Yesterday I sat in on a session run by STL Partners and Intel — How generative, agentic and sovereign AI are reshaping telco growth in MENA. I’m not going to summarise it. It got me thinking, and one observation in particular has stuck with me since.

It was STL’s, not mine. They said they’d been genuinely surprised by how fast telcos have moved on AI — and not in one region. Across both Europe and the Middle East. Specifically, by how quickly operators have gone from using AI to cut their own costs to selling it as a top-line enterprise service: the leap from internal efficiency to revenue that the industry has talked about for years and rarely pulls off.

That’s worth sitting with, because it quietly kills the lazy version of this story. The lazy version says the Gulf “gets it” and Europe doesn’t. STL’s read says the opposite — the appetite, and the commercial pivot, moved fast more or less everywhere. Telcos on both sides woke up to the same opportunity at roughly the same time.

So if intent isn’t the differentiator — and STL are telling us it isn’t — then what is?

The answer is the boring, physical question of whether you can actually act on the intent at scale. And that’s where Europe and the Gulf stop looking alike. Building sovereign AI capacity takes capital and it takes power, and those two things are handed out very unequally across the map. Telcos everywhere can move fast on strategy. They can only move fast on deployment where the capital, the mandate and the megawatts are already on the table.

1. The demand is real

Start with the part that isn’t in dispute. STL mentioned a survey on the call: ask enterprises which partner they’d turn to for AI implementation, and in MENA around half name their telco — against roughly a quarter elsewhere in the world. That gap is the whole regional story in one statistic. A remarkable vote of confidence for an industry that spent twenty years being told it was a dumb pipe — and a sign that here, the demand to be a trusted, in-jurisdiction AI provider is pointed straight at operators.

e& put up the clearest picture of it — a sovereign-AI enablement spectrum running from in-country infrastructure, through inference-as-a-service, to “SLM-in-a-box” at up to 80% lower cost, out to industry copilots for government and regulated sectors. Abu Dhabi’s 19MW unified government data centre, serving 40-plus public entities, sits underneath it as the anchor.

What struck me most was where STL placed the telco opportunity in the value chain: not in raw global compute, but in distributed AI infrastructure and the connectivity, orchestration and control layer above it. That’s the bit I keep coming back to. The money isn’t only in owning the boxes — it’s in orchestrating where intelligence runs, who governs it and how it gets billed. The control plane. That’s the layer telcos are uniquely placed to own, and it’s the one the Gulf operators are already reaching for.

2. Why the Gulf can act fast

Three forcing functions sit under the speed, plus an uncomfortable fourth.

The first is capital — sovereign, patient and abundant. The numbers Intel put up give you the scale of the bet: something like 20 gigawatts of new AI compute demand every year to 2030, at roughly $30–40 billion per gigawatt, most of it going to compute. You can already see the cheques being written — a five-gigawatt Stargate build in the UAE, a 6.6-gigawatt commitment in Saudi Arabia by 2034. When the state and the operator share a balance sheet and a horizon, numbers like that are a decision, not a fundraising round.

The second is mandate. National AI strategies in the Gulf aren’t aspiration decks; they’re instructions. They give an operator cover to build ahead of demand, because the demand is effectively policy.

The third is anchor demand — government and regulated sectors that must keep data in-country and are ready to be the first big tenant. You don’t have to find the market. The market is told to show up.

And the fourth, which an IDC line in the session named more plainly than most panels would: “The conflict in the Middle East has introduced new layers of complexity for CIOs, from supply chain disruption to data sovereignty concerns and IT budget pressure.” In the Gulf, sovereignty isn’t an abstract preference. It’s been sharpened by real disruption and real geopolitics — enterprises actively pulling workloads off public cloud and back onto infrastructure they control. Urgency is its own forcing function, and the region has it in a way Europe, for now, does not.

Stack those four together and the speed stops looking like cultural dynamism. It’s what happens when capital, mandate, demand and urgency all point the same way at once.

3. Europe runs on different physics

Now run the same checklist for Europe — and specifically for a market like the Netherlands.

Capital here is opportunistic, not sovereign. It wants the business case before it moves, not after. There’s no single national mandate telling an operator to build ahead of demand — there are twenty-seven national conversations and a fragmented enterprise market. Public-sector anchor demand exists, but it’s procured, not handed over.

And then there’s the constraint the Gulf never has to think about: power.

