
On January 15, 2026, AWS [launched the European Sovereign Cloud](https://press.aboutamazon.com/aws/2026/1/aws-launches-aws-european-sovereign-cloud-and-announces-expansion-across-europe): a separate cloud, not a separate region, with its first location in Brandenburg, Germany and more than 7.8 billion euros of committed investment behind it. Six months of general availability is enough time for the marketing dust to settle and for the architecture to be examined on its merits. Both deserve it.

Here's my position up front, because sovereignty posts tend to hide theirs: this is the most serious sovereignty engineering any hyperscaler has shipped, it fixes real problems that "EU region" checkboxes never touched, and the one question that will decide whether it means anything is legal rather than technical. Nobody can answer that question yet, including AWS.

## What AWS actually built

Credit first, and specifics, because the engineering is genuinely unusual. The European Sovereign Cloud isn't a fenced-off corner of the existing AWS; it's a [separate partition](https://www.infoq.com/news/2026/01/aws-european-sovereign-cloud/), which in AWS vocabulary means a different universe with no shared control plane.

| Separation dimension | What the ESC does |
|----------------------|-------------------|
| Control plane | Independent partition; no dependency on US-based AWS regions to operate |
| Identity | Its own IAM, separate from global AWS accounts |
| Billing | Independent billing systems inside the EU |
| DNS and certificates | Separate DNS infrastructure and its own certificate authority |
| Operations staff | EU residents only, employed by EU entities |
| Corporate structure | A new German parent company with three subsidiaries incorporated under German law |

Compare that with what the [region dropdown](/sovereignty/sovereignty-is-not-a-region-dropdown) gives you and the difference is a category, and it answers real operational questions. Who can touch the hardware? EU residents. Where does the control plane live? Germany. Can a global AWS outage take it down? By design, no; the partition stands alone, which quietly also makes it an interesting resilience story after [last autumn's outage cluster](/multi-cloud/multi-cloud-overrated-portability-isnt).

## The price of the moat

Isolation costs, and the costs are visible in three currencies. Services: the ESC launched with an [initial set of around 90 services](https://aws.amazon.com/blogs/security/announcing-initial-services-available-in-the-aws-european-sovereign-cloud-backed-by-the-full-power-of-aws/) against the 240-plus of commercial AWS, so teams accustomed to reaching for any managed service will find gaps, and gap-filling is exactly the work the cloud was supposed to remove. Money: pricing runs at roughly a 15% premium over the commercial regions, which is the bill for duplicated everything. Geography: one region in Brandenburg at launch, with EU-wide expansion announced rather than delivered.

None of this is a gotcha. A partition with its own certificate authority can't cost the same as a shared one, and anyone selling you isolation for free is not selling isolation. The premium is honest. What it does mean is that the ESC is a deliberate purchase for workloads that need it, not a default, and AWS prices it accordingly.

## The question a GmbH can't answer

Now the hard part. The ESC's operating companies are incorporated under German law, and its marketing leans on that structure. The [unresolved question](https://www.infoq.com/news/2026/01/aws-european-sovereign-cloud/) is whether any subsidiary of a US parent, however carefully incorporated, sits beyond the reach of US legal process aimed at the parent, because the CLOUD Act's obligation attaches to what a provider can be compelled to control, and control is a corporate-law question rather than a datacenter one. We walked through that statutory text [in the previous post](/sovereignty/sovereignty-is-not-a-region-dropdown); nothing in a partition diagram amends it.

To be fair about the mechanics: the structure raises the practical and legal cost of any such demand enormously. EU-only operators can't be quietly ordered around, German directors face German law, and a US court order against the parent would collide with EU law in ways that guarantee years of litigation. That's real friction. Friction isn't immunity, though, and the difference between the two is precisely what a sovereignty buyer is paying to not worry about. The honest statement of where things stand: untested. No court has ruled on this structure, and until one does, every confident answer in either direction is a prediction wearing a suit.

## The steelman: unfalsifiable criticism is cheap

There's a fair objection to pieces like this one: if no corporate structure can satisfy the critics short of full divestiture, the criticism is unfalsifiable, and unfalsifiable criticism is a rhetorical trick rather than an argument. For most European organizations' actual threat models, EU-resident operators plus a separate partition plus German entities is protection beyond anything previously available from a hyperscaler, and treating the residual legal question as disqualifying lets the perfect strangle the good. There's truth in that. Plenty of buyers should take the ESC and sleep fine.

But notice who the ESC is for. Its price premium and its marketing aim at exactly the buyers whose threat model is the residual legal question: ministries, defense-adjacent industry, critical infrastructure, the customers European law is starting to sort into sovereignty tiers. For that audience, "dramatically more friction, ultimately untested" is the honest label, and the tier frameworks now moving through Brussels will decide formally whether a US-parented structure can reach the upper shelves. That's not my call to make. It will get made, in regulation and eventually in a courtroom, and the ESC's commercial fate rides on it.

Our stake in this is small and simple to declare: we're a European vendor, sovereignty tailwinds help us, discount accordingly. What we'd say to anyone evaluating the ESC is the same thing we say about every platform, ours included: the exit is the part of sovereignty you control unilaterally, so whatever tier of cloud you buy, keep your workloads portable enough to leave it. AWS built a remarkable moat in Brandenburg. Before you move inside, ask yourself the question the GmbH can't answer for you: if the legal prediction goes the wrong way in three years, how long would leaving take?

