Rise of AI Conference 2026 · Berlin · Meta Stage

From Industrial Powerhouse to AI Powerhouse

Thomas Jarzombek · Parliamentary State Secretary at BMDS (Germany's ministry for digitalization and state modernization)

At Rise of AI 2026 in Berlin, Thomas Jarzombek, Parliamentary State Secretary at Germany's digitalization ministry, reviews his first year in office and argues Germany should turn its industrial strength into AI competitiveness. He describes ongoing EU trilogue negotiations to simplify the AI Act, including extending high-risk-system deadlines by a year and working to exempt the Machinery Directive from the Act's scope, alongside a national German AI Act and a planned one-stop-shop regulatory model. He presents three government AI projects: Spark, an open-source tool that checks infrastructure-project paperwork in hours instead of about half a year and can cut approval time in half; the Agentic AI Hub, which pairs startups with municipalities to modernize legacy government software; and the Deutschland Stack, a sovereign compute platform built on European cloud providers. He argues against building an independent European alternative to US large language models, favoring what he calls blended sovereignty, integrating strong German startups into global AI supply chains instead. In Q&A he discusses SME adoption, shadow AI use by public servants, and the upcoming EU Digital Identity Wallet.

Key takeaways

  • Jarzombek says EU trilogue talks aim to exempt the Machinery Directive from the AI Act's scope and have already made rules easier for SMEs and extended high-risk-system deadlines by a year.
  • His ministry's open-source tool Spark checks infrastructure-project paperwork in hours instead of about half a year and can cut approval time in half.
  • The Agentic AI Hub matched nine startups with 20 municipalities from a pool of almost 400 startups and about 200 interested municipalities to modernize legacy government software.
  • Jarzombek argues for 'blended sovereignty', integrating strong German startups into global AI supply chains, rather than building an independent European alternative to US large language models.
  • He says the EU Digital Identity Wallet, a central ID and credential tool, launches on 2 January with about 150 business partners already testing it.

Detailed notes

Simplifying the EU AI Act

Jarzombek opens by marking Rise of AI's tenth anniversary and reflecting on his first year as State Secretary. He says the day of the talk coincides with a crucial EU trilogue session on simplifying AI regulation, part of what he calls the omnibus format, and lists results already achieved: easier rules for SMEs, a one-year extension of the deadline for high-risk systems, and the deletion of unconditional requirements such as mandatory training on AI tools. He says the current fight is over integrating the Machinery Directive with the AI Act, with Germany proposing a lex specialis approach to get machinery out of the AI Act's scope, though he expects only a partial "landing zone" rather than a full solution. Domestically, he describes a national German AI Act and a planned one-stop-shop model, where an industry's existing regulator also handles its AI questions.

Three AI projects: Spark, the Agentic AI Hub and the Deutschland Stack

He presents Spark, an open-source AI tool built by his ministry that checks whether documents for large infrastructure projects, such as bridges or railroads, are complete and consistent; a check that used to take about half a year now takes hours, and he says the tool can cut the overall approval process time in half, with more releases planned. He describes the Agentic AI Hub, created because officials previously had no incentive to work with startups: it matched a first batch of nine startups with 20 municipalities, addressing a landscape of more than 20,000 fragmented, often legacy municipal software systems, out of a pool of almost 400 interested startups and about 200 open municipalities, with plans to scale successful pairings to 50 or 100 municipalities. Third, he describes the Deutschland Stack, a standard framework and sovereign compute platform his ministry is building for other government departments, tendered to European cloud providers for digital sovereignty.

"Blended sovereignty" instead of a separate European AI stack

Jarzombek pushes back on the idea that Europe should build a complete alternative technology ecosystem to the US, saying he does not believe Europe can compete head-on in large language models given how far ahead US players are in model quality, GPU access, talent and venture capital. Instead, he advocates what he calls blended or connected sovereignty: integrating strong German and European startups, such as a Siegen-based quantum computing company backed by Schwarz Digits and the photonics startup Black Semiconductor from Aachen, into the global supply chains of providers like Azure, AWS and Google, creating mutual dependency that strengthens both sides.

The government's role: experimentation, not subsidies or bailouts

On whether government should subsidize AI adoption at large enterprises, Jarzombek is skeptical, saying industry knows best what to do for itself; he sees government's role as being a lead user, reducing regulatory hurdles, and supporting SMEs through existing support programs. He argues industry adoption of AI is uneven across sectors and warns that companies which wait too long risk losing competitiveness, stating that no government can protect companies that stop being competitive.

