We’ve done this before. About five years ago, “autonomy” started showing up everywhere in AI marketing. Autonomous agents. Autonomous vehicles. Autonomous decision-making. The word traveled fast, got attached to a thousand product launches, and almost nobody stopped to ask whether the systems being described were actually autonomous. They weren’t. They still aren’t. Autonomy, properly understood, requires a kind of self-direction that current AI doesn’t possess. But the word kept moving, because it was useful.

Now the same thing is happening with sovereignty. Sovereign AI is the phrase of the month. It’s in posts, panels, headlines, and ministerial briefings. People nod when they read it. They share it. They build entire arguments on top of it.

Almost nobody stops to ask what it actually means.

And the more I sit with the phrase, the less it holds up. Because once you notice the pattern, it’s the same one we already fell for with autonomy. A serious-sounding word, doing rhetorical work, the underlying technology can’t actually back up.

Sovereign is being distorted

When you say Sovereign AI, the word “sovereign” becomes the adjective. It modifies the AI. The AI holds sovereignty. Independent. National. Self-determining. Whether or not anyone means it that strongly, that’s what the phrase implies.

Now flip it. AI Sovereignty puts AI as the domain and sovereignty as the condition being asserted over it. Suddenly, we’re asking different questions. Who controls the AI? Where does it run? Whose laws govern it? Who answers when it breaks?

One of those is a marketing posture. The other is a governance question. And the strange thing is, the industry picked the first one.

That isn’t an accident. Sovereign AI, as a slogan, was popularised by Nvidia, and the commercial logic writes itself. If AI is sovereign, every nation needs its own. That’s a lot of GPUs. AI Sovereignty, by contrast, would mean someone has to actually own the thing, govern it, and be accountable for what it does. One framing sells hardware. The other creates obligations. Guess which one ended up on the keynote slides.

AI Sovereignty sits on top

Long before Sovereign AI started trending, we already had Digital Sovereignty and Data Sovereignty. Those terms have real roots. Indigenous data rights movements. EU privacy law. Twenty-plus years of arguments about jurisdiction on the internet. They were pointing at genuine problems. Where does my data live? Whose courts can reach it? Who holds the keys?

Most of those questions can be answered, at least partly, with engineering. Encryption. Key custody. Residency controls. Confidential computing. The hyperscalers have already built much of this, and post-quantum algorithms are now standardized and being deployed. You can run UK data on US infrastructure today, encrypted end-to-end, and the privacy question is largely solved.

What encryption doesn’t solve is the dependency question. The CLOUD Act allows the US government to compel a US operator regardless of where the data physically resides. That’s a legal reality, not a technical one. No amount of cryptography fixes a jurisdictional problem.

So, digital and data sovereignty are partial. Achievable in pieces. Honest framings if you’re willing to be honest about what they can and can’t do.

AI Sovereignty sits atop all of that and inherits every limitation, plus a few new ones. Models. Weights. Training data. Inference compute. Evaluation pipelines. Agent protocols. Each layer is a dependency. Each layer has a vendor, a jurisdiction, and a supplier. You can be sovereign at the top of the stack. Almost nobody can be sovereign at the bottom. The UK can decide where weights are hosted. It cannot decide who builds the lithography machines.

This is harder for companies than for nations, by the way. A country can at least pass laws and write an industrial policy. A company trying to be digitally sovereign has to buy the entire stack at retail. No tax base. No eminent domain. Most of what enterprises call “digital sovereignty” turns out, on inspection, to be vendor diversity and contractual leverage. Useful, but not the same thing.

And then there’s the other version

Here’s the part I find genuinely interesting. If you take Sovereign AI literally, properly literally, it would describe an AI system that has gained its own independence. Self-governing. Self-owning. Operating under its own authority rather than someone else’s.

We are nowhere near that. And I’d argue we can’t be, by definition. Sovereignty in any meaningful sense requires self-ownership, and self-ownership requires a kind of evaluative independence that current AI systems simply don’t have. They don’t choose their goals. They don’t own their weights. They don’t decide their training data. They run because someone funds the compute and someone signs the deployment.

Real Sovereign AI, in the literal grammatical sense, is a category that doesn’t have any members yet. It might never. And the day it does is the day the conversation shifts from governance to something much stranger.

Which makes it a peculiar phrase for a billboard.

The policy is being written on the slogan

Much current legislation has been built on this vocabulary without anyone questioning it. The EU AI Act. Parts of the CLOUD Act framing. Various national strategies. They all assume Sovereign AI is a coherent goal that nations can pursue.

I don’t think this is a conspiracy. I think it’s how language captures policy. A phrase becomes popular. It sounds serious. It shows up in briefings, then position papers, then draft legislation. By the time anyone questions whether the underlying idea actually holds up, the regulation is already shaping markets.

The result is a policy written on the premise that overstates what sovereignty can deliver and badly understates the cost.

Now, imagine we got it

Suppose, for a moment, that full AI sovereignty were achievable. Every nation runs its own stack. Its own chips, models, standards. What does that world look like?

Less investment in the companies building global infrastructure. More closed networks. The splinternet, but at scale and on purpose. Knowledge transfer slows because research stops crossing borders cleanly. Scale economics break because no model can amortize across global demand. Capital concentrates on nationally favored champions instead of competitive markets. The largest knowledge-sharing system humans have ever built fragments into national silos.

In a capitalist economy, that path doesn’t even mathematically work. Capital follows scale. Scale requires openness. The two are incompatible at full sovereignty. You can have one or the other, not both.

So the question I keep coming back to isn’t whether we can reach this. It’s whether we’d want to.

Not the version in the press release. The real one. The one with the closed networks and the slowed research and the consolidated national champions and the quiet loss of the open flow that makes most of the modern internet useful in the first place.

If we got there, would we recognize it as the world we asked for?

My honest read is that Sovereign AI, AI Sovereignty, Digital Sovereignty, and Data Sovereignty are not four problems. There’s one problem with four marketing wrappers. And the conversation has been shaped, almost completely, by the wrapper that sells the most hardware. Real sovereignty in any of these forms is a utopian belief, achievable only through a level of isolation no capitalist economy can absorb. The vocabulary is doing the persuading. The reality, when you look at the stack, refuses to cooperate.

So, before the next speech, before the next regulation is written on the next slogan, the question worth putting on the table is this. Are we trying to govern the technology, or are we trying to own it? Have we noticed yet that those are not the same thing?

---

If you find this content valuable, please share it with your network.

Follow me for daily insights.

Book me to speak at your next event.

Start managing your agents for free.

Chris Hood is an AI strategist and author of the #1 Amazon Best Seller Infailible and Customer Transformation, and has been recognized as one of the Top 30 Global Gurus for Customer Experience. His latest book, Unmapping Customer Journeys, will be published in 2026.