---

Related: [Sovereignty Is Not a Region Dropdown](/sovereignty/sovereignty-is-not-a-region-dropdown), the legal groundwork under this post.
More about what we're building at [light-cloud.com](https://light-cloud.com).


Real engineering, honest gaps, and one question a GmbH can't answer

AWS European Sovereign Cloud, Six Months In


There's a dropdown in every cloud console that radiates more false comfort per pixel than any other UI element in software. Region: eu-central-1. Frankfurt. Select it, deploy, and a whole compliance narrative writes itself: the data is in Germany, Germany is in the EU, therefore we're covered. Meetings end early on the strength of that dropdown.

Here's the claim this post defends: where your data sits and who can be compelled to hand it over are two different questions, the dropdown answers only the first, and an expanding genre of sovereignty marketing depends on you not noticing the difference.

## Jurisdiction follows the company

In 2018, the United States passed the CLOUD Act, and its core provision is worth reading in the original because no summary improves on it:

<figure class="research-quote">
  <blockquote>...regardless of whether such communication, record, or other information is located within or outside of the United States.</blockquote>
  <figcaption>18 U.S.C. 2713, added by the CLOUD Act, 2018. <a href="https://www.congress.gov/crs-product/R45173">Congressional Research Service overview</a></figcaption>
</figure>

The obligation attaches to the provider, wherever the provider's US legal entity can reach: data within its "possession, custody, or control" is in scope, Frankfurt datacenter or not. Whether an EU subsidiary of a US parent falls outside that reach is precisely the disputed question, and the honest answer is that it hasn't been cleanly resolved; corporate structure, not server placement, is where the argument happens. Your bytes sleep in Germany. The subpoena lands in Washington.

To be precise about what the law is and isn't: it establishes legal process, with warrants, review, and a mechanism for providers to challenge conflicts with foreign law. It's not a secret free-for-all, and describing it as one is its own kind of washing. But the process runs through US courts, applying US standards, to data your German neighbors assumed was governed by German ones.

## Two legal systems, one server

The collision isn't theoretical; it has a case name. On July 16, 2020, the EU's top court decided [Schrems II](https://www.europarl.europa.eu/RegData/etudes/ATAG/2020/652073/EPRS_ATA(2020)652073_EN.pdf), striking down the Privacy Shield transfer framework because US surveillance law didn't offer EU citizens protection the court considered essentially equivalent to EU law. A successor adequacy framework has since arrived, and few people in Brussels would bet on it being the last word, given that its predecessor's predecessor died the same way.

Notice what that whole line of cases is about. Not datacenter locations. Legal orders: which country's authorities can reach the data, under which safeguards, reviewable by whose courts. The entire dispute would look identical if every byte involved had spent its life in Frankfurt, because the question was always jurisdiction over the companies holding the keys.

| The region dropdown controls | It does not control |
|------------------------------|---------------------|
| Latency to your users | Which governments can compel disclosure |
| Where data rests physically | Where the operator is legally answerable |
| Which local failure domain you're in | Who employs the people with admin access |
| A data-residency checkbox | Whether encryption keys leave the jurisdiction |

## How to spot sovereignty-washing

"Sovereign" is becoming a product adjective, and some of what it decorates is genuine engineering while some is a region with a flag sticker. The test is a short list of questions that marketing pages tend not to answer. Who owns the operating entity, all the way up? Which courts can bind that owner? Who holds the encryption keys, and can the operator's parent be compelled to use them? And the question that gets skipped most: if the answers change, how fast can you leave?

Key custody deserves its own line, because it's where the washing gets technical. Hold-your-own-key arrangements genuinely narrow what an operator can be compelled to produce, but only if the operator truly can't reach the keys; a key management service run by the same corporate family is a promise, not a boundary. Ask where the key material physically lives and who can rotate it. The answer sorts real designs from brochures quickly.

That last question on the list matters most because exit is the sovereignty mechanism you control unilaterally. Every other protection on the list is something a vendor or a legislature grants you and can amend. The ability to take your workloads and data elsewhere on short notice is the one that doesn't require anyone's permission, which is why we treat portability as a sovereignty feature and not a procurement nicety. It's the same argument we made about [multi-cloud](/multi-cloud/multi-cloud-overrated-portability-isnt), pointed at a different threat.

## The steelman: residency isn't nothing

The dropdown's defenders have real points. Data residency solves genuine problems: latency, local failure domains, and a handful of sector rules that literally require data to remain in-country. Residency also narrows practical exposure, since data that never leaves a jurisdiction at rest is harder to sweep up incidentally. And for a lot of businesses, a pragmatic read is fine; not everyone's threat model includes foreign legal process, and pretending otherwise is its own theater.

Conceded, all of it. Residency is a floor. The error is selling the floor as the house: compliance narratives built on the dropdown quietly assume location implies jurisdiction, and both the CLOUD Act's text and a decade of transfer-framework litigation say it doesn't. A law can say so in one sentence. Marketing needs you to not read that sentence.

We're a European company, and I'd love to tell you that alone settles something. It doesn't; jurisdiction applies to us the same way, which is rather the point. What we can do is build so the exit door stays open, and describe the legal terrain plainly instead of decorating it. So, about your own stack: do you know, today, whose courts can reach it? Not where it runs. Who can reach it.

---

Related: [Multi-Cloud Is Overrated. Portability Isn't.](/multi-cloud/multi-cloud-overrated-portability-isnt), the operational half of this argument.
More about what we're building at [light-cloud.com](https://light-cloud.com).