Q&A highlights

Asked why startups struggle to sell AI into SMEs, Jarzombek says the market itself will sort winners from laggards, illustrating adoption with an anecdote about craftsmen using a drone to inspect a leaking roof before repairing it. On working with municipalities first, he says local government is citizens' first point of contact with the state and a lower-risk place to experiment than, for instance, interior-ministry systems, and that the Agentic AI Hub favors adapting existing solutions rather than building new ones. Asked about a public app store for municipal services, he says more competition is needed, since public IT offices currently often face none, and warns about "shadow AI," public servants using personal AI tools with internal or personal data because official tools are not yet good enough. Asked to reflect on his first year, he contrasts the previous fragmentation of digital policy across several ministries with the current dedicated ministry, and highlights the EU Digital Identity Wallet, a central ID and credential tool launching on 2 January with about 150 business partners already testing it.

Quotes

The business of the future is the business of AI.

Thomas Jarzombek

No politician and no government can protect companies that are not being competitive anymore in Germany.

Thomas Jarzombek

Today you can experiment with AI. Today you can make mistakes.

Thomas Jarzombek

Questions answered in this talk

What is Germany doing to simplify the EU AI Act for industry?

According to Thomas Jarzombek at Rise of AI 2026, Germany pushed for a one-year extension on high-risk-system deadlines, deletion of some unconditional training requirements, an effort to exempt the Machinery Directive from the AI Act's scope, a national German AI Act, and a planned one-stop-shop model where an industry's existing regulator also handles its AI questions.

What is Spark, the German government's AI tool?

According to Thomas Jarzombek at Rise of AI 2026, Spark is an open-source AI tool built by his ministry that checks whether the paperwork for large infrastructure projects, like bridges or railroads, is complete and consistent, cutting a check that used to take about half a year down to hours and roughly halving overall approval time.

Does the German government want to build a European alternative to US AI models like ChatGPT and Claude?

According to Thomas Jarzombek at Rise of AI 2026, no; he argues Europe cannot currently compete head-on on large language models given US leads in model quality, GPU access, talent and capital, and instead favors 'blended sovereignty', integrating strong German startups into global AI supply chains.

Mentioned

EU AI Act, trilogue, Machinery Directive, one-stop shop, Spark, Agentic AI Hub, Deutschland Stack, blended sovereignty, shadow AI, EU Digital Identity Wallet

Keep exploring

More from Thomas Jarzombek

More from Rise of AI Conference 2026

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 3,876 words

Show the full transcript

50? 10 10 10 10 No, okay. So, I just wanted to find out how long your want >> Stay on the on the plan you have now even 30 if you want. >> 30, okay. You would like to listen 30 minutes to what we're doing in our government? Ah, okay. Some like to. Great. So, truly a pleasure to be here especially right now. And so, yes, first of all many greetings from our secretary from Kastilberger and so, and first of all a big congrats to the 10th anniversary of this conference here, Rise of AI. I think it's kind of family meeting. So, I'm lucky to be here to have been here for several times and see a lot of known faces here.

Great and Fabian, Daniel, all of you great congratulation on that. So, last year when I was here we just started with our new ministry and now one year later I think it's almost been today one year ago when the chancellor was elected and we started Kastilberger got minister. And so, just to to address some things that we have done and then to address some things I believe that's important for Germany and also on the industrial side. So, maybe let's start when I was here last year I said so we want to make regulation easier and have we delivered yet? It depends on how this week will go because right now today is the crucial day.

Today is the day of trilog and we pushed very hard in the last year in this trilog format in this omnibus format of the European Union. So, the basic idea is to make Europe more competitive and to make regulation easier and better applicable. And therefore, we had very long negotiations with all all other member states of the EU Commission and we achieved some goals. We make it easier for SMEs. We will bring the the the deadline for this high-risk systems. It will be extended by 1 year. The third thing is that unconditional words like training on AI if you implement AI tools are being deleted. So, these are three things that we already achieved. Right now, there is the debate specially about machines.

How can we integrate all the necessities necessities of the machinery directive together with the AI Act and there are a lot of concerns in the industry that these both regulations do not fit together. And therefore, there are still today very hard negotiations on the on on Brussels about trilog. The first approach was 2 weeks ago. They couldn't find a common solution. I'm confident that today we will make it better and that we will see a result here today because basically the idea is to get the machinery directive out of the scope of the AI Act. There will be some further mitigations on that. It will not be 100% by this and so we also had as Germany an idea.

