Swiss-jurisdiction AI · non-EU · non-Cloud-Act
Your AI. Local and private.
AI infrastructure under Swiss law — outside EU and US Cloud Act reach. Dedicated compute for autonomous pipelines, domain-specific retrieval, and private model hosting.
Hybrid, Private or Air-Gapped
Why Silex Business
Built for regulated organisations
Swiss data residency
Your data never leaves Switzerland. Fully beyond the reach of the US Cloud Act.
Regulatory compliance
nDSG and GDPR compliant. Your data is encrypted — only you hold the key. Audit-ready.
Memory-dense architecture
384 GB unified memory per cluster. Multiple large models remain resident simultaneously — no reloading between autonomous agent calls.
Flat-rate pricing for AI pipelines
Fixed pricing for autonomous pipelines generating 1M+ tokens daily. Per-token APIs are designed for human usage patterns, not agents.
Auditable inference
Model version pinning, retained inference logs, and jurisdictional residency certificates — packaged with every contract for regulator inspection.
Strategic AI Implementation
Our implementation approach
While others pursue blanket AI rollouts, we believe in targeted deployments that deliver measurable impact with controlled investment.
Targeted deployment
Focus AI where it matters most. Identify high-impact use cases rather than implementing AI everywhere. Strategic placement delivers better ROI than broad, unfocused rollouts.
Measured investment
Prove value with controlled pilots before scaling. Our approach minimises risk while maximising learning and demonstrable business impact.
Controlled scaling
Expand success, contain failure. Maintain control over costs, compliance, and quality throughout your AI journey.
Industries
Purpose-built for your sector
FinTech & financial services
Cloud Act exposure, FINMA obligations, cross-border data transfers.
Swiss-only inference with no data export. Flat fee instead of variable API costs.
MedTech & healthcare
Patient data, HMG requirements, strict access control.
Dedicated environment with strict data residency. No prompt leaves the data centre.
Legal & compliance
Attorney-client privilege, confidential client documents, professional secrecy.
Sovereign AI processing for contract review and compliance analysis — plus a curated Swiss legal knowledge base with source-cited answers from BGE, fedlex, and cantonal legislation.
Swiss SMEs
AI without US hyperscaler dependency. SaaS vendors need Swiss inference for regulated end clients. ML teams need a place to host fine-tuned models.
White-label inference for SaaS products, bring-your-own-model hosting for ML teams, and private endpoints — all at flat rates with no lock-in.
Services
Our B2B offerings
Nine specialised offerings for Swiss organisations and universities with data residency and sovereignty requirements.
01
Sovereign inference endpoint
Private OpenAI-compatible API on dedicated Swiss hardware. One tenant per cluster.
Target clients
Swiss SaaS vendors, ISVs, internal platform teams with nDSG or FINMA obligations.
Delivery model
Private REST API — OpenAI-compatible, no client-side code changes.
Pricing model
Fixed monthly subscription per cluster (CHF/month), not per-token.
02
Sovereign RAG stack
Language model and vector database on the same Swiss infrastructure — no cross-border data flows.
Target clients
Law firms, insurance companies, cantonal administrations, compliance teams.
Delivery model
Vector database and language model on the same cluster. Documents never leave the data centre.
Pricing model
Fixed monthly subscription, configuration by index size.
03
Domain fine-tuning
LoRA/QLoRA fine-tuning on open-weight models with your domain data, entirely in Switzerland.
Target clients
Organisations with domain-specific data — legal German, Swiss German, FINMA taxonomy.
Delivery model
One-time project: LoRA adapter or merged weights as deliverable.
Pricing model
Project rate (CHF/day), fixed scope.
04
Biomedical environment
Dedicated environment for healthcare workloads with enhanced data-handling under HMG and DSG.
Target clients
Biotech startups, private clinics, university hospital spin-offs, medical devices.
Delivery model
Private API with biomedical base model and stricter data-handling and access-control requirements.
Pricing model
Monthly subscription plus DPA addendum for health data.
05
Swiss SaaS integration
White-label inference for SaaS vendors offering Swiss AI features to their clients.
Target clients
SaaS vendors in legal, accounting, HR, payroll — regulated end clients.
