Preparing to take pole position on the AI grid
Summary
A video interview on how enterprises scaling AI deployments are straining traditional cloud infrastructure, and why an "AI Grid" model — distributing AI inference across edge locations rather than concentrating it in a handful of facilities — changes the economics of running AI in production for UK and international organisations.
The interview covers why the first wave of AI infrastructure — centralised GPU clusters built for training — doesn't fit the latency and geography requirements of inference at scale. John explains the AI Grid concept as CDN architecture applied to AI: distributing GPU compute across edge locations and routing each request to the nearest node that meets its latency, capability, and sovereignty constraints. The conversation also addresses the three-tier deployment model (edge, regional, core), the workloads where placement actually changes outcomes, and why the commercial advantage in AI infrastructure is shifting toward whoever already owns a distributed global footprint.
What I'd add
Eight minutes is enough to land an idea and not enough to defend it. Three objections come up every time I have this conversation in person, and none of them got airtime.
The first is the fair one: isn’t distributed inference just marketing from edge vendors? For a lot of workloads, honestly, yes — a batch pipeline or an internal copilot doesn’t care where it runs. But latency isn’t the only test. The other one is whether your current deployment is paying a fragmentation tax: every team standing up its own model instance, each with its own cold context and its own underused card.
The second is what “distributed footprint” actually means. That’s a claim about capital structure, not about scale. Edge locations take a decade and serious capital to build; adding GPUs to a network that already exists is an upgrade cycle.
The third is sovereignty, and it’s the one people get wrong in both directions. The law is more permissive than the folklore suggests, and most organisations’ own policies are stricter than the law requires. Both facts change where you can route a request — see the AI Grid page for how that lands in the routing decision.
I'm speaking on this — The Compute Infrastructure Questions Every AI Buyer Should Ask →
Looking for a speaker?
Keynotes, panels, and workshops on cloud, AI, edge, and security.