- Sovereign infrastructure promises safety—until local-only becomes a single point of failure
- A Middle East flashpoint reveals how geopolitics can reach a data center fast
- CIOs chase sovereignty, but a hidden legal web could turn compliance into risk
The whole point of sovereign cloud and AI infrastructure is to protect data, whether to comply with regulatory demands or simply to preserve a competitive edge. But it turns out sovereignty actually introduces new risks of its own.
These risks seem to fall into two main categories: physical risks tied to infrastructure and business risks related to a complicated and ever-changing global regulatory landscape.
On the physical front, it turns out storing data in an isolated asset can actually be a vulnerability as much as it is a safeguard.
The war in Iran presents an interesting case study in this area. Over the past year or so, the Middle East has emerged as not only a rapidly growing data center and AI hub but also a region at the forefront of sovereignty efforts. As we’ve noted before, AWS Microsoft and Google have all inked deals to invest heavily in sovereign infrastructure in the region.
It’s exactly this confluence of hefty investment and a focus on sovereignty that makes what’s now happening in the Middle East a cautionary tale.
The downside of physical isolation
In response to U.S. and Israeli attacks, Iran over the past month has bombed AWS data centers in two neighboring countries. It has also directly threatened to level OpenAI’s Stargate campus in the United Arab Emirates and continue targeting assets of U.S. technology companies.
All this is to say, folks running sovereign workloads in the region are likely stressed (for many reasons, let’s be honest). From a tech perspective, the situation exposes the inherent vulnerability of an isolationist approach to the cloud, AI and data. If you’re operating in a region where there’s one primary facility providing sovereign services, that’s a risk.
It’s not just bombs that could take your operations offline, but a regional blackout, an accidental deletion, a cyberattack or one of a million other scenarios.
“Running everything everywhere used to be a resiliency play,” Megaport CEO Michael Reid told Fierce. “If you trust the cloud to do everything for you and there’s an issue – what’s occurred in the Middle East or an accidental deletion – you actually want some sort of resilient backup outside of the cloud or outside of that environment.”
But this is where things get sticky. Data protection rules in some areas – Germany, for instance – are stringent. So, while you might want a full cloud backup stored outside of the country, you may not be allowed. The question then becomes whether you spring for a backup outside the cloud.
It’s true that sovereign AI doesn’t necessarily need to sit in a sovereign cloud, but the same ideas apply: you likely want a backup somewhere, but have to be very careful about how and where it is located.
Navigating the regulatory web
Ron Babin, a professor at Toronto Metropolitan University and adjunct research advisor at IDC, told Fierce that a recent survey of 134 CIOs conducted by the analyst firm found 100% are dedicating time and resources to sovereign AI. But while most are viewing it as a security initiative, Babin said it’s actually much broader than that. Not only do CIOs need to be familiar with the technology, but they also need to be keenly aware of geopolitical environments.
For instance, they need to understand how their AI usage may be working contrary to laws in other jurisdictions and what happens when a company using a sovereign AI model in one jurisdiction (either a state or country) is subpoenaed by another.
“CIOs cannot ignore this and executive teams must acknowledge that they’re operating in a different world,” Babin said. “Everyone must use AI – there’s no doubt about that – but where do you start to hit the boundaries of legal and regulatory requirements that different countries, different jurisdictions, different states will have as you roll out your AI model.”
“If you don’t know what the compliance requirements are, you might be taking on risks that are very dangerous for your organization,” he warned.
Babin said a foundational assessment by CIOs should include understanding not only where their home base of operations is, but where they sell their services, where the supply chain is located, what the laws are in those areas and what industry-specific regulations might apply. The latter is particularly relevant for telcos, financial services, healthcare and other sensitive industries. CIOs also need to have a firm grasp on how AI is being applied within their organization and why.
Babin argued in favor of a hybrid approach to sovereign AI, one that mixes global public cloud infrastructure (sovereign or otherwise) and models with local options. The idea is to reap the benefits of the economies of scale that companies like AWS, Google and Microsoft have while preserving the ability to run jurisdictionally isolated workloads and models.
The downside, he acknowledged, is that doing this is neither cheap nor easy.
But the alternative is being fined or potentially losing control over critical data. He pointed to what happened to consumer DNA test provider 23andMe as an example. The company eventually went bankrupt and ended up selling its massive genetic database, creating a huge data vulnerability for anyone who had used the service.
“No company lives forever,” Babin noted. “The risk of you providing detailed and confidential information to a large AI provider, and then for whatever reason they’re hacked, they go bankrupt, someone buys them, they’re bombed – all of a sudden, you’re vulnerable to somebody that you relied on...you need to have a plan, and this is where the hybrid architecture comes in."