A working table in Rome, in the days around the symposium. Across from me, a European technology investor who runs a fund with several billion euros behind it, and who had just come back from opening a research centre in Beijing.

I have learned not to lead with this idea. It sounds far-fetched the first time, and the people who grasp it immediately usually think I have said something illegal. So I go slowly. The embassy of Canada in Italy — or in any country, for that matter — is inviolable. Italian police cannot walk in. Italian courts cannot reach inside. It sits on Italian soil and Italian law stops at the door. That status is not a metaphor and not a courtesy — it is one of the oldest and most robust legal technologies we have, and it has survived wars.

Now apply it to a building full of servers.

I get four sentences in. He does not need the fifth.

“So you mean a Canadian embassy data centre in China.”

Yes. That is exactly what I mean.

The reason the idea lands hard is that it inverts the question everyone has been asking. For a decade, digital sovereignty has meant keeping our data at home — data residency rules, national clouds, localisation mandates. That framing has one fatal property: it makes sovereignty a function of geography, and geography is the one variable a mid-sized country cannot optimise. You cannot conjure hyperscale capacity inside your borders by legislating that it must be there. You can only make it more expensive not to have it.

The data embassy inverts it. Stop asking how to keep the data at home. Start asking how to keep your law attached to the data wherever it happens to sit.

This is not speculative, which is the part that surprises people. Estonia — a country that has been the subject of a real cyberattack and drew real conclusions from it — has operated a data embassy in Luxembourg since 2017, holding critical state systems under Estonian jurisdiction on Luxembourgish soil. Monaco followed in 2021. The World Economic Forum wrote the definition down this May, in a paper on AI infrastructure and sovereignty: such arrangements “enable the creation of secure digital spaces in a host country where a guest country’s (or organization’s) data, workloads storage and compute remain governed under its own jurisdiction.”

The same paper contains the sentence that, to my mind, ends the whole self-sufficiency debate: “for most economies, self-sufficiency is unrealistic given the scale, complexity and concentration of AI infrastructure required. Therefore, AI sovereignty must be pursued through strategic interdependence.”

Strategic interdependence is not a compromise position. For a country whose geography will never hold hyperscale capacity — most of the Global South, and most small states — it is the only position that survives contact with the physics.

There is also a resilience argument that gets lost because it is boring. The instinct of a government told to build sovereign infrastructure is to build one very good facility — Tier 4, maximum redundancy, ribbon-cutting. A Tier 4 facility costs three to four times a Tier 2. For the same money you can distribute several Tier 2 facilities across several jurisdictions, and the distributed arrangement survives things the single facility does not: a grid failure, a legal change, a sanctions regime, a war. The Holy See is the extreme illustration — more than half its territory sits outside Rome. It has been doing distributed sovereignty since long before there was anything to distribute.

Now the objection, and it is the right one.

This is a legal fiction stapled to somebody else’s cloud. A diplomatic note does not encrypt anything. If the host’s engineers hold the keys, if the network egress is controlled by the host’s carrier, if the operational runbook is written by the vendor, then your “sovereign” facility is a normal facility with a flag on the door — and the flag will not help you on the day it matters.

That objection is correct, and it is the hinge of this entire series. Legal status without technical detachability is theatre. What makes a data embassy real is a set of unglamorous engineering properties: you hold the keys; you control the network path; you can operate it with your own people if you have to; and you can leave — pick up the workload and move it to another jurisdiction without rebuilding it. Detachability is the property. Everything else is packaging.

Which is why this is a two-part act, and most countries will do only the first part. Declaring that data centres may hold sovereign status is a stroke of a pen for a head of government — an afternoon’s work, a good press conference. Building the architecture that makes the declaration enforceable is three years of procurement, key management, network engineering and contract law that nobody will ever write about.

The encyclical is blunt about what is at stake if the second part never happens. Colonialism, it says, “no longer dominates only bodies, but appropriates data… These have become the new ‘rare earths’ of power.”

You can put your flag on a building. The question is whether you can carry it out.

Previously: The Intelligence Delta. Next Tuesday: Beyond the Hegemon.

Tile 2 of 9: Sovereignty travels with the law, not with the geography — but only if the architecture can carry it.

Next Tuesday, 3 November: Episode 3 — Beyond the Hegemon.

All episodes: https://richardstpierre.com/mosaic-doctrine

Palazzo Lateranense, the twenty-fifth of May. I am on a panel with James Muller, who co-founded International Physicians for the Prevention of Nuclear War — the organisation that won the Nobel Peace Prize in 1985 — and with a Chinese scholar of classical culture. The question put to us is about inequality.

I begin where anyone would begin in that building. The Church has spent more than a century on inequality: the dignity of work, the universal destination of goods, the preferential option for the poor. It is the most sustained body of thought on the subject that any institution has produced. I say so, and I mean it.

And then, somewhere in the middle of my own sentence, I stop. Because I realise that the word I have just used with such confidence no longer covers the thing I have spent the last two years watching.

We spent the twentieth century measuring inequality in wealth. We are going to spend the twenty-first measuring it in something else.

