Summary
- The brand page sells sovereignty and prints no company name in the extract retrieved on 21 November 2026; the legal-notice path tested that day returned HTTP 404, and the notice that does exist belongs to another domain and names two separate companies at one address.
- AS61136 carries the as-name
Cloudeo-AS, but the aut-num object itself names no holder and no registration number; the registered name lives in a separate organisation object, ORG-ACIS2-RIPE, which printsOCI Cloud SASand its RCS number. - PeeringDB declares 50 IPv4 and 50 IPv6 prefixes for that ASN; RIPE's own announced-prefixes response for the fortnight to 21 November 2026 lists five IPv4 prefixes and no IPv6 prefix, and one of those five is registered to a different legal person from the ASN's own organisation.
The sentence that names a country
Start with the sentence the marketing rests on. The homepage reads: « Un cloud souverain pour maîtriser votre infrastructure, vos données et vos choix stratégiques » (Cloudéo homepage). Beneath it the same page reports a service hosted 100 percent in France, eight datacentres, four availability zones, a 99.9 percent service commitment, ISO 27001 and HDS certification, and nine service categories running from infrastructure and storage through digital workplace, security, colocation, cloud platforms, backup and disaster recovery, AI and GPU capacity, to managed services. All of it is self-reported, and none of it is the subject of this article.
The subject is what a sovereignty claim has to be, structurally, before a reader can check it. A claim about where machines sit is a claim about a legal person's facilities. A claim about who can be compelled to answer for a route is a claim about a legal person's registration. Sovereignty as a property rather than an adjective is a chain: brand to company, company to registry identity, registry identity to address space, address space to the routes actually announced. Each link is a document, and each document either prints a name a reader can verify or it does not.
On the homepage the first link is missing from the material I have. The retrieved extract contains no company name, no SIREN or RCS number, no autonomous system number and no prefix list. That is a reading limit as much as a finding: the extract was truncated and the footer was not captured, so I cannot tell you the page carries no legal notice anywhere. I can tell you what the obvious place returned. The address https://www.cloudeo.fr/mentions-legales/ (the tested legal-notice path) answered HTTP 404 on 21 November 2026. One path, one date, and no legal-notice text to quote.
The group domain does better, and in doing better it lengthens the chain. The legal notice on oci.fr (Groupe OCI legal notice) states that the site is edited by « La société GROUPE OCI SAS, au capital social de 613 500 €, immatriculée au RCS de Strasbourg sous le numéro 502 770 118 », with its seat at 2 rue Ampère, 67450 Mundolsheim, and names M. Frédéric VACHER as publication director. The same notice states, separately, that the site is hosted by OCI Cloud SAS at that same address. Two legal persons, one address, one of them hosting the other's website. And this notice is for oci.fr: it does not mention Cloudéo, and it does not mention AS61136. A reader who starts at the brand and wants a company has to cross a domain boundary to find one, and arrives at two.
The brand's owner calls it a brand
The group is not coy about the relationship. A Groupe OCI recruitment page (OCI recruitment page) describes Cloudéo as the sovereign-cloud brand of Groupe OCI, and describes the platform as three availability zones inside seven datacentres, staffed by more than forty infrastructure, systems, network, security and DevOps engineers, certified to ISO 27001 and HDS, and working toward SecNumCloud. A Groupe OCI interview published on 21 July 2025 (OCI interview with Sébastien Tinetti) presents Sébastien Tinetti as director of Cloudéo within Groupe OCI and reports an intention to apply for SecNumCloud qualification. An intention to apply is not a qualification, and no document I read records one.
Note the estate arithmetic the group publishes about itself, because it does not close. The homepage says eight datacentres and four availability zones. The recruitment page says seven datacentres and three availability zones. The directory record I reach below lists seven interconnection facilities. Three counts, three different objects — self-reported datacentres, self-reported datacentres, and interconnection facilities — with no document reconciling them. I am not going to resolve that by subtraction.
It is three statements about different things, and the honest reading is that the group's own pages describe its estate at two different sizes without saying which is current.
