models, your own access architecture. The advantages: maximum control and no dependency on a provider. The limits: high upfront and operating costs, specialist staff for operations and security, and model quality tied to hardware that becomes expensive quickly. For a financial firm with three to fifty employees this variant is usually oversized. It mainly suits larger institutions with their own IT department.

Which criteria decide?

Five criteria decide in practice, weighted by what we hear in first conversations:

Criterion	Global cloud, Swiss region	Swiss provider	Own infrastructure
Data location	Switzerland, physically in the country	Switzerland	Switzerland, your own site
Access from abroad	Parent abroad, govern by contract	Low, Swiss company	None, full control
Cost	Variable, low entry	Mostly low to medium	High
Model quality	Current models as a service	Smaller choice, more self-operation	Depends on your hardware
Effort	Low, provider operates	Medium	High, own team required

Comparison of the three hosting variants, as of October 2026. Provider information changes and should be verified before a decision.

Which variant fits which company?

• Small team, first pilot: a global cloud with a Swiss region. Low entry, fast delivery, extendable towards hybrid later. What that means in numbers is covered in What does AI cost for financial firms?.
• High data sensitivity, Swiss law preferred throughout: a Swiss provider, possibly combined with model APIs under a business contract.
• Larger institution with its own IT: your own infrastructure or a hybrid combination, when existing systems must not or cannot move to the cloud.

For all three variants the same applies: location is one of several protective conditions. At least as important are the contract, access control, and the question of whether your inputs are used to train models. How to frame these points for everyday tools like ChatGPT is shown in ChatGPT in wealth management: what is allowed?. What the supervisor expects in terms of governance and documentation is covered in FINMA's supervisory notice 08/2024 explained.

Note: We are not a law firm. All statements on legal and regulatory topics, as well as on providers and prices, are subject to change and do not replace legal advice. For specific questions, please consult a lawyer.

Conclusion: your next step

Do not start with the provider; start with your data: classify what is public, what can be anonymized, and what is identifiable. The revFADP checklist walks you through the points. Then ask every provider the same three questions: where are the servers, who can access from abroad, and what happens to my inputs during training? Anyone who gets these three answers in writing makes the decision in weeks rather than months. For the full picture on use cases, costs and regulation, read the 2026 guide: AI for financial services in Switzerland. And if you want to see what a protected Swiss environment looks like in practice: on AI for wealth managers we show our solution on Microsoft Azure in Zurich North.

FAQ

Is a US cloud with a Swiss region compliant with data protection law?

Cautiously speaking: the Swiss location prevents storage alone from triggering a cross-border disclosure. What remains open is the question of possible access by the corporation or by authorities from third countries; these risks can be reduced contractually and technically. The assessment for your firm belongs with a professional.

What does Swiss hosting for AI cost?

Orders of magnitude, as of October 2026: entry tariffs of Swiss cloud providers start at around CHF 16 per month for infrastructure, with model costs on top. A complete compliant setup including the connection of your systems sits in the mid five-figure range plus a four-figure monthly amount. The details are in the cost article.

Can I switch providers later?

Yes, if the contract includes an exit clause and the export of your data. Pay attention to this when signing, not when leaving.

Do I need my own GPU servers?

For most small financial firms: no. The models run as a service in the cloud. Own GPU infrastructure pays off only at very high volumes or under special protection requirements.

Is a Swiss data center alone enough protection?

No. Location is one of several conditions. At least as important are the contract with commissioned processing and a no-training clause, access control, and logging. The full overview is in the revFADP checklist.