Intel’s own GM on the call was explicit about it — the region is winning the world’s data-centre builds in part because of its power capacity. That’s the quiet asymmetry. The Netherlands is living through real grid congestion. Securing a large new connection, in the right place, at the right time, is no longer a formality — in parts of the country it’s a queue, and in some it’s a flat no. You cannot brute-force a gigawatt-scale build when you can’t get the gigawatts. The thing the Gulf treats as free and infinite is, here, the binding limit.

Here’s the turn that’s been sitting in front of me all day. In the Gulf, sovereign AI is a capital story. In Europe, it’s a power story — literally, electrical power. Same appetite, different physics. When the unit of ambition is gigawatts and your grid is rationing megawatts, the megabuild simply isn’t a road you’re allowed to drive down.

4. The constraint is the strategy

The trap is to look at the Gulf, see the 19MW build, and assume that’s the only way in — just slower and harder for Europe. Run a procurement cycle for a centralised AI factory, discover you can’t fund it ahead of demand or power it on time, and watch the whole thing stall. The build everyone admires is the one Europe is worst equipped to deliver.

So flip it — and here’s where I get optimistic rather than gloomy. If capital is scarce and power is the ceiling, the answer isn’t a bigger data centre. It’s a smaller one, everywhere: inference distributed across infrastructure operators already own, in places already connected and already powered, sold against demand instead of ahead of it. Lighter on capital, lighter on the grid, faster to stand up, and far more replicable across a fragmented continent than any one bespoke national facility.

And this isn’t a consolation prize — it’s where the economics are heading regardless. The direction of travel in the session was clear: agentic AI could push token consumption up by two orders of magnitude, with around 80% of compute cycles going to inference by 2028 — distributed, latency-sensitive and brutally cost-driven. The hardware mix is rebalancing with it; in these workloads the GPU-to-CPU ratio slides from something like 8:1 toward 1:1, as more of the work turns out to run perfectly well on general-purpose compute and smaller models. Smaller models, lower cost, closer to the data. Europe’s forced move toward distribution lines up almost exactly with where the technology is going anyway.

Hyperscalers centralise AI. Telcos distribute it. In the Gulf, that’s a strategic preference. In Europe, the grid is going to make the decision for you whether you like the phrasing or not.

So the lesson from MENA isn’t “move faster” — STL just told us European telcos already moved fast on the part that’s about will. The lesson is about shape. Europe can’t conjure sovereign capital or a national mandate, and it certainly can’t conjure spare gigawatts. So stop trying to import a playbook built on conditions you don’t have, and build the one your constraints are quietly pointing you toward.

So what should European telcos actually do?

If the megabuild is off the table, the answer isn’t to sit and wait for capital and power that aren’t coming. It’s to play the hand your assets already deal you. Four moves.

Distribute by default. Stop scoping a centralised AI factory you can’t fund or power, and start putting inference where the power and the footprint already are — central offices, edge sites, the customer premises. The grid won’t let you centralise; your own network already lets you distribute. Make that the starting assumption, not the fallback you reach for after the big build stalls.

Own the orchestration layer, not just the racks. Renting space, power and tokens is the easy part, and the first thing to get commoditised. The defensible position is the control plane — coordinating where workloads run, governing them, keeping them in jurisdiction, and billing for all of it. That’s the layer telcos are uniquely placed to hold, and over time it’s worth more than the iron underneath it.

Sell outcomes, by vertical. Enterprises don’t want infrastructure. They want a problem solved with their data kept where the regulator insists it stays. Pick the verticals where you have real local credibility — financial services, healthcare, public sector — and package the whole thing: sovereignty, compute, the model, the integration, and the professional services that make it actually land. The AI workload pulls connectivity, storage and security through behind it. Lead with the outcome, not the box.

Move in an ecosystem, and move now. No European operator builds this alone, and none has the time to try. Partner for the silicon, the edge hardware, the software layer — and get a reference customer live while sovereignty is still an open commercial question, before centralised sovereign cloud arrives to fill the gap. The advantage here is a window, not a moat. Windows close.

The Gulf gets to answer the AI question with a chequebook and a power station. Europe has to answer it with what it already owns: distributed infrastructure, regulatory trust, and proximity to the customer. That’s the harder hand. It’s also, for once, the one telcos are genuinely built to play. The operators that win here won’t be the ones that build the biggest thing — they’ll be the ones that distribute the smartest. And they’ll start now, not when the gigawatts show up.