Where the network identity actually lives
Now the registry, which is where an infrastructure claim becomes checkable. The RIPE Database aut-num object for AS61136 (RIPE REST aut-num AS61136) prints the as-name Cloudeo-AS, references organisation ORG-ACIS2-RIPE, carries status ASSIGNED, was created on 4 February 2013 at 11:02:55 UTC, was last modified on 8 July 2026 at 12:49:33 UTC, is maintained by RIPE-NCC-END-MNT and fr-alsace-1-mnt, and declares an import policy accepting any route from AS29075, AS174 and AS48213, and an export policy announcing AS61136 to those same three.
Read that object twice and notice what it never does. It never names a holder, and it never prints a registration number. The as-name Cloudeo-AS is a label a network gives itself. The org attribute is a pointer — a handle, ORG-ACIS2-RIPE — and a handle is not a company. This is the ordinary architecture of the RIPE Database rather than a peculiarity of this record: the identifying name lives in a second object, and the second object is the one that carries the legal person.
That second object (RIPE REST organisation ORG-ACIS2-RIPE) prints org-name OCI Cloud SAS, country FR, registered number 908 563 307 R.C.S. Strasbourg, org-type LIR, an address at 2 Rue Ampere, 67450 Mundolsheim, a creation date of 29 March 2018 at 08:04:24 UTC, and a last modification of 13 May 2026 at 07:38:58 UTC.
So the chain reads: brand Cloudéo, as-name Cloudeo-AS, handle ORG-ACIS2-RIPE, legal person OCI Cloud SAS. Four strings for one network, and only the last is a company with a number a reader can look up in a commercial register. The handle keeps a stem, ACIS, that appears nowhere in the registered name it now prints. Two dates are on the record: the aut-num was created in February 2013, the organisation object in March 2018, five years and two months apart by my subtraction. What the registry does not print is any history of what that handle pointed to at creation.
The objects say what they say, and the rest is inference I am declining to make.
Three prefixes, three kinds of answer
The address space is where the chain either continues or stops, and it does both depending on which prefix you ask about. I read registry objects for three of the five observed prefixes; for 193.104.154.0/24 and 185.253.134.0/23 I did not, and I am not going to characterise what I did not read.
Take 91.216.19.0/24 (RIPE search for 91.216.19.0). The inetnum prints netname OCI-Connect, status ASSIGNED PI, organisation ORG-OA325-RIPE, and a sponsoring-org of ORG-ACIS2-RIPE; the route object for that prefix lists origin AS61136. The organisation object named there prints org-name Groupe OCI SAS and registered number 502 770 118 R.CS Strasbourg. Which means the ASN's own organisation sponsors a provider-independent assignment held by a different legal person — the holding company whose number appears on the group legal notice — and that company's space is announced through the ASN. Same address, same group, two registrants. If you want to know who is accountable for that /24, the answer is a different company from the one the AS object points at.
Take 185.17.240.0 (RIPE search for 185.17.240.0). The search returned an inetnum covering 185.17.240.0 to 185.17.240.127, netname Groupe_OCI-Network, status ASSIGNED PA, maintained by fr-alsace-1-mnt — and no org attribute at all. That is a /25 sitting inside a routed /22, covering one eighth of it by my arithmetic, and the object prints a nickname and a maintenance pair instead of a legal person.
Take 185.253.132.0 (RIPE search for 185.253.132.0). The inetnum covers 185.253.132.0 to 185.253.133.255, netname Cloudeo-SXB-Network, status ASSIGNED PA, maintained by fr-alsace-1-mnt, no org attribute. A nickname that carries the brand, and nobody named.
Two of the three prefix objects I read therefore do not identify a registrant. That is not an irregularity in itself: provider-aggregate assignments routinely inherit identity from the aggregate they sit inside. It matters here because the layer a reader would audit for a jurisdictional claim is exactly the layer that prints Cloudeo-SXB-Network and stops. The one string that recurs at every layer of this chain is fr-alsace-1-mnt, which appears on the aut-num and on the prefixes alike. A maintenance object is a set of keys and a mailbox, not a registrant. It is the thread that runs through the whole chain, and it is not a person.
