Domyn CEO Uljan Sharka: ‘We're a few quarters away from $1bn ARR’

0:02John Thornhill:Hello, and welcome back to the Sifted podcast. I'm John Thornhill, Sifted's founder and podcast host, and I'm delighted to be joined this week by Uljan Sharka, who has surely one of the most remarkable backstories in European tech. Ulian came to Italy in 2008 from Albania without knowing a word of Italian. He developed a fascination with tech, later worked at Apple in Silicon Valley, and is now the co-founder and chief executive of Domin, the Italian AI startup unicorn which builds models and agents for enterprise use cases as well as AI infrastructure. We're going to be talking about how Domin is trying to apply AI very differently from many other AI companies, the virtues of small language models, and how Europe has a real chance to compete against the US and China if it gets things right.

0:51John Thornhill:Welcome to the show, Ullian. Thank you for having me, John. First, I'd love to hear about your personal story. Tell us, why did you leave Albania at age 16? Studies. So I was looking for a journey where I could study abroad, and this brought me to Italy. and then you know I was completing my high school studies and ready to get to university but this internship at Apple changed my life once I started there. This was the period when Apple was launching the iPad and there was I think a peak of interest around the company and that internship then eventually became a full-time job dropping out of my plans for university.

1:37John Thornhill:How did you get to Italy in the first place? So I got to Italy through Albania. So basically this was, you know, not an easy journey because back in the days Albania was not yet in the Schengen group. So it was, you know, a very, very, I would say complicated experience. But studies, as you know, are something that Albanians pursue abroad, a lot in the UK as well. So my initial plans were to come in the UK, but then I happened to be in Italy and that's it. Okay. And how did you develop your fascination with tech? So it started early on, already in Albania. I was literally trying to change every piece of technology that I had on my hands.

2:27So I was quite obsessed with gaming. but also any device that I was buying, you know, from little radios to computers, I was trying to change them and to make them mine. I was fascinated by how all of this works. So beyond what I could learn at school, I wanted to put my hands into it. And I remember I created my own loading screen for my personal laptop back in the days, which had its own, you know, logos and music. And this idea that you can actually teach the computer to do things back then was very basic. but still it was somehow expressing your desires was so fascinating that that brought me to dive deeper into it and turn it into my main passion.

3:09John Thornhill:And did you teach yourself to code or did you study somewhere or how did that work? So during high school, I had the chance to hop on a program, which was a course for coding. This was the time where coding was becoming popular and schools were starting, you know, side programs for people to apply. And I did that. I had a very good professor who gave me a chance already back in Albania. This was as easy as, you know, some basic programming of computers or nothing special. But then after I developed a passion on it, I self-taught myself by spending nights and nights to build the programs, small games, give it to my friends, get feedback.

3:52And that became an obsession eventually. And how did that lead to you going to Apple? So I started this job as a freshman, you know, after high school to literally back up my finances. And it was just another job. But once I got in touch with the culture, once I saw, you know, how this technology was put together and what it was starting to represent for people and how much people were falling in love with the technology because it was enabling them to do things they were not able to do before. It was some sort of freedom realization that you could see in people's eyes and their passion for this technology.

4:33It got me so obsessed with the idea that these tools can change people's lives that initially I wanted to do it more and more at Apple. And at some point, I just decided to start a new company.

4:46John Thornhill:And what were you doing at Apple? What was your job there? So I was helping large enterprises to adopt Apple technology in a time where management teams were trying to enable work out of office. So they were buying lots of iPhones and iPads so that people could connect to their email out of office securely. And iPads were being deployed, you know, in broader operations. So engineers, for example, in manufacturing plants, they were leveraging these screens, you know, to interact with machines in a new way. So that process envisioned companies not just buying these devices, but also integrating them in very sophisticated knowledge systems.

5:27And I was helping them to achieve that.

5:29John Thornhill:I've heard you say that you were quite struck by how incredibly smooth and sophisticated consumer tech is and how, frankly, pretty lousy enterprise tech is. Is that right? That's correct. So we were, you know, selling these hundreds, thousands of devices to each individual customer from time to time. And we were getting feedback that the iPhone was such a delightful experience. You know, you can just turn it on and it works. And this was the time when Siri was being introduced. So you could ask it things like, you know, not just how's the weather going to be tomorrow, but also show my next meeting in the agenda.

6:08and people were asking like, why can I not do that with my work data? And that was primarily because they were trying to solve a problem, which was information systems at work were so complicated that they were creating a digital divide. So on one side, these organizations were spending lots of money to digitize and on the other side, they were creating a gap because some of their best people could not use this because of lack of knowledge or because how complicated they were. So Apple was introducing to them an opportunity and a vision, which was simplifying access to that digital infrastructure as easy as turning on an iPhone or asking Siri, how's the weather going to look like tomorrow?