Delivery model
Wholesale inference, SaaS vendor is the operator.
Pricing model
Wholesale rate, 10–20 % below retail price.
06
Swiss Hybrid AI
Tailor-made architecture from fully air-gapped operation to secure integration with hyperscalers or external AI providers.
Target clients
Organisations looking to combine private AI with existing cloud services or external AI providers.
Delivery model
Architecture consulting and implementation: from air-gapped operation to secure integration with Azure, AWS or other providers.
Pricing model
Project-based or monthly retainer, depending on integration depth.
07
Research Compute
Dedicated GPU cluster at a flat monthly rate for universities, universities of applied sciences, and SNF-/Innosuisse-funded research projects.
Target clients
Research leads and PIs at universities of applied sciences (e.g. HSLU), university hospitals, and SNF-/Innosuisse-funded projects.
Delivery model
Dedicated cluster with SSH and API access. Pre-configured with common frameworks, or bare-metal access for custom stacks. Infrastructure letters for grant applications included.
Pricing model
Fixed monthly subscription at a preferential academic rate (below commercial rate, contact us regarding your idea). Predictable for budget requests and grant reporting.
08
Domain knowledge retrieval
Curated, continuously maintained Swiss knowledge base — co-located with inference on dedicated hardware. Query domain-specific content with source-cited answers.
Target clients
Law firms, tax advisors, compliance teams, FinTech companies, cantonal administrations — any organisation with recurring domain-specific retrieval needs.
Delivery model
Managed vector index with periodic corpus refresh. Co-located with inference — no cross-border data flows. Example domains: Swiss case law and legislation (BGE, fedlex, Amtsblatt); financial regulatory corpus (FINMA circulars, SIX listing rules, AML guidance).
Pricing model
Monthly subscription, scoped by corpus size and query volume.
09
Bring-your-own-model hosting
You supply the model — we host it on dedicated Swiss hardware. Hosting-only contract: you own the weights and outputs, we own the uptime.
Target clients
ML teams at Swiss SMEs with fine-tuned models, university research groups, AI consulting firms training models for clients.
Delivery model
Client uploads model weights (safetensors, GGUF, HuggingFace format). Served via vLLM or TGI on dedicated cluster. Version management and rollback included.
Pricing model
Monthly subscription. Hosting-only scope — lighter compliance surface, explicit liability split.
How we work
Every project starts with a personal conversation. We take the time to understand your use case and build a solution together that fits.
Direct contact
No ticket system, no call centre. You speak directly with the person who manages your environment.
Tailored solution
No off-the-shelf offering. Every configuration is adapted to your needs.
Long-term partnership
We think in relationships, not contract terms. Our clients know us personally.
Business pricing
Cluster pricing
Dedicated Swiss infrastructure with predictable monthly rates. Run sovereign — or integrate with Azure, AWS, or other providers via BYOK and secure API bridging. You control the boundary.
Silex 1×
CHF 900
/month
1 node · 128 GB unified memory Models up to 70B parameters
RAG, chat, single-model inference
Get in touch Most popular
Silex 2×
CHF 1,500
/month
2 nodes · 256 GB unified memory Models up to 235B parameters
Multi-agent workflows, fine-tuning
Get in touch Silex 3×
CHF 2,400
/month
3 nodes · 384 GB unified memory Models up to 400B parameters
Full model residency, sovereign agent operations
Get in touch What is included
- Dedicated cluster — 1, 2, or 3 nodes depending on configuration
- Unlimited tokens, flat fee — no metering
- Swiss data residency (Root, Lucerne — Tier-3 data centre)
- OpenAI-compatible endpoint
- RAG stack and agent orchestration
- nDSG / GDPR compliant, your data encrypted — only you hold the key
- Self-serve onboarding (docs + API reference)
- Training pipeline for LoRA/QLoRA (up to 400B parameters)
- Multimodal training support (text + vision)
Add-on options
Tailored to your requirements
- Guided setup — tailored to your requirements
- Managed onboarding + SLA — tailored to your requirements
- Private Cloud isolation — tailored to your requirements
- On-prem / air-gap deployment — tailored to your requirements
- Domain fine-tuning (LoRA/QLoRA)
- Multimodal training (vision + text)
Cost comparison
How Silex compares
Monthly cost comparison against Swiss cloud and infrastructure providers — by workload tier.