We call it lex specialis, but nevertheless, I think we will have here some kind of landing zone. So, that makes things easier. From our point of view, is that enough ambition? No. We need to get further. There will be upcoming fitness checks in the EU and we have to make it easier and I believe specially for startups, for science, for SMEs, it's still pretty complicated to adapt to all the things that are addressed in the EU AI Act. The second thing is that we implemented the German AI Act for the national regulation out of that and so we try also to make it as easy as possible. And what's what's possible with this European framework.

And basically one idea is to have a one-stop shop so that wherever if you're in an industry that's already regulated, go to your existing regulator and he's the one that will care for also all the AI topics. And the rest that will happen has to happen inside the regulation system and should not be of your concern so that you don't have different faces that you're that you have to deal with when it comes to implementing AI. So, this is part of regulation. Maybe it's bothering you but nevertheless it's our core business and I think with regulation you can do a lot of harm and we believe must use it more. And so this is the second part.

What are we using AI for? Let me start with Spark. Spark is a tool if we're building large infrastructures. Today it's the approval process is pretty complicated. To build a new bridge for instance on new railroads, you have really tons of documents that have to be matched. Is everything for the application complete? Do all the different papers fit together? And therefore in the beginning of this process just to find out if everything is complete and if it fits together, it takes half of a year. And we built an AI tool that can realize this within hours. So, this is real advantage and also in the further process we can speed up we can speed up the approval process for large infrastructure by by 50% so we can reduce the the amount of time to to the half of it today.

Minimum. And there will be coming more updates and it will increase. So this is Spark. Uh It's published as open source on open code. It's our own platform. Maybe you have looked at it. There's release one already uploaded and release two and three will be coming. So this is one thing. Another thing is we want to implement more startups. So this is for us really important because in the past government was not good in making business with startups. Yeah, because all the people in the in the bureaucracy they had no incentives at all to work together with startups. There were a lot of risks, but nobody was incentivized for the advantages, for getting in innovation, for strengthening young companies, to help them as being a first customer to acquire venture capital.

And this is the reason why we started a Gentic AI Hub. And the first thing we are doing is bringing together municipalities. We have a real problem in our municipalities because they have more than 20,000 different software applications on the out there. And and a lot of them are really legacy and they don't have interfaces or standardized interfaces. And so over years and years and years nobody found a solution how to modernize all this. And so we thought if there is a impossible mission we need startups for that. Yeah? And so basically the idea is with this a Gentic AI Hub to bring together startups with this municipalities. And it works out. So we have a first batch with nine startups, 20 municipalities, three months of time.

We are now in week I believe week eight. And so it works out quite well because it was clear in the beginning that we find out a lot of startups that would be interested in cooperating. Almost 400, but it was not so clear that we found enough municipalities that are open for cooperation with startups and it was really good was really good for us to to realize that there were almost 200 municipalities that wanted to cooperate in that project. And so we see you an openness. And the interesting fact is that it's not the Lord Mayors or the Chiefs of these of these municipalities that come to our ministry and talk to our minister. We discovered the AI enablers somewhere in this building of a municipal government.

And so I think these are the right persons to address because their decisions are really made. And so we want to do further batches. We also want to extend this agentic AI app also to other businesses. And the basic idea is to scale this. It will never work if we only take one startup and one municipality even if it's a big one like Berlin or Cologne or Munich or whatever. Düsseldorf have to say and suddenly MP for Düsseldorf here, yeah. We are also in Cologne, yes. And never nevertheless it's important that it can scale and it will be a success if we find not not everyone. So it's a portfolio idea and not every of these corporations will succeed.

That's obvious. So but we want to see some of them that can be scaled to 50 or to 100 of municipalities. So that's what we try to achieve. So see Anna Gründler there. She's also part of this with Deep Set. So yes, very appreciate. These are two projects. The third thing is that we are pushing our ministry as being so to say AI enabler enabler in the government. And um what we are building is a platform for all the other departments for processing AI. We have a thing that's called Zustimmungsvorbehalt. That means that every other department that wants to deploy digital projects, they need green light from us. And so, we built a standard framework that's called the Deutschland Stack, the D-Stack.

So, there are principles, there are core components, and also we are going to define AI in that. And but we want we want really to enable all the other departments to use more AI. This is our aim. And therefore, we made a tender for a platform for a compute platform. And when we're talking about more sovereignty, digital sovereignty, it's obvious that we are aiming for European providers, European cloud providers to get contracts with them for the compute of this kind of project. We have with Keep it a front end for that, and this is what we bring out to the other departments, and we want to push that on a sovereign technology. So, these are three three projects I would just want to address here to show what we have done in the first year, and what's our aim now for the upcoming time.

