ATAILA Newsroom · Event · Budapest · 2026-08-28

AI Summit Budapest is next week. We will be there — and here is what we are listening for.

AI Summit Budapest 2026 runs on 7–8 September in Városliget — close to 300 speakers across eleven rooms on the campus: the Magyar Zene Háza, the Néprajzi Múzeum and the zones around them. ATAILA will be there for both days. Not with a stand and not with a pitch: we are going to listen. This is the guide we made for ourselves, published in case it is useful to you too.

What we are listening for

We read the whole programme. Two sessions carry “sovereign AI” in the title, and one practitioner session is about the risks of cloud AI and deploying locally instead. Meanwhile the large cloud vendors hold several slots, including main-stage keynotes.

That is not a criticism of the organisers — it is an accurate reflection of the market. But it does mean the word “sovereign” will mostly be spoken next week by people who sell capacity, and much less often by anyone who operates it. Sovereignty is being discussed as a strategy. It is lived as an operational discipline: who holds the keys, who can restore the system on a bad Tuesday, and whose invoice arrives when the model gets bigger.

That gap between strategy and operations is the thing we will be listening for in every room.

The sessions we would not miss

Grouped by what we care about, quoted from the public programme. Times and rooms are as published on 28 August — the programme is still moving, so check the official schedule on the morning of each day.

Running AI locally

The closest thing on the programme to the question our customers actually ask.

  • Hívatlan hallgatóság? A felhőalapú AI veszélyei és lokális megoldások a gyakorlatban

    Gál Tamás, Safescribe·Tue 11:00–11:40, Prompt Aréna

  • ESA Copernicus is Open. Why Can’t Anyone Use It? An Open-Source AI Stack for Real Earth Observation

    Somogyi Viktor, Netwerk·Mon 17:20–17:40, Quantum Stage

Sovereignty

Two of the three carry the word in the title. The third is about data assets, which is the same question in different words.

  • Szuverén AI a számítási kapacitáson túl – Európa stratégiai mozgástere az új AI-értékláncban

    Palkovics Bálint, 4iG·Mon 12:15–12:35, Visionary Stage

  • Sovereign AI: The Pathway towards Equal Development of Artificial Intelligence

    Zhang Shuyan, Fudan University·Tue 13:50–14:10, Visionary Stage

  • Restart Hungary! Adatvagyon és vállalkozói szabadság

    Balogh Petya, STRT Holding·Mon 17:25–17:45, Visionary Stage

Regulation, as an operational problem

Monday morning in Future Lab is the densest hour of the two days if this is your question.

  • A szabályokat el lehet halasztani. A felelősséget nem. Az AI Act az Omnibus után: a működő AI-tól a jogilag védhető vállalati AI-ig

    Dr. Kopasz János, Taylor Wessing Hungary·Mon 09:20–09:40, Future Lab

  • Amikor az algoritmusért már felelni kell: AI-szabályozás 2026-ban

    dr. Vajda Viktor, Magyar Mesterséges Intelligencia Tanács·Mon 09:40–10:00, Future Lab

  • Compliance by design: building AI-Agents that survive the EU AI act

    Fauszt Gábor, Zenitech·Mon 12:55–13:15, Future Lab

  • AI-korszak, új kockázatok – Hogyan tegyük biztonságossá és irányíthatóvá a mesterséges intelligenciát?

    Dr. Barta Gergő, Deloitte·Tue 12:40–13:00, Innovation Hall

  • Az AGI-küszöb: Nemzetbiztonság, geopolitika és a szigorodó AI-szabályozások

    Frész Ferenc, CyEx Group·Tue 10:50–11:05, Innovation Hall

The hardware underneath

Compute is the part of the AI conversation that cannot be improvised, and it gets the least airtime.

  • Adatközpontok szerepe az AI ökoszisztémában

    Prof. Dr. Gál Zoltán, Debreceni Egyetem·Tue 11:55–12:15, Visionary Stage

  • Kritikus infrastruktúrák védelme mesterséges intelligenciával – a hibák utáni reagálástól az előrelátó, prediktív és rendszerszintű védelemig

    Pálfi József, Delta Sentinel·Tue 10:30–10:50, Méta Open Lab

  • A szuperszámítógép első éve: lehetőségek, tapasztalatok és kutatási irányok

    speaker not yet published·Tue 11:15–11:35, Quantum Stage

Hungarian-language models

Directly relevant to anything a Hungarian company wants to run on its own hardware.

  • Magyar LLM fejlesztések

    Dr. Ligeti-Nagy Noémi, ELTE Nyelvtudományi Kutatóközpont·Tue 17:25–17:45, Innovation Hall

  • Magyar nyelvre magyar modell

    Szakács Benedek, Trendency·Tue 12:10–12:30, Cinematic Studio

If anyone asks us about local AI

Short version, and it is the same thing we wrote about the Nvidia–Hugging Face deal this week: open models are free to download, free to run and free to fine-tune, and we think it has to stay that way. We run open models on our own GPUs, in our own datacentre. The weights are pulled once and live on our storage; inference never calls a hub. We deliberately run more than one serving engine, because being one runtime away from a rewrite is not a bet worth making.

The harder half is not the model. It is everything after go-live: who patches it, who is on call, who proves a year later which model version produced which answer. That is the part we sell, and it is the part the programme talks about least.

The question we will ask everyone

One question, asked of everyone we talk to over the two days:

“Which of your workflows would you not put into ChatGPT today?”

It is a better question than any pitch, because the answer is always specific and always real. Whatever that workflow is — the client files, the payroll, the test results, the contracts — it is the one that has to run somewhere you control. Everything we build exists to make that possible without hiring a platform team.

We will publish what we learn here afterwards, including the parts that contradict us.

Practical details