ATAILA Newsroom · Budapest · 2026-08-28

Process sovereignty is the right question. The answer is smaller than a five-year programme.

In an interview with index.hu ahead of next week's AI Summit, Nagymajtényi Gábor — who wrote the last government-adopted e-governance strategy — draws a distinction we think is the most useful sentence published in Hungarian on this subject this month. Sovereignty, he argues, is not only about where the data sits. It is about whether you can run, take over and restore your own systems. We agree completely. We part company on the timeline.

The distinction that does the work

Nagymajtényi separates data sovereignty — which jurisdiction holds the bytes — from process sovereignty: whether the organisation can independently operate, take over and restore what it depends on. It is a better frame than the one the market usually uses, because it survives contact with a bad day.

Data residency is a contract term. Process sovereignty is a question you can only answer by trying: can you rebuild this system from source, with your own keys, without anyone's permission, before the business notices? A company can hold every byte inside the EU and still fail that test completely, because the thing it cannot do without a vendor is not storage. It is operation.

The interview is a conference run-up piece and its headline is domestic politics, which is not our business and not our argument. We are answering the engineering claim, which stands on its own.

His numbers, and whose they are

Worth stating plainly: the figures below are Nagymajtényi's own assessments in an interview, not audited statistics, and we cite them as his. He describes roughly 3,800 isolated public-administration applications, with no comprehensive picture of which systems exist, who maintains them or where the records are kept. He puts 60–70% of the Hungarian IT market as driven directly or indirectly by state orders, much of it reselling imported products. He argues the state should run open-source or licensed models, further trained on Hungarian data, on controllable and preferably domestic infrastructure. And he estimates five years as the minimum for meaningful progress.

We have no basis to check the first two, and no interest in arguing with them. It is the last number — five years — that we want to take seriously.

Five years is the right answer to the wrong unit

If the unit is the Hungarian state — thousands of applications, decades of procurement, an inventory nobody holds in full — then five years is probably optimistic, and we would not claim otherwise.

But the interview's own conclusion points somewhere much smaller: that Hungary needs at least two or three sectors with a working, continuously developed AI service running on its own infrastructure. That is not a five-year national programme. That is a unit-level engineering problem, and the unit is one organisation with one confidential workflow.

At that scale the components are ordinary and available today: open-weight models, downloaded once and stored on hardware you own; GPUs in a room whose door you control; your own identity system, your own keys, your own DNS; an estate defined in code so it can be rebuilt rather than remembered; and someone whose job is to be on call for it. We know the parts work, because that is what we run. Every one of them is available to anyone reading this — the scarce ingredient is not technology, it is the willingness to operate rather than subscribe.

Where we would push back

The article reaches for Poland's Bielik as the positive counter-example, and it is a fair one — a national open model is a real asset. But a model is the cheapest part of sovereignty, and by some distance the most photogenic.

A country can commission a Hungarian model and still have no process sovereignty at all, if the thing serving it is somebody else's platform, on somebody else's schedule, under somebody else's terms. The model is the part that gets the press conference. The operation is the part that decides whether it is still running in 2029.

What we would do with the five years

Not wait for them. Take the single workflow your organisation would never put into a public AI service, and move that one — properly, with an operator, on hardware you control. Then do the next one. Two or three sectors with something real and running beats a national programme that is still in consultation, and it is the version that produces people who actually know how to do this.

We will be at AI Summit next week, where this conversation continues on stage. Our contribution is smaller than a country strategy and more useful than a slide: the sessions we think are worth your time.