And so, if we look at Germany from industry perspective, um the industrial powerhouse, I think these days I often hear that it's going that bad, and we are in a bad position, and whatever. I mean, we are economy number three in the world, and we are not the third biggest country in the world. So, quite Germany is quite successful as an industry location. And we need to charge this with AI. So, this is important. And we want to see that industry is also bringing out a higher uptake on AI. So, that's really important. We see in every sector of the industry companies that are very open, that are who are going forward, but we need more of them.

So, that's pretty important. And in the debate about shall we build an European ecosystem, an European tech ecosystem as a complete alternative to the American ecosystem? Uh I believe this was proposed for for many many years. I don't believe in that. I think that what we need is kind of connected or blended serenity. That What does it mean? I believe the If you look at the AI models, you see there are some companies, but they are coming all from the US. They are so far away in the meantime. When it comes to the quality of the model, when it comes to the GPU power they have, when it comes to the talent they can onboard, the whole amount of venture capital that's there.

Uh I'm not convinced that we can build kind of LLM on the same technology that Claude and and ChatGPT and others are using. Uh I don't believe that we can compete in that field. Maybe there comes a kind of disruption. We will see new algorithms, new technologies. Very interesting also what SAP is right now doing with Priapus to go in different kind of technologies. There is really an opportunity. When it comes to these basic LLMs, I see that these US players are pretty far away. And so, is it the right move to say so we want to force our industry to use models that are maybe not that good as we see the US models?

I don't believe that the industrial players will will play this kind of game because everybody wants to be equipped with the best technology. But I think what we need is to in to implement or to infuse more of the power of our startups into the whole global ecosystem. That's what I call blended or connected serenity. So, we have very strong companies. For instance, maybe you have seen this week there was an announcement about series A of electron. It's come It's a small company building quantum computers. They're coming out of Siegen in North Rhine-Westphalia. And Schwarz Digit is investing also in their early bird. Um so we have different players in quantum computing and I believe that the German quantum computing ecosystem is globally really strong.

When you look at photonics, a further technology, see startups there from Aachen, uh Black Semiconductor. They're building a chiplet with photonic connections and so um a lot of different technologies in the whole AI field. There are really interesting startups all around the place. To integrate them into these global supply chains of uh of Azure or AWS or Google, I think this will bring success. And so then Serenity is connected because um it's a win-win for the big players to uh to to benefit from the innovation power of the startups here. But on the other side, that we have connected dependencies and nobody is able to switch off the other side. And I think that's what makes businesses strong, that's what forms ecosystems, and I think this is the right way also for the industry to go.

Is it the right way for us to say we go to the industry and give them subsidies for implementing AI? There is also a debate ongoing. We have some support programs for SMEs, and maybe that's helpful even for very small companies to enable the ones in this in in these companies who wants to start AI projects to help them, to give them some authority also by the government to say we have also some subsidies for that. But is this uh a smart move in a in a in a large-scale for the for the large enterprises? I don't believe that. I believe very much in there There a sphere for the government. We want to be lead user for applications.

We want to implement things. We want to reduce regulatory hurdles. But when it comes to the industry, it's an industry business. And I think the industry for themselves they know best what to do. But one thing is for certain, the business of the future is the business of AI. And today you can experiment with AI. Today you can make mistakes. But I think there will be a time and maybe that's pretty near where there will no mistakes be forgiven anymore because others are such much farther away. And so everybody has to step in. That's my message in the end when it comes to the industry. And now thanks for listening and I'm happy to have some Q&A.

>> >> Thank you so much, Thomas. Really inspiring. I think that because of what you said in the in in the last part about the industry, one thing that we realized in the German AI Association when we talk with our members is that it is still hard for startups, scale-ups to sell AI into SMEs, for example, or in in in that. That we we we think that there's a certain how can I say, not technology openness in in in the economy. Do you see that, too? Or or how can we overcome that? >> I see in every industry, in every branch, there are companies that are ahead and others who are waiting. Yeah. Waiting is dangerous, but um uh I don't believe that um So it's it's a decision also of the market.

And you see that the ones that are not going with the innovation, they will at some point also have to leave the market. And they they bring their resources to others who are smarter. And when I had a hole my when the roof was leaking in the house of my mother, I I there and uh craftsmen, they were already there and they had a drone flying over the roof and there when I entered the site, the place, yeah, they told me, "Okay, we flew with our drone over the building and you know know there is the reason why it's leaking." I said, "Great." And you see all these uh applications everywhere. If we can do something to to to to to to really help or to encourage uh these companies or decision makers in these companies, we are totally open to do that.

