Your industry: Manufacturing and industry
Tailored to manufacturing: the traffic light rates trade secrets rather than personal data, which is why yellow also covers US processing with a contract and confidentiality. Exception: export-controlled technology (dual-use, defence, US-origin goods) stays red outside Switzerland until the licensing question is resolved.
Works with client data: the setup is proven, contract and governance are in place. This is exactly how you should do it.
Depends on the contract, configuration and industry: feasible with a documented risk assessment, not without it.
Not defensible for client data in our risk assessment: no contract, no control or no documented safeguards. Often still the best choice for research, marketing and anonymised content.
In manufacturing the traffic light primarily rates trade secrets rather than personal data: yellow here also covers processing in the US, provided a contract with confidentiality and a no-training clause exists. Exception: export-controlled technology (dual-use, defence, US-origin goods) counts as red for any processing outside Switzerland until the licensing question is resolved; even with Swiss data centres, check whether support staff can access from abroad. As soon as personal data is affected, the general legend applies, and machine data can be personal data, for example when a system assigns identifiers to operators.
Filter the table by traffic-light colour. On a smartphone it scrolls sideways.
| Solution | Client data | Models | Data location | Model quality | Effort | Cost | What to watch |
|---|---|---|---|---|---|---|---|
Azure OpenAI Switzerland (Standard) Microsoft Azure, Switzerland North region | Green | GPT-4o (until April 2027), Embeddings | Processing Switzerland Storage Switzerland | Solid, several generations behind the frontier | Medium, integration by IT or partner | Low, per token, no base fee | A Swiss data location protects construction data most easily: choose the Standard deployment type (not Global or Data Zone), contract with confidentiality and a no-training clause. For export-controlled technology, additionally check whether the provider's support staff can access from abroad. For new customers, GPT-4o (until April 2027) is effectively the only option; GPT-4.1 is deprecated. |
Azure OpenAI Switzerland (Provisioned) Microsoft Azure, Switzerland North, reserved capacity | Green | GPT-5.1 (until May 2027) | Processing Switzerland Storage Switzerland | Very good, strongest model with CH guarantee | Medium, integration by IT or partner | From approx. USD 13,000 per month (minimum 50 PTU) | Like the Standard variant, but with regional Provisioned capacity in Switzerland North (not Global or Data Zone Provisioned); with high volume, e.g. quote calculation or technical documentation. GPT-5.1 (until May 2027) costs from approx. USD 13,000 per month (minimum 50 PTU), and a successor in Switzerland is not guaranteed. |
Swiss AI platforms e.g. AIgent, Swisscom, Phoenix Technologies (kvant), Infomaniak | Green | Open models such as DeepSeek, Qwen, GLM, Apertus | Processing Switzerland Storage Switzerland | Good to very good, a step behind the frontier | Low, ready-made interface, some with agents | Medium, licence per user, sometimes plus usage | Well suited and the preferred option for export-controlled data, provided operations and administration are in Switzerland; selectively disable optional US models. |
AI in your industry software e.g. practice, law firm, CRM or portfolio systems | Depends on provider | Usually GPT or Claude in the background | Processing Depends on provider Storage Depends on provider | Good, tailored to the use case | Low, the system is already in use | Low to medium, often as an add-on module | CAD, PDM or ERP vendors: clarify in writing which model runs where and whether training on your data is excluded. |
Microsoft 365 Copilot Microsoft, with EU Data Boundary | Yellow | GPT models from OpenAI, operated by Microsoft | Processing EU/EFTA Storage Switzerland (Data Location or ADR) | Very good, strong in everyday office work | Low, licence and configuration | Medium, licence per user and month | Defensible for trade secrets: Flex Routing off, Anthropic models off, permissions cleaned up. |
Azure OpenAI Data Zone EU Microsoft Azure, resource in Switzerland North | Yellow | Latest GPT generation, e.g. GPT-5.4 to GPT-6.1 | Processing EU Data Boundary, confirm countries with Microsoft before use Storage Switzerland | Frontier, latest GPT models | Medium, integration by IT or partner | Low, per token | EU processing defensible with contract and confidentiality, but not for export-controlled technology (dual-use, defence, US origin): red applies there until the licensing question is resolved, and a contract does not replace an export licence. For personal data, respect the DPA rules. The EU Data Boundary can also include EFTA countries: confirm the affected countries with Microsoft before use. |