6:47But we're not able to deliver that at Apple, being Apple a consumer company and not focusing on integrating, you know, ERP data or CRM data.

6:55John Thornhill:So you returned to Italy, inspired by that kind of Apple genius, as it were, and you launched iGenius and Domin, which became Domin back in 2016. What was the initial founding thought for your company? So it was quite obvious to me by interacting with these clients that they were having a problem with this digital divide. So the more technology they were introducing in the enterprise, the more people were feeling uncomfortable because they had to learn something additional to their expertise, something new. and this was not easy for everyone. So this clear division between enterprise technology and consumer technology was part of feeding that problem because people were used at home to literally talk with devices.

7:47They were talking to machines after Siri, Alexa became a thing. Then they were going back to work and they had to learn these very complicated interfaces. So I saw that as a huge risk for society because it means that technology was not just an opportunity, but it was also not an equalizer anymore. It was actually enabling people that knew how to use it to become superhumans and everybody else left behind. So I wanted to close that gap. Coming from Albania where, you know, we had civil war in 1997, a very complex social and economic environment, I had this feeling that technology was going in their own direction.

8:27It was actually dividing people instead of helping people to enjoy opportunity on a par. So this is what pushed me to start this company and to leverage this idea of AI, which was already solid back then, even though we're talking more than 10 years ago. And why I came back to Italy? First of all, because I just love Europe. I love to live in Europe and it's part of culture, I guess. so there was a personal aspect to it. And where is Domain based? In Milan, Italy. But the second part to it was that I strongly believe that in order to make technology human you need to sort of start a renaissance, a digital renaissance, something that Italy has already done before in a different context.

9:16But I got obsessed with this idea that to solve the digital divide we need a digital renaissance where we don't develop technology for the sake of it which is what I was seeing in Silicon Valley, and I think is even more today with AI. But we needed an environment where people are valued more, people are central to everything, and then this renaissance could use technology as a tool to enable people.

9:39John Thornhill:And so you're very much kind of trying to ground the technology in specific sectors like finance and healthcare and government and so on. Can you tell us more about that? Give us some examples of what are the kind of use cases that you're helping customers use this technology for? First and foremost, we're helping them to own it. We believe that there is no future with artificial intelligence where a few private companies own and control it and everybody else is consuming it from them. So owning it means putting AI in a box, giving it to them, handing it over so they can control it and they can be sovereign.

10:15No remote switch, no dependency even on us at Dominic. Once we enable them to own it by literally transferring legal ownership into the AI models, then we give them a toolbox where they can build use cases. For example, we have clients in financial services that are building AI products out of their data, whether that is better risk models, whether that is dynamic pricing systems that enable them to price their products in a more efficient and personalizable way. or when it comes to governments, we're helping them to digitize public infrastructure faster or to implement it in mission-critical areas like defense.

10:59So it's as easy as putting it in a box and giving it to them to innovate software. So the main use case here is to upgrade old software, which is the operating system of governments and the private sector, into something that is more intelligent.

11:13John Thornhill:And what are the biggest challenges these organizations face when they're trying to adopt AI? I'd say security is the first one. So they don't want to hand over their data to a third party. They are dealing with very mission-critical information that is regulated to some extent, but to other extents, it also creates systematic risk for society. So helping them with making AI safe is more important than the use case at the beginning. And once you have enabled that infrastructure, then it is possible for them to not just invent new things, but just improve what already exists. This can be, you know, things like innovating front office.

11:52Imagine, you know, those long calls with public administrations when you're trying to do anything from healthcare to, you know, getting your passport. All of these things are frustrating. Talking to the government today is frustrating. And AI can solve for that. It can make the experience much more delightful, fast, and it can enable governments to actually make public resources available at scale.

12:19John Thornhill:Are there examples of companies that have used OpenAI or Anthropic models and regretted it that you know of? I think there have been a few cases that are also public. I think it happened with Samsung that some of their patents at some point leaked to a public user of ChatGPT. There was another case where there was a public chatbot which sold the car for free at$0 because it was manipulated by an end user. And this happened because the company that was providing it had no control over the system. It was just handed over to them as an API, so as an application programmable interface where they could just put some rules, but they don't own that infrastructure.

13:02So these are examples of how these mission-critical use cases can be material for businesses and governments. Imagine if that same problem happened, you know, in energy. National energy grid company that could potentially turn off an entire country just because of similar mistakes.

13:22John Thornhill:I saw an interview the other day with Alex Karp, the chief executive of Palantir, who made a very similar case. who was warning a lot of client companies that by working with proprietary frontier model companies, they were in effect transferring their business alpha to somebody else. They were making customers want to control of their own compute, their own models, their own data stack and their own alpha. And that's a risk if you use a frontier model, a proprietary frontier model produced by one of these big US tech companies. You would agree with that, I assume? 100%. This is the risk of disrupting the rule of law and intellectual property because what keeps the world fair today is that through the rule of law, we can enable freedom and everyone can enjoy their inventions.