That's that's true. But um I think uh we have to stay by the truth and no politician and no government can protect protect companies that are not being competitive anymore in Germany. >> making stupid mistakes. Yeah, yeah. That's that's definitely it. In the beginning you you talked about the communities and and and which you're talking about to get uh cities and online and and and having having AI there. It is the first I think it is that important because the citizens their first contact with the government in this is via the local community and I think for that reason you do that or >> And you can experiment better. I mean, if you are talking about uh businesses of the Ministry of Interior Affairs, everybody says, "So, okay, you need the highest BSI certification and privacy is absolutely key and so uh you see that maybe Russian hackers will attack that and so on and so on."

If you're talking about very low-level services from a municipal level, you're able to experiment better. And what we try also with these startups is to implement existing solutions, not to build something new. This is basically also an idea uh behind Deutschland steckt, not to build everything by yourself, to buy solutions from the market. And what we try to find out is uh can we adapt existing solutions to these municipalities? And in first step uh they can come as they are, but if they want to scale, then we will come also and ask for more sovereignty in some things and say, "Okay, what kind of cloud are you using? What LLMs are you possessing?" But I see solutions that are also LLM agnostic.

And so, what I hear from all these startups is, "Hey, we are startups. We are flexible. We can do that." But it's obvious if you deliver this for 50 customers, you can adapt your product. If you're testing it with the first customer, uh I think it's not realistic. >> Is the end goal, uh let's say to have an public app store for the for the communities and they say, "Bauamt, Ordnungsamt, Friedhofsamt." And you you get your >> >> Yeah, that's a good question. You get the right Yeah. >> I think we need more competition. So, what we see is that every every time in these public offices, you have somebody building IT solutions and they have no competition at all.

And the users, they tell us, "Okay, we find we we see that there are software out there that is years ahead. And why does it take so long that these innovations are implemented in our government applications?" >> Mhm. >> And when we're talking specially about AI, we see this off shadow AI that uh uh the the public servants, they're not stupid, yeah? They're using their smartphones, their personal uh cloud services, their personal AI services. But this is dangerous in some way, yeah? Because they upload whole documents, internal documents, private personal data. Uh no, I'm sure nobody will do this, but maybe. >> >> Maybe some of them thought about that, yeah? Maybe some of And so, this is the reason why we have to we we we need to be faster.

We need to be faster to deliver internally for our public servants uh solutions that are on the same level as the ones out outside of the market. And the only path to this is to be open, to have competition, and not only to have a single person to decide because that normally does not lead to more speed. >> Last question, 1 year being State Secretary of the Ministry of digitalization and state modernization, how can you describe was it a whirlwind? Was it different from the things you've done before in in in this case? How did you make this year? >> Yeah, absolutely. So, in the past, it was fragmented. All the topics of broadband and mobile networks were integrated in the Ministry of Transportation.

All the digital business and startups in the Ministry for Economic Affairs and everything of government technologies was in the Ministry for Interior Affairs. So, it was fragmented and in every of these departments, it was never in the in the core of the attention of the minister. >> Mhm. >> And to be honest, I mean, look who were the persons that were ministers in these houses these years ago. Could you imagine debating with them about EU DI wallet or about LLM models or something like this, yeah? And with Carsten Wilberger, you can. >> Yeah. >> And he's asking that. And he has some projects like the EU DI wallet that's really important for us. It will start on the 2nd of January next year.

So, it's the central ID tool. It's also central wallet where you can bring credentials from your from your business also inside. We have in the meantime 150 business partners that are testing with us and want to start with our EU DI wallet. Also because of privacy and privacy means that we will not take your data and recreate your business model on your own platform as maybe some of these hyperscalers are doing. You can trust that your data and your business will stay your data and your business and will not be ours. So, this is a project where Carsten Wilberger brings together all participants every 2 weeks and he goes there in detail. And if we see problems, he does not end up the meeting that somebody says "Wir sind auf a good way.

Then he says, "Okay, describe exactly describe exactly what steps are we going uh to solve that problem?" And that's a complete different philosophy as everything I have seen now in the meantime of 16 years here in this ministry business 15 years. >> Well, 10 years Rise of AI always a good friend of this conference. Say thank you to Thomas Jarzombek. Thank you so much. >> >> Thank you.

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.