Claude via AWS Bedrock (EU) Amazon Web Services, EU regions | Yellow | Opus 5 and Sonnet 5 in Ireland or Stockholm, Opus 5.5 and Sonnet 5.5 via EU inference | Processing EU Storage No prompt storage by Bedrock | Frontier, especially strong for text and analysis | Medium to high, AWS account and development needed | Low to medium, per token | Like Data Zone EU, including the exception for export-controlled technology. |
Gemini via Google Vertex AI (EU) Google Cloud, EU regions | Yellow | Gemini models | Processing EU Storage EU or Switzerland | Frontier, strong with long documents | Medium to high, Google Cloud project and development | Low to medium, per token | Like Data Zone EU; verify the processing location. The same exception for export-controlled technology. |
Claude directly from Anthropic Claude API, Team and Enterprise | Yellow | Claude Fable 5.1, Opus 5.5, Sonnet 5.5 | Processing US or global Storage US | Frontier, newest Claude models first | Low, ready to use | Medium, licence or per token | Yellow here applies only to content without personal data, without an NDA bar and without export control: contractually agree confidentiality and a no-training clause, and weigh the residual US CLOUD Act risk for construction data. Export-controlled technology (dual-use, defence, US-origin goods) stays red, and customer NDAs often exclude disclosure to third parties or abroad. As soon as personal data is involved, the red rating applies again. |
Claude in Microsoft Foundry or Copilot Microsoft, operated on Anthropic infrastructure | Yellow | Claude models | Processing Outside the EU Data Boundary Storage Not guaranteed | Frontier, same as Claude direct | Low, switchable at a toggle | Medium, in licence or per token | Yellow here applies to technical documents without personal data, without an NDA bar and without export control, provided a contract, confidentiality and a no-training clause are in place. For dual-use, defence and US origin the rating stays red. |
Copilot with Flex Routing enabled Microsoft 365, setting available since April 2026 | Yellow | GPT models | Processing At peak load also US, Canada, Australia Storage Switzerland | Very good, same as Copilot | Low, a single setting | Medium, licence per user and month | Yellow here applies only with a contract, confidentiality and a no-training clause and only to content without personal data, without an NDA bar and without export control. For dual-use, defence and US origin it stays red; with personal data or critical intellectual property, disable Flex Routing. |
Private AI accounts ChatGPT, Claude.ai, Gemini with a private or free account | Red | All of them | Processing Unclear, mostly US Storage US | Frontier, which makes it so tempting | None, and that is exactly the problem | Low, free or privately paid | Off limits even without personal data: no contract, no confidentiality, possible training on your construction data. For inventions before the patent filing, the unclear disclosure additionally endangers novelty. |
Microsoft, AWS and Google are US corporations and are subject to the US CLOUD Act, even with data centres in Switzerland or the EU. This residual risk belongs in every risk assessment.
Hardly any organisation needs just one tool. What has proven itself is the combination of an AI employee for daily work and a secure interface for automations.
For manufacturing we recommend the high-performance setup: with a contract and confidentiality, frontier models are fine; personal data runs separately through the Swiss pipeline. For export-controlled data, the Swiss pipeline is the only option, regardless of model quality.
In manufacturing it is mostly about trade secrets rather than personal data: construction data, calculations and customer lists are protected by contracts and the Unfair Competition Act (UCA), not by the DPA. The first topic, however, is often export control: drawings, specifications and manufacturing know-how of dual-use goods count as technology whose transfer can be a licensable export, including electronic transfer. There is no SECO guidance on cloud and AI yet; when in doubt, ask SECO directly. On top come customer NDAs, which can exclude disclosure to third parties or abroad, and patent law: if an invention is disclosed before filing, it is no longer novel. Where personal or customer contact data is involved, the DPA applies in addition, including machine data with operator identifiers.
This overview does not replace legal advice. Model availability and regions change constantly; October 2026 edition. Verify details with the provider before use.
Taskstreams builds and operates the pipeline that fits your policy: Switzerland only with Azure OpenAI Switzerland North, or the frontier model in the EU while your client data stays in Switzerland. Contract and governance are included; you receive finished AI employees for your processes, from document review to client reporting, instead of your own infrastructure project.