Fifty declared, five observed
The directory record is the layer most people actually read, and it is the layer where this network's description and this network's behaviour can be put side by side.
PeeringDB network record 37001 (PeeringDB API network 37001) prints name Cloudéo, also-known-as Cloudéo, ASN 61136, organisation 38934 named Groupe OCI with a website at oci.fr and an address in Mundolsheim, network website cloudeo.fr, IRR as-set RIPE::AS-CLOUDEO-AS, network type Enterprise, a traffic band of 10-20Gbps, traffic ratio Balanced, scope Europe, an open peering policy with locations and contracts not required, zero internet exchanges, and seven facilities. Those seven entries are the most concrete estate in the whole corpus: three at ETIX Nantes, one at nLighten Lyon, one at nLighten Strasbourg, one at UltraEdge Lyon-Venissieux, one at UltraEdge Strasbourg. The record was created on 7 August 2024, last updated on 10 October 2025 at 06:20:19 UTC, and carries a RIR status of ok dated 8 August 2024. Its metadata object is empty.
And it declares fifty IPv4 prefixes and fifty IPv6 prefixes. These are declaration fields filled in by the network. No document I read defines what they count, whether they mean allocated, announced, planned or something else, or who checks them. The traffic band above them is a declaration on the same footing, and there is no ratio to be had between a declared bit rate and a counted prefix: the two are different units filled in by the same hand.
Against that, RIPE's own observation. RIPEstat's announced-prefixes response for AS61136 (RIPEstat announced prefixes for AS61136), for the window from 7 November 2026 at 08:00 to 21 November 2026 at 08:00, with a latest_time of 21 November at 08:00 and a status of ok, lists exactly five IPv4 prefixes: 91.216.19.0/24, 193.104.154.0/24, 185.17.240.0/22, 185.253.134.0/23 and 185.253.132.0/23. Zero IPv6 prefixes. The response states that its results exclude routes with very low visibility, specifically those seen by fewer than ten RIS full-feed peers, so this is a filtered observation over a stated fortnight rather than a census of every path in existence.
The arithmetic is mine, not the company's, and it is worth stating flatly. Fifty declared IPv4 prefixes against five observed is a factor of ten. Fifty declared IPv6 prefixes against zero observed is not a ratio at all; it is a sentence. The five observed prefixes carry 2,560 IPv4 addresses — 256 + 256 + 1,024 + 512 + 512 — of which the single /22 carries 1,024, or 40 percent. The /25 inetnum returned for 185.17.240.0 covers 128 addresses, 12.5 percent of the /22 it sits inside. And the directory record was 407 days old when I read it, measured from its own last-updated field.
One more document, because it is the only place in the corpus where the brand string, the registered company name and the route meet in a single field. RIPEstat's prefix-overview for 185.17.240.0/22 (RIPEstat prefix overview, 185.17.240.0/22) returns announced true, is_less_specific false, an empty related-prefix list, and an origin list containing AS61136 with the holder string Cloudeo-AS OCI Cloud SAS. The ALLOCATED label in the same response belongs to the covering 185.0.0.0/8 IANA block, not to the /22 — worth saying, because that label is the one most easily misread as a statement about this prefix. And note what the holder string is: a label built on the observation side out of registry data. It joins the brand's as-name to the registered company name, which is precisely the join the registry's own objects decline to make in one place. It inherits the split rather than resolving it.
Where the documents agree
It would be easy, and wrong, to read all this as a record that cannot be trusted. Check the parts that can be checked.
The directory's ASN matches the registry's ASN. The directory's organisation name, Groupe OCI, corresponds to a legal person the registry records at the same Mundolsheim address under number 502 770 118, the number printed on the group legal notice. The aut-num's declared policy names exactly three external ASNs, and the network's peering posture is open with no contract requirement, which is the same stance described two different ways by two different documents. The seven facilities in the directory are named, specific, and dated by the record's creation.
