Bleu, Delos, and the Ceiling of a Licensed Sovereign Cloud

!A cinematic gold figure of Atlas holding a marble globe aloft against a black void, gold embers drifting

# Bleu, Delos, and the Ceiling of a Licensed Sovereign Cloud

By Micky Irons

Two of Europe's most-watched sovereign cloud experiments are now live. Bleu, the joint venture in which Orange and Capgemini operate a Microsoft Azure stack for the French public sector, has moved from announcement to operation. Delos, an SAP subsidiary running an Azure-based platform for the German government, has done the same. For procurement teams who have spent three years asking where the sovereign option is, this is real progress. We want to say that plainly before we say anything else. These are serious operators, serious engineering, and a genuine improvement on shipping regulated data straight into a hyperscaler tenancy with foreign staff holding the keys.

We also want to be honest about where the model stops. A partner cloud built on a licensed platform inherits a ceiling. You can move the operator, the staff, the datacentre, and the support contract onto home soil. What you cannot move is the source of the platform itself. That distinction is the whole argument, and for a public-sector buyer it is the difference between sovereignty as a service level and sovereignty as a property right.

What a partner cloud actually gives you

The Bleu and Delos designs solve a specific, legitimate problem. Under a standard hyperscaler contract, the platform is operated by the vendor, administered by the vendor's global engineering teams, and reachable, in theory, by any foreign legal instrument that binds that vendor. A national partner cloud breaks that chain at the operational layer. Local staff run it. Local entities hold the contracts. The physical estate sits inside the jurisdiction. Access by the original platform owner is contractually and technically constrained.

That is worth having. It answers the immediate data-residency question, it satisfies a large share of procurement checklists, and it lets an agency point to a named domestic operator rather than a distant cloud console. If the concern is where the servers sit and who logs in day to day, the partner cloud model addresses it.

The ceiling nobody prints on the brochure

Here is what the operating agreement cannot change. The platform is still licensed. The code that defines how the cloud behaves, updates, patches, and evolves originates outside the operator. The operator runs a copy under terms. Those terms can be excellent and still be terms. A licence can be renewed, repriced, amended, or, in a genuine geopolitical rupture, allowed to lapse. The roadmap belongs to the licensor. The security model belongs to the licensor. When the platform changes, the sovereign operator adapts to a decision made elsewhere.

This is not a hypothetical worry for a procurement team pricing a fifteen-year commitment. Ask the practical questions. If the licensing relationship deteriorates, who controls continuity? If a future release removes a capability an agency depends on, who decides? If the platform's core is subject to an export control or a sanctions instrument in its country of origin, does the domestic operating agreement survive it? A partner cloud can guarantee that today's copy runs on home soil. It cannot guarantee that the source of tomorrow's copy stays willing, unchanged, and reachable. Operated under licence is a strong floor. It is also a ceiling.

Ownership is a different property, not a better tier

We built Mickai as a Sovereign Intelligence Operating System because we think the honest answer to that ceiling is ownership, not a better rung on the same ladder. When a regulated organisation runs Mickai, it holds the substrate. The system installs inside the organisation's own walls, runs air-gapped where the workload demands it, and writes a cryptographically-signed audit record on every action so the institution can prove, not merely assert, what its intelligence did and when. There is no external console the operator adapts to. There is no roadmap owned somewhere else. The institution is not a tenant of a sovereign-flavoured tenancy. It is the owner of the thing it runs.

!A gold figure of Hades enthroned in a marble vault against a black void, gold light in the depths

The defensibility behind this is filed, not aspirational. We hold 104 UK patent applications spanning roughly 2,340 claims across 13 families, named inventor Mickarle Wagstaff-Irons, building toward examination and grant. Those filings describe the mechanisms that make owned sovereign intelligence work in a regulated setting: the signed-audit record, the per-department reasoning fleet, the regulator-mode switching, and the continuity design that keeps the system running when an outside connection is pulled. This is the layer a licensed platform cannot offer by its own structure, because a licensed platform's core belongs to whoever wrote it.

We are not arguing against Bleu and Delos

We want to be careful here, because it would be cheap to frame national partner clouds as failures. They are not. For a great many public-sector workloads, a domestically operated hyperscaler stack is a proportionate, defensible, and cost-sensible choice. Almost every regulatory regime across the UK and EU permits cloud with the right controls in place. The genuine no-cloud line sits at specific workloads: classified material, isolated operational technology, export-controlled programmes, and cases where a data-protection assessment comes back negative. For those, and for any institution whose sovereignty requirement is a property right rather than a residency preference, an operating licence is the wrong instrument.

So the two models are not rivals so much as different tiers of the same question. If your requirement is where the data lives and who administers it, a partner cloud clears the bar. If your requirement is who owns the intelligence, who controls its future, and who can prove its behaviour without trusting an external party, ownership is the only structure that answers it. Bleu and Delos raise the floor. We think the ceiling is where the interesting procurement decisions actually get made.

The takeaway for procurement

Read the sovereignty claim in any bid as two separate questions. First, where does the data sit and who operates the platform. National partner clouds answer this well. Second, who owns the platform's future and who can be compelled to change it. That second question a licensed operator cannot answer in your favour, because the answer lives with the licensor. For most workloads the first question is enough. For your most sensitive, longest-lived, and least-substitutable systems, insist on the second. Ownership is not a premium version of a partner cloud. It is a different property, and for some workloads it is the only one that holds.

Frequently asked questions

Is a national partner cloud like Bleu or Delos genuinely sovereign?

It is sovereign at the operational layer. Domestic staff run it, domestic entities hold the contracts, and the estate sits inside the jurisdiction, which addresses data residency and day-to-day access. It is not sovereign at the platform layer, because the underlying cloud is licensed from an external owner whose roadmap, security model, and continuing availability sit outside the operator's control. For many workloads the operational answer is sufficient. For the most sensitive, it is not.

What is the difference between operating under licence and ownership?

A licensed operator runs a copy of someone else's platform under terms that can be renewed, repriced, amended, or, in a rupture, allowed to lapse. Ownership means the institution holds the substrate itself, controls its evolution, and depends on no external party to keep running. Mickai is built for ownership: it installs inside the organisation's own walls, runs air-gapped where required, and signs an audit record on every action.

Does this mean regulated bodies are barred from using cloud?

No, and we would not claim otherwise. Almost every UK and EU regime permits cloud with the right controls. The real no-cloud line is workload-specific: classified material, isolated operational technology, export-controlled work, or a case where a data-protection assessment comes back negative. The wider driver is preference for control, cost, and provable behaviour, not a blanket legal prohibition.

Where does Mickai fit alongside the hyperscalers?

As a complement, not a rival. The large AI vendors run the public frontier well. Mickai runs the owned, sovereign substrate for the regulated boundary those vendors' own terms exclude. An institution can keep both, using public services for open work and Mickai for the data and decisions it must own outright. Related reading: our pieces on the Sovereign Intelligence Operating System as a category, the signed-audit record as a compliance property, and why we position Mickai as an ally to the giants rather than a competitor.

!A gold figure of Apollo raising a lantern against a black void, gold embers rising toward clear light

Micky Irons is the founder of Mickai, the Sovereign Intelligence Operating System that regulated organisations own and run inside their own walls.