There are three things a nation must control to project power in the modern world: its borders, its resources, and its data.
The first two are centuries-old struggles, well-suited to diplomacy and guns. The third is newer, slipperier, harder to detect when breached—and even harder to enforce without gutting the very openness that made AI possible in the first place.
At NVIDIA GTC 2025, we gathered for a lunch panel to discuss what happens when the ideal of global-scale AI runs headlong into the non-negotiables of sovereignty, security, and compliance.
This was not a theoretical session. It was a rare, unvarnished glimpse into the architecture of sovereign intent, as engineered by Core42 and made real by VAST Data.
The heart of the chat was how to build AI that respects borders, obeys laws, resists espionage, and still works in real time.
The First Truth: Data Does Not Like to Stay Still
Core42, the UAE’s sovereign AI platform builder, began with a grand, almost audacious premise: to construct an AI intelligence grid that would wrap the globe like a second nervous system—data in, insight out. Powering governments. Enabling digital sovereignty. Delivering intelligence at the scale of entire populations.
But that dream hits its first immovable object almost immediately: the very nature of AI demands data mobility, while sovereignty demands data locality.
You cannot both silo and scale.
Or at least, you couldn’t—until someone redefined what a data platform even was.
Chris Morgan from VAST put it in terms more elegant than brute force: traditional storage was built for a world where data went to rest, where the workflow ended in tape or archive.
But in AI, data never rests. It’s always on the move—query, fine-tune, infer, update, iterate. Every cycle demands access. Every access demands policy enforcement. Every policy must travel with the data, or sovereignty is broken by latency.
And latency, in this case, is not just time. It’s risk. It’s exposure. It’s, quite literally, a violation.
The Architecture of Trust
Core42 didn’t start by buying racks of GPUs. They started by defining the trust model.
What do nations need? Control.
What do regulators require? Visibility.
What does AI require? Velocity.
What does sovereignty demand? That none of these be compromised.
And so they built what Core42’s Raghu Chakravarthi described for the audience as landing zones—fully sovereign enclaves of compute, storage, and security—bound by encryption, ringed with policy, and connected directly to the oversight infrastructure of each participating nation.
These are not “clouds.” They are trust containers.
Inside each is a blueprint for inference: not just storage and compute, but metadata layers that enforce who can see what, where, and when.
When Raghu spoke about exporting his WhatsApp thread to ChatGPT and getting roasted by his friends, he was smiling—but the lesson was deadly serious. If you can’t keep your group chat sovereign, how will you manage a national biometric database?
This is why Core42 partnered with VAST. Not for mere storage as the market would have defined it, but for a namespace that carries law.
The Passport Layer
In VAST’s architecture, data doesn’t just sit. It moves, and when it does, it brings its papers.
Each data object carries with it the policy by which it was created, the audit trail of who touched it, and the encryption boundaries that travel with it. This is metadata as governance, and it’s the missing layer in every storage system built before.
Chris described it—without saying the word—as a kind of vectorized citizenship. Data becomes self-identifying. It does not need to be managed; it declares its own terms of use.
This allows Core42 to do something unprecedented: take sensitive data from Kenya or India, keep it encrypted, run inference on it inside UAE-based secure clouds, and return results without ever breaking policy containment. It’s sovereignty-as-a-service. And it works.
The War Against Fragmentation
Most enterprise stacks today resemble abandoned subway stations: layers of dead protocols, silos built for workloads that haven’t run in years, pipelines of duct tape and PDFs.
Now imagine trying to run a multimodal AI inference engine on that. You need fast object storage. Also file. Also block. Also a vector store. Also something Kafka can hit in sub-200ms.
Oh, and it better be air-gapped and regulator-approved and available across three continents.
This is where the old world fails.
What Core42 is building—with VAST underneath—is a unified platform, where all these abstractions converge in a single namespace. Object, block, file—they’re not separate products. They’re views on the same data. Same policy. Same auditability.
The result is a system where data can live in one place, be accessed from another, inferred upon in a third, and still remain, legally and cryptographically, under control.
That’s the trick. You don’t scale by replicating. You scale by abstracting control across location.
Watch the full video here:
