Model as a Service

Enterprise AI models via a secure API ready-to-use - sovereign infrastructure, no GPU ownership, no data leakage. Built for regulated industries and privacy-first teams. 

Swiss Sovereignty by Design 

• All models, inference infrastructure, and gateways are hosted on sovereign Cloud resources in Switzerland
• Operated under Swiss law, aligned with Swiss data protection and regulatory expectations 

Data Protection First 

• No prompt or response logging
• No use of customer data for model training
• Designed for workloads involving highly sensitive enterprise and regulated data 

Enterprise-Grade Control and Economics 

• Access via secure API keys
• Clearly defined TPM (Tokens Per Minute) and RPM (Requests Per Minute) rate limits
• Configurable monthly budgets for predictable and auditable costs
• Usage-based pricing aligned with token consumption
• No upfront GPU investment, no capacity planning, no hardware lifecycle risk  

Operational Simplicity 

• No GPU procurement or management
• No model lifecycle or scaling infrastructure to operate
• Enterprise-ready catalogue of open-weight LLMs to choose from 

Generative AI is easy to experiment with, but running it responsibly, securely, and economically at scale is not. 

Many organizations experiment with AI using public APIs or first hardware purchases. In practice, this quickly leads to challenges: 

• GPU hardware is expensive, evolves rapidly, and is poorly suited for POCs or early production phases
• Upfront GPU CAPEX creates risk when usage patterns, models, and requirements change quickly
• GPU capacity on hyperscalers is often limited or unpredictable, delaying projects
• The data enterprises want to use with AI is often their most sensitive data, which should not go out uncontrolled
• Building and operating an AI platform from scratch is complex, time-consuming, and costly 

Model as a Service (MaaS) from ELCA Cloud Services addresses these challenges by providing controlled, sovereign access to AI models without hardware ownership and without platform complexity. 

MaaS enables organizations to adopt AI with clarity and speed - not compromise. 

• Inference Location and Jurisdiction: Switzerland
• API Interface: OpenAI-compatible REST API
• Models: Curated catalog of enterprise-ready open-source models, including Large Language Models (LLMs), multimodal models, and embeddings
• Pricing Model: Token-based, usage-aligned, budget-controlled
• Governance Controls: Budgets, rate limits, team separation, and full data control
• Operations: Managed Cloud service, available 24/7, with business hour SLA
• Support: Best effort support access included, extended support options available 

MaaS fits seamlessly into existing application architectures with minimal integration effort. This allows teams to: 
 

• Start quickly with POCs
• Scale into production without infrastructure redesign
• Align costs to actual usage, not peak hardware capacity 

MaaS provides access to a curated and continuously evolving catalog of enterprise-ready open-source AI models, covering the most common enterprise AI workloads: 

• Chat and Reasoning: Apertus 70B, DeepSeek V3.2, GLM 4.5 Air 110B, GPT OSS 120B, Granite 3.3 8B, Mistral 7B Instruct v0.3, Qwen3 8B, QwQ 32B
• Multimodal: Gemma 3 12B IT, Gemma 4 31B, Granite Vision 3.2 2B, Llama 4 Maverick, Llama 4 Scout 17B, Qwen3 VL 235B
• Embeddings and Reranking: BGE M3, BGE Reranker v2 M3, Granite Embedding 278M
• OCR and Speech-to-Text: DeepSeek OCR, MinerU 2.5, Whisper Large v3 

Models follow defined update, deprecation, and replacement policies to ensure stability, security, and predictability. 

MaaS is designed as the foundation of a broader sovereign AI platform, not a standalone silo. 

Depending on your needs, MaaS can be extended with: 
 

• LLM Chat Interface as user chatbot
• Retrieval-Augmented Generation (RAG) for enterprise knowledge search
• Model Context Protocol (MCP) to interact with data
• AI Agent Workflows and integrations
• MLOps Tooling for own AI development
• Dedicated LLM Server or Gateway for special requirements
• AI Landing Zone to create a secure environment
• Managed AI Platform services for end-to-end operations 

This enables organizations to progress from model access to a fully sovereign enterprise AI platform — without re-architecting or losing governance. 

Organizations typically adopt AI through different paths: 

• Public AI APIs
• Self-hosted models
• Hyperscaler AI platforms
• Sovereign Model as a Service 

In practice, most enterprises combine and evolve these approaches over time. 

ELCA Cloud Services supports organizations across all of these paths - and helps design, build, and run pragmatic, compliant, and economically sustainable architectures.