The Layer Europe Can Keep
Two weeks ago I wrote that someone else holds the off-switch: a single directive out of Washington disabled Anthropic's models for every foreign national on Earth, eleven days after Europe was first let in to test one. The lesson was that the frontier is borrowed, and the defensible ground is the layer one step behind it: applied systems, governance, sovereign deployment, the parts you can actually own. A model you rent can be switched off by memo. The layer above it cannot.
This is the case for that layer, and what it looks like in practice. We build it. It is called Ingram Cloud, and it runs today.
Sovereignty is not where the data centre sits
The comforting version of AI sovereignty is a map with a server on it. Build a data centre in Frankfurt, the story goes, and the problem is solved. The off-switch proved that wrong in an afternoon. The model was withdrawn by the company that controls it; a data centre in Frankfurt would have sat there holding nothing. Hosting matters, but it is the floor, not the building.
Real sovereignty is freedom from coercion: the ability to choose which model runs, to switch provider without rebuilding your work, to keep your data under your own law, and to prove afterwards what the system actually did. That is a software problem, not a real-estate one. It needs a control layer that sits above the model, and that is the part of the stack Europe has spent the least time building.
So we built it. Here is what it has to do, and how we do it.
1. Choose any model. Swap it without a rewrite.
The fastest way to be coerced is to marry your business logic to one provider's proprietary surface. Then a price change, a deprecation, or an export-control memo lands on a system you cannot move.
Ingram Cloud is built on open standards. In plain terms: nothing about the way you build on it is specific to one model vendor, so no single vendor can hold your system hostage. For your engineers, that means you connect through the OpenAI-compatible API most of the industry already speaks, tools plug in over the Model Context Protocol, and you bring your own model keys and point them where you want. Because an agent's design is held separately from the model behind it, changing provider is a configuration change, not a rebuild. You can run the best American model today and replace it next week. Prompts may need tuning for the new model; the system around them stays put.
2. European hosting, European jurisdiction.
Ingram Cloud runs on European infrastructure, under European law. Your data, your run state, and the orchestration around your agents stay within EU jurisdiction, inside the GDPR and the EU AI Act rather than adjacent to them. There is one honest caveat: if you point an agent at a model hosted abroad, that single call leaves jurisdiction, because the model does. Everything that holds your data and your history stays in Europe, and you can keep inference itself in jurisdiction by running European or open-weight models. Each end user runs as an isolated tenant with its own memory and connections, so isolation is the default rather than something you remember to switch on.
This is the part most sovereignty pitches skip: not a European label on an American backend, but the system itself held where your regulator can reach it and a foreign directive cannot.
3. Prove what happened.
A regulated institution cannot deploy what it cannot account for. Under the EU AI Act and an ISO/IEC 42001 management system, "the model decided" is not an answer; you need evidence of what ran, under which configuration, on whose data.
Ingram Cloud records the run. Every execution leaves a trail of events; an agent's published design is versioned, and each running instance keeps its own revision history, so a model swap or a policy change is a dated, inspectable fact rather than something you reconstruct after the audit. The trail your risk function asks for is built into the platform.
4. Decide what the agent may do, in the moment.
Governance that lives only in a policy document changes nothing once the system is running. Ingram Cloud enforces it where the work happens: agents act through scoped tools, and sensitive actions can pause for human approval before they run. You decide what an agent is allowed to do while it is doing it, and that decision is recorded with everything else.
What we are not trying to win
We do not train the frontier, and we are not asking Europe to. That is the expensive race the off-switch piece argued Europe should stop trying to win. What we own, and what you can own through us, is the layer that makes any frontier model, American or European, open-weight or hosted, usable inside a real institution without surrendering control of it. Many models, with fallbacks, under European jurisdiction, and accountable.
A memo can switch off a model. It cannot switch off the layer above it. That layer is the one worth keeping, and it is the one we built.
Jerome Leclanche<br /> CEO, Ingram Technologies
About Ingram Technologies
Ingram Technologies is a European AI research and development lab, founded in 2022 and based in Brussels. We build applied, agentic AI systems and advise on AI governance and ISO/IEC 42001 compliance, with a focus on responsible, sovereign AI for European organisations. Ingram Cloud is our platform for deploying and governing AI agents under European jurisdiction.
Media enquiries: press@ingram.tech