So the failing field is the prefix count, not the record. The structure of that failure is the instructive part: a directory field that presents itself as size was filled in by the network, while the size a reader can verify was produced by route collectors the network does not control. Where a claim and an observation can be compared, the comparison is the finding. Here the comparison is a factor of ten on IPv4 and a declared fifty against an observed zero on IPv6.
What an auditable sovereignty would need
Put the layers in a column and ask each one for a name.
The brand page names a country. The group legal notice names two companies at one address, for a different domain. The aut-num names nobody. The organisation object names OCI Cloud SAS, with a registration number. One prefix names Groupe OCI SAS as holder and OCI Cloud SAS as sponsor. Two prefixes name a nickname and a maintenance handle. The directory names a group, an ASN, seven facilities and fifty prefixes. The routing observation names five prefixes and no person at all.
A registry handle is a pointer, not an accountability document; the accountability document is the legal person printed in the same object, which is why the organisation object carries more weight here than the aut-num, and why the two org-less inetnums matter more than they look. A self-declared directory field is a claim and the routing system is an observation, and where the two can be compared the comparison is the finding. Judged by that standard, this corpus neither proves nor disproves that Cloudéo's machines are in France.
What it shows is that the documentary chain a customer would walk to confirm it has two gaps and one cross-link, and that the cross-link runs through a domain the brand does not own.
That is not a small omission in a market where the word sovereign is doing commercial work. A public buyer selecting a sovereign provider is being asked to accept a jurisdictional claim, and jurisdiction lives in a company registration rather than in a datacentre address. This corpus gives that buyer one registration number, on a notice for the wrong domain, for a holding company, alongside a second company named only as a host — and no document that attaches either of them to the brand on the brand's own site.
The questions to put in writing
If you are buying, the useful output of this reading is not scepticism. It is a list.
Ask which legal person signs the contract, and whether that is OCI Cloud SAS, Groupe OCI SAS, or a third entity. Ask which legal person holds 185.17.240.0/22 and the two /23s, since the registry objects for them print no org attribute. Ask what the directory's two fifty-prefix fields were intended to count and when they will next be updated, given that the record's last update predates the observation by 407 days. Ask whether IPv6 is announced at all, given that RIPE's collectors saw none in the fortnight to 21 November 2026, and what that means for a buyer whose own estate is dual-stack. Ask who controls the maintenance handle fr-alsace-1-mnt and who can add an org attribute to a prefix object. Ask for the SecNumCloud timeline in writing, since the group's own pages describe an intention to apply and work in progress rather than a qualification.
None of those questions requires the vendor to concede anything. Each of them produces a document with a name and a date on it, and a document with a name and a date is what an auditable sovereignty claim is made of.
What this article does not establish
Four limits belong beside the claims they qualify.
The homepage extract was truncated and its footer was not captured, so the absence of a legal person from that extract is a reading limit, not a finding that the page carries none. The legal-notice address was tested once, on one date, and returned 404; that result belongs to that address on that date and says nothing about any other page on the site. RIPEstat's list excludes routes seen by fewer than ten full-feed peers, so a tenfold gap cannot be proved complete from it: the true announcement count could be five, or it could be twenty and still short of fifty.
The stem ACIS in the organisation handle does not appear in the registered name, and the two object dates are five years and two months apart — a predecessor company is an inference the objects do not support, and I have not made it. The three datacentre and facility counts in the corpus are three different objects and are not directly comparable.
What the corpus does establish is narrower and steadier. The network identity behind this brand resolves through a handle to a named company. The objects holding most of its address space do not name that company. And the number most readily quoted as this network's size is a self-declared field that sits a factor of ten away from what the routing system observed over a stated fortnight.
There is a page for this entity in the BTW directory, worth opening beside the sources: Cloudéo in the BTW directory. It records a cloud-service platform, a global region, and a data gap of no ASNs — which is a small illustration of the whole problem. The ASN exists, and it has existed since 2013. What the record lacked was a document connecting the brand to it, and after four research passes the connection still travels through a registry handle, a legal notice on another domain, and one derived holder label.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