CHF 900
Silex flat fee
CHF 1,584
Highest comparable
CHF 684
Potential saving
CHF 1,500
Silex flat fee
CHF 6,336
Highest comparable
CHF 4,836
Potential saving
CHF 2,400
Silex flat fee
CHF 6,336
Highest comparable
CHF 3,936
Potential saving
For small workloads, per-token services are cheaper. Silex Radix becomes cost-effective at medium volume — and offers dedicated infrastructure, a Swiss legal entity, and no Cloud Act exposure from day one.
Estimates based on blended token rates (3:1 input:output ratio). Infomaniak: published pricing (2024). Nine: nine.ch calculator (April 2026). Oriented: published GPU server pricing oriented.net. Cloudscale: published GPU pricing cloudscale.ch (~CHF 2.20/hr RTX Pro 6000). As of April 2026.
Questions
Frequently asked questions
Is this GDPR compliant?
Yes. Swiss nDSG and EU GDPR. No data leaves Switzerland. No inference data is used for model training. Your data is encrypted at rest — only you hold the key. Silex Radix cannot access it.
What model do we get?
The platform supports any open-weight model that fits your cluster's memory envelope. The default deployment is Qwen3.5-397B on a dedicated tensor-parallel cluster. Alternative models (Llama, DeepSeek, domain fine-tuned variants) can be deployed on request.
Can we connect our own tools?
Yes. The OpenAI-compatible endpoint works with LangChain, LlamaIndex, any OpenAI SDK, and custom pipelines.
Is there an SLA option?
Yes. Every subscription includes automated monitoring with incident acknowledgement within two hours on business days. A managed SLA add-on provides a 99.5% monthly uptime target, defined resolution targets, proactive notifications, and service credits. Details are shared during the scoping conversation.
Can we train our own models?
Yes. 384 GB unified memory enables LoRA/QLoRA fine-tuning of models up to 400B parameters. You upload your training data to an encrypted volume and manage the training process via API — only you can access the data. Training and inference remain entirely in Switzerland.
What training data do you support?
Text, code, structured data, and multimodal (text + image). Specialised for Swiss German, legal texts, and FINMA-regulated content.
How does your stack compare to other providers?
Our stack (Ubuntu, Docker, CUDA, PyTorch) is very similar to oriented.net GPU servers — but fully in Switzerland with no Cloud Act exposure.
What does "private hosting" mean — isn't it still a cloud?
Any hosted infrastructure is technically a "cloud" — but not all clouds are equal. Our clusters are located in a Swiss data centre, operated by a Swiss GmbH, with no access by foreign jurisdictions. For a clear-eyed look at cloud terminology, we recommend the article «Die Cloud gibt es nicht» on dnip.ch. «Die Cloud gibt es nicht» — dnip.ch
What is domain knowledge retrieval?
A continuously maintained index of domain-specific Swiss content — co-located with inference on your cluster. You query it like a search engine but receive synthesised answers with source citations. Available domains include Swiss case law and legislation, and financial regulatory content (FINMA circulars, SIX rules). Additional domains on request.
Can we host our own model?
Yes. Upload your fine-tuned or proprietary model weights (safetensors, GGUF, or HuggingFace format) and we serve them on dedicated Swiss hardware. You retain full ownership of the model and its outputs — our contract covers infrastructure availability only. Version management and rollback are included.
Service level
Availability and support
Every subscription includes baseline monitoring. The managed SLA add-on adds contractual commitments.
| Standard (included) | Managed SLA (add-on) | |
|---|---|---|
| Availability target | Best effort | 99.5% monthly |
| Incident acknowledgement | ≤2h on business days | ≤2h business days · ≤4h weekends |
| Incident resolution | Best effort, typically same day | 4h software · next business day hardware |
| Planned maintenance | 72h notice · pre-agreed windows | 72h notice · pre-agreed windows |
| Service credits | — | 5% per 0.1% below target (capped at 30%) |
| Monitoring | Client dashboard · proactive notifications | Client dashboard · proactive notifications |