{"slug": "run-frontier-llms-on-sovereign-eu-infrastructure", "title": "Run frontier LLMs on sovereign EU infrastructure", "summary": "Corti, a Danish health-tech company, launched Corti Models, a service that runs frontier large language models on sovereign EU infrastructure, with the entire stack open source and verifiable down to the GPU. The service is OpenAI-compatible, allowing developers to switch from US providers by changing the base URL and API key, and offers $50 of free credits for first-time users. Corti Models runs on Kommodity, an open source infrastructure layer, ensuring no US cloud provider is in the request path and compliance with ISO 27001, ISO 42001, GDPR, NIS2, DORA, and EU AI Act.", "body_md": "# Run frontier LLMs on sovereign EU infrastructure\n\nCompliant, secure, and verifiable down to the GPU. Frontier models on European hardware, in your coding agent or called directly from your application. Your data never leaves your region.\n\n## You shouldn't choose between frontier models and control\n\nNo operational control\n\nInference location, security, and underlying changes sit with the provider - not with you.\n\n**One-vendor lock-in**\n\nPrompts and workflows harden around a single API. Unwinding costs more than building did.\n\n**Costs you can't forecast**\n\nRates and limits move on their schedule, so capability gets rationed to budget.\n\n**Foreign jurisdiction**\n\nContract and access terms can shift with a political decision, not a commercial one.\n\n## In the coding agent, or in your app\n\n### Agentic coding in the terminal\n\nPoint OpenCode, ForgeCode, Crush, or Pi at Corti Models with the Corti CLI. Keep the workflow you already have.\n\n```\n# run the setup wizardnpx @corti/cli init models# load credentials, then launch your agentset -a; source ~/.env; set +aopencode\n```\n\n### Direct API in your product\n\nOpenAI-compatible. Existing app code works with a base URL change and a new key.\n\n```\nclient = OpenAI(    base_url=\"https://ai.eu.corti.app/v1\",    api_key=\"<your-api-key>\")r = client.chat.completions.create(    model=\"corti-s1\", messages=msgs)\n```\n\n## Pick the model.\n\nKeep the jurisdiction.\n\n## AI projects can stall because of the provider\n\nA working prototype\n\nSomeone on your team builds the feature against OpenAI or Anthropic. It demos well. The business case is obvious.\n\nSecurity, legal, procurement\n\nData residency cannot be answered. The transfer cannot be justified. The cost at scale cannot be approved. The feature sits in review and the quarter ends.\n\nThe same code, cleared\n\nChange the endpoint and the credentials. Keep your application logic, your prompts, and your evaluation set. Run it on infrastructure that passes review the first time.\n\n## Verifiable at every layer\n\nMost vendors claim sovereignty in the contract, then rent the stack. Corti Models runs on Kommodity, an open source infrastructure layer, so the entire stack, including encryption, isolation, and security, is public and auditable, not just claimed.\n\n- Attestation before inference - isolation enforced by hardware, not policy\n\n- No US cloud provider in the request path\n- Infrastructure code open source on\n[GitHub](https://github.com/kommodity-io/kommodity)for anyone to review\n\n## Private by architecture\n\nPrompts and completions process in memory and end with the request. Nothing is retained, nothing trains a model, nothing is visible to another tenant.\n\n- Central control over access, data handling, and per-team budget\n- Sovereign cloud or on-premises - same platform as clinical AI in production\n- ISO 27001, ISO 42001, GDPR, NIS2, DORA, and EU AI Act posture by default\n\n## The weights are the easy part\n\n## Today’s leader isn't tomorrow’s\n\nCorti evaluates and operates the best available models as the market moves - without you rebuilding integrations every leaderboard cycle.\n\n## Pay for tokens, not seats\n\nPer-seat coding assistants charge whether developers use them or not. Governed consumption charges for what runs, at European infrastructure cost.\n\n- One credit balance across Corti APIs\n- Same credits for development, testing, and production\n- Per-team limits and request-level reporting\n\n**Jesper Carøe , CEO, Trifork Digital Health**\n\nEarly adopter and implementation partner\n\n## First request in under five minutes\n\nCreate a key, change the base URL, make the call. OpenAI-compatible - any existing SDK works without a rewrite.\n\n- Generate an API key in the console, start with $50 of free credits\n- Point any OpenAI SDK or curl at ai.eu.corti.app\n- Or install the Corti CLI and connect your terminal agent\n\n## Frequently asked questions\n\n**How is this different from a European company reselling US models?**\n\nA reseller changes who invoices you. The inference still runs on infrastructure under US jurisdiction and the transfer still happens. Corti operates the hardware, so your request is processed on machines we run, in a European facility, under European law.\n\n**What happens if we want to leave?**\n\nThe API is OpenAI-compatible, so moving off Corti Models is the same base URL change as moving on. We do not hold your prompts, completions, or tuning data. The compatibility that makes adoption cheap is what makes exit cheap.\n\n**Can we run this on our own hardware?**\n\nYes. Corti Models is available in sovereign cloud and on-premises deployments, on the same platform used for on-premises clinical AI today. On-premises changes the commercial and deployment timeline, so raise it early.\n\n**How do you handle model updates and deprecation?**\n\nModel versions are pinned. You choose when to move, with advance notice and an overlap window on the previous version. Nothing changes underneath a running workload because a vendor shipped a new default.\n\n**What is the migration effort?**\n\nFor direct API use, a base URL and a key. For terminal coding agents, installing the Corti CLI and pointing an existing agent at it. Teams typically have a working integration in an afternoon and a governance model agreed in a few weeks.\n\n**Some of the strongest open-weight models were developed in China. How is that sovereign?**\n\nSovereignty is about who operates the infrastructure and whose law applies, not where the research was done. Open weights are a static artefact. Corti deploys them on hardware we operate in Europe, and no data reaches the original developer. You get the capability without the dependency, which is not possible with a closed model you can only reach through its owner's API.\n\n**We are not in healthcare. Is this for us?**\n\nCorti built this for healthcare because healthcare is the hardest compliance environment there is. The requirements that follow, data residency, auditability, and strict privacy controls, are the same ones facing financial services, energy, public administration, defence, and aviation.\n\n## Run your next coding task on European hardware\n\nInstall the CLI, keep the agent your team already uses, and point it at models hosted in the EU.", "url": "https://wpnews.pro/news/run-frontier-llms-on-sovereign-eu-infrastructure", "canonical_source": "https://corti.ai/models", "published_at": "2026-08-20 09:19:21+00:00", "updated_at": "2026-08-20 09:44:18.975722+00:00", "lang": "en", "topics": ["large-language-models", "ai-infrastructure", "ai-products", "ai-policy"], "entities": ["Corti", "Kommodity", "OpenAI", "Anthropic", "Trifork Digital Health", "Jesper Carøe", "EU AI Act"], "alternates": {"html": "https://wpnews.pro/news/run-frontier-llms-on-sovereign-eu-infrastructure", "markdown": "https://wpnews.pro/news/run-frontier-llms-on-sovereign-eu-infrastructure.md", "text": "https://wpnews.pro/news/run-frontier-llms-on-sovereign-eu-infrastructure.txt", "jsonld": "https://wpnews.pro/news/run-frontier-llms-on-sovereign-eu-infrastructure.jsonld"}}