5 Comments
User's avatar
Melanie Goodman's avatar

This is such a clear-eyed snapshot of where cloud infrastructure is heading- and why. For years, the centralised model was gospel: big, efficient, far away. But the priorities have flipped. Now, it’s about where and who just as much as how much. Latency and sovereignty aren’t abstract concerns anymore; they’re deal-breakers. What’s fascinating is how deeply this shift reaches, right down to staffing and fibre routes.

Expand full comment
Laura Ferraz Baick's avatar

This is an excellent framing of what I'd call the "unbundling of the cloud monopoly";though it's worth questioning whether this distributed sovereignty model actually delivers on its promise or simply creates new points of failure and operational complexity.

Expand full comment
Neural Foundry's avatar

Fantastic analysis of how AI is forcing this architctural shift. The Indonesia example really clarifies why this isn't just theory. When you've got seventeen thousand islands and data residency mandates, you physically can't build a single centralized cloud and call it a day. That geographic constraint basically forces the layerd model you're describing, which honestly applies to most regulated markets now.

Expand full comment
Sam Illingworth's avatar

Thanks for this excellent post. Do you think there's a role for edge data centres in institutes like universities, for example, and that this is one way that they could potentially commodify existing infrastructure?

Expand full comment
Mark S. Carroll ✅'s avatar

This is a clear-eyed take on something too many discussions still separate. Sovereignty and latency are no longer competing priorities. AI makes them interdependent.

What I especially appreciated here is how grounded this is in physical reality. Power density, fiber routes, control planes, staffing. These aren’t abstract debates. They shape what can actually be built and operated.

The layered model you describe feels durable. Centralized regions for training, sovereign environments for governed workloads, and edge for inference and real-time systems. That framing maps cleanly to how AI is being deployed in practice, not how people wish it worked.

Thoughtful work. This is the kind of analysis that holds up beyond the news cycle.

Expand full comment