ComputerWeekly

How sovereign do we need to be? Data sovereignty as a spectrum


The US Cloud Act and FISA 702 give Washington the power to compel US companies to hand over data, wherever it sits. Meanwhile, Five Eyes intelligence sharing widens potential data snooping to US, UK, Canada, Australia and New Zealand collaboration.

So, what sovereignty guarantees remain when such mechanisms potentially allow cloud data to be accessed by law or bilateral agreement, and when a US provider can be compelled to write or push a patch into systems anywhere in the world?

That’s the question Google executives were asked at its Sovereign AI Summit in Brussels this week, where the company argued sovereignty is a spectrum and that customers should choose their point on it.

Jai Haridas, vice-president of regulated and sovereign cloud at Google Cloud, did not dodge the legal point. “We are a US company, and we have to follow the US laws,” he said.

But he pointed to architectural responses made possible in Google white-labelling of its technology to national cloud offerings. There aren’t many deployments to point at, but one is PREMI3NS – a sovereign cloud offering built on Google Cloud Dedicated technology and operated by S3NS – which received SecNumCloud 3.2 qualification from France’s National Agency for the Security of Information Systems (ANSSI) in 2025.

Under SecNumCloud, for example, “all updates go into quarantine where they are tested by the operator Thales, and they can actually stop an update”, Haridas said.

If there’s a spectrum, such measures are near the strong end of it.

At the summit, Google announced functionality that aims to make such a spectrum realistic for customers. These included a preview of confidential external (ie, customer) key management and audit frameworks, which extend the three-tier sovereign portfolio – data boundary, dedicated cloud and distributed cloud – it launched last year.

Meanwhile, analysts at Omdia are working on a five-level maturity model to help organisations work out where they sit:

  1. At level one, sovereignty is not a focus. We find reliance on third-party infrastructure, in unregulated firms with speed and cost unrestricted by sovereignty concerns.
  2. Level two offers improved visibility, but still no control, through public cloud offers that add privacy and guaranteed local residency features.
  3. Level three describes regulated industries that are still mostly reliant on cloud provider infrastructure but may need direct control over critical data and software via sovereign cloud solutions where law and regulation compel them.
  4. Level four gives public bodies with high geopolitical sensitivity – such as in the public sector – direct control over the full stack through a local partner that operates the same code base as the cloud provider, with extra-territorial access still in place.
  5. Level five is “very strong” with full control of data, software and infrastructure via national cloud providers or air-gapped solutions, and is reserved for organisations where the fundamentals of their activity would be at stake, such as defence organisations or Swiss banks.

Panel contributors at the event (pictured above) showed how such a spectrum accords with real-world practice.

Gertjan Dewaele, vice-president of product and technology at payment product company Worldline, described its dual setup – global payment workloads on public cloud, and a sovereign European stack for clients and projects such as the digital euro.

“Sovereignty for payments is not new,” he said. Regulation already pushes in this direction, such as under Dora, the EU’s Digital Operational Resilience Act, where financial firms must show exit plans and fully redundant systems.

Roman Kislyi from Ukraine’s Ministry of Digital Transformation described the far end of the spectrum. His ministry runs Gemini but strips out identifying data before it leaves the country, while keeping critical systems in distributed datacentres and a data embassy located abroad.

“Will your solution save the data of your citizens? Will your solution work if your datacentres are hit by a missile? Will your solution work if there is a blackout? Those are the questions we ask ourselves every day,” Kislyi said.

For manufacturers, the trade-off is different. Hamza Charrouf, chief customer officer at Pelico, said industrial customers want to keep operational knowledge out of models they do not control, and increasingly want innovation and sovereignty together rather than have to make a choice between them.

Panel members converged on a single point – the right level of control depends on what the data is for. “Take the emotion out of the domain, understand what you need, and then develop that,” said Robert Kernahan, vice-president and head of sovereign cloud for northern and central Europe at Capgemini.

“And when you break away from this one-size-fits-all idea, you go to optionality of choice, and you build that in,” he said.

Marcus Mühling, chief executive of Secure Cloud Germany, put the same point more bluntly. “If your clients have to make extreme choices – either sovereignty or innovation – that’s broken,” he said.

But cost is a key constraint. A partner-operated dedicated cloud carries a premium of 15% to 20% above standard public cloud rates, Google’s Haridas said. He described it as “a luxury product if you don’t need it”, but not if regulation requires it.

Sovereignty is clearly moving up the stack towards artificial intelligence, although that wasn’t a major topic of many conversations at the event.

But, said Mauro Capo, digital sovereignty lead EMEA at Accenture, clients now often ask who controls the models, the enterprise knowledge they draw on, and “the decisions and the actions that those models will increasingly influence”. Sovereignty, he argued, is shifting from infrastructure to data and knowledge.

For UK IT leaders, the debate may have taken place on the continent, but the spectrum of responses exists here too. The UK Ministry of Defence runs Google’s air-gapped distributed cloud variant, for example, and UK organisations that use US hyperscalers face the same Cloud Act question but without any EU-type security certification as a response.

The practical lesson is the one the panel kept returning to: Classify the data, decide the level of control each class needs, and build exit routes in from the start.



Source link