Here is the shortest way I can put it. Tokens are the currency of artificial intelligence. If your country does not have tokens — if it does not have the compute, the energy, the models and the capital to generate them at scale — then in the terms that will matter, it does not have a GDP. Not a small one. Not a lagging one. It has a number that has stopped measuring the thing everyone assumes it measures.

That sounds like rhetoric. It is arithmetic.

Wealth inequality was a distribution problem inside a shared system of production. Two countries made things the same way; one made more of them. You could argue about trade, about tariffs, about aid, about the terms on which the surplus was divided, and the argument was coherent because the underlying activity was comparable.

Intelligence inequality is not that. It is a difference in the production function itself. If one economy operates several times more efficiently because it has embedded machine cognition into how it decides, designs, procures and manufactures, then comparing its output to its neighbour's is not comparing two runners. It is comparing a runner to a vehicle. The unit of measurement quietly stops working, and the number on the page keeps being reported as though it still does.

And it compounds. Compute buys capability. Capability buys revenue. Revenue buys compute. The loop closes, and each turn of it widens the gap by more than the last. Look at what is being committed: a scale of investment in the United States over a single year that the entire European Union does not approach, that Canada does not approach by an order of magnitude. The training economics make it worse — most of the cost of building a frontier system is compute, and the compute is somewhere else. For every dollar of value created, the great majority leaves.

The Holy See has already conceded the measurement point, more directly than most finance ministries have. Magnifica Humanitas says outright that development metrics tied to GDP have, for more than eighty years, "almost systematically neglect[ed] aspects essential to the overall wellbeing of people and the environment," and that complementary parameters are now crucial. The encyclical is more precise still about the illusion underneath the optimism: "To think that new technologies will automatically benefit everyone is to ignore the evidence."

Now the objection, which is a good one and deserves better than a dismissal.

Efficiency is a virtue. It is the mechanism by which almost every improvement in human welfare has actually arrived, and a five-fold improvement in how a society allocates its resources is, on its face, wonderful. Nor is technological divergence new: printing, electricity, containerised shipping and mobile telephony all opened enormous gaps that then closed, sometimes with startling speed. Kenya leapfrogged retail banking. Betting against diffusion has been a bad trade for two centuries.

I do not think this one diffuses the same way, and the reason is specific. What diffused in every previous case was the artefact — the press, the handset, the container. What is scarce here is not the artefact but the substrate: gigawatts of firm power, capital at hundred-billion scale, and physical facilities that take four years to build. Software crosses borders in seconds. Substations do not. You can hand a country a model. You cannot hand it the ability to train the next one.

So the delta is not a forecast. It is already in the spreadsheet, for anyone who runs the numbers.

The problem is that the countries on the wrong side of it are still running the old numbers, in the old units, and reporting them to parliaments that believe the units still mean something. The first act of sovereignty is not building anything. It is admitting that your instruments have stopped working.

Tile 1 of 9: Sovereignty must first be measured, or it cannot be defended.

Next Tuesday, 27 October: Episode 2 — Data Embassies.

All episodes: https://richardstpierre.com/mosaic-doctrine

In May I sat on a panel at the Palazzo Lateranense with James Muller, who co-founded the physicians' movement that won the Nobel Peace Prize in 1985 for its work on nuclear war. In July I chaired Working Group 7 of the Global Nobel Laureates Assembly at Borgo Laudato Si', inside the Vatican gardens. On the sixteenth of July, at the Campidoglio, the Assembly adopted the Rome Declaration.

What made those weeks unusual was not the guest list. It was that two populations who almost never occupy the same room were, for about three days, in the same room: the people who decide how the world should govern artificial intelligence, and the people who understand how it is actually built.

I went in expecting the hard problem to be disagreement. It wasn't. Nearly everyone in those rooms wanted approximately the same things — human control over consequential decisions, a technology that serves people rather than sorting them, an international order that survives the next ten years. The hard problem was that the two populations were describing the same object in incompatible languages, and neither could hear the other's constraints.

That gap is the thing I actually work on. Not the ethics of AI, which are being argued about competently by people better qualified than me. The gap between what a technology does and what the people deciding about it believe it does — and the way that gap compounds, quietly, through every procurement, every regulation, every architecture choice, until a country wakes up and finds that its options were foreclosed years ago by decisions nobody thought were strategic at the time.

I have spent the summer going back through it. What follows is not a report on a conference. It is an argument, built in nine parts, that ends somewhere I did not expect when I started.

I am going to publish it the way you would watch something: as a season. Nine episodes, every Tuesday, starting 20 October.

Season One — Rome

1. The Intelligence Delta

2. Data Embassies

3. Beyond the Hegemon

4. The Automation Trap

5. The Un-Glamorous Revolution

6. Democracy's New Crisis

7. The Colleague, Not the Tool

8. Ethics Over Legality

9. The Mosaic Doctrine

Each one stands on its own. But they are built in order, and the ninth is the reason for the other eight.

Episode 1 lands on 20 October, and the last one on 15 December. Subscribe and I will send them to you.

Next Tuesday, 20 October: Episode 1 — The Intelligence Delta.

All episodes: https://richardstpierre.com/mosaic-doctrine