14:08But if with the current architecture of AI, those inventions are being transferred for free to a provider of AI, this is like feeding a beast. This is not just one entity or multiple entities, whether those are public or private, losing IP, but also reinforcing a monopoly that is going to be very, very hard to beat in the future. even for new ideas or new startups that can emerge in several areas of the economy. So this is a two-sided problem. On one side, you eliminate competitiveness for who's using the centralized models, as you just mentioned. And on the other side, you're eliminating freedom for the future because it's going to be very hard to bid five or 10 companies that own the world's IP in every sector.

15:02John Thornhill:So that clearly has massive implications for sovereignty. Is that a big advantage being a European company at the moment, the fact that there is so much focus on trying to reassert European sovereignty? I think the short answer is yes, even though I don't like the idea of isolating ourselves as Europeans and saying us against, you know, the US or other regions. I think we should fight for an integrated global platform that embodies European values, values for which we have been fighting, you know, in terms of democracy, in terms of regulation. So we should not get in a trap where these values are now being weaponized against Europe itself.

15:43And I think we should keep it up with just pushing this in the right direction and everything will just follow. This is my strong belief.

15:52John Thornhill:Who are your biggest customers at the moment? So we have customers from different sectors, from the Italian government to leading manufacturing companies like Fincantieri, one of the largest shipbuilding companies in the world. But even US customers like Bank of New York or other banks in Europe like Rabobank in the Netherlands. And what are they using your products for, your services for? So they are getting an AI operating system to power better software for different solutions. Some of them we cannot even see because as part of our actually aim to enable sovereignty, we give them with the tools.

16:31We provide them the tools to actually build these use cases, but then they can do whatever they want with it.

16:36John Thornhill:What is your annualized recurring revenue at the moment? So we have crossed the 200 million mark as of July 2026. Okay. And how ambitious are you for the company? How big do you want to get? So our next big milestone is crossing 1 billion ARR, and that's going pretty fast. When do you think you might hit that? I think we're a few quarters away. Okay. You're also working on a frontier language model in Europe, Evropa. Can you tell us about that? Yeah. So you talked about ambitions and, you know, one billion ARR is the next short term milestone, but we believe there is a chance here to build a one trillion plus market cap company from Europe with global ambitions.

17:29So now global ambitions means you need to be at the frontier all the time. Europa is exactly that. It's a consortium that aims to leverage a typical European partnership approach where public and private companies work together to build open science and open technology. So Europa is tasked with the mission to build Europe's frontier open model. This is going to be a 1 trillion parameter plus model, equally representing over 24 languages and collaborating with 27 countries and working with their governments to source best -in-class data so that to make sure that this model is going to benefit all of Europe but also be an open alternative for the world.

18:19And who is in the consortium? So we're leading it. And the other member is Fraunhofer, a German university, which has always been at the forefront of AI science.

18:29John Thornhill:And are you going to include other members in this consortium? We're talking to other members. Might be in the future. Right. And how quickly do you think you could develop a one trillion parameter model? So we have a hard deadline, which is one year from July. But we believe we can pull it off even faster. So it will depend on the results and how satisfactory, you know, the different training jobs are going to result. But we're fully focused on it, so you can rest assured that this is coming earlier than one year. And it looks like, you know, people like Mistral have obviously were trying to get into the frontier model game and appear to be moving away from that because they realize it's very difficult to compete with these giant U.S.

19:12John Thornhill:companies with such enormous amounts of capital. So how are you going to compete? We have a different vision for deployed engineers and this idea that you need to decouple technology from implementation. First of all, we believe that implementation should be done by the customers themselves with external support, but not outsourcing their tacit knowledge. I mean, asking them to outsource their data through centralized models and outsource their tacit knowledge through external professional domain experts literally means them being fully outsourced. So they'll be just a design and engineering company and they would have all of their supply chain and value chain completely outsourced.

19:53This is unacceptable for the future of every economy. That's high risk. So the way we see it is that we can build vertically integrated AI systems that integrate anything from AI clouds or the chips and the infrastructure with the models and the application layer and hand it over to our clients so that they can customize them for their needs. And this is exactly what we're doing. So we have been measuring this as a KPI for the company before starting to scale and accelerate our go to market globally. So we started with projects that required six months to implement in 2024, came down to seven weeks in 2025.

20:33And as of today, we're able to implement an end-to-end mission-critical AI in just two weeks. And by the end of the year, we believe that's going down to one day. So this means that by focusing on a product-led deployment instead of a forward deployment engineer model, We're going to help our clients to implement this faster, bringing down the cost of implementation and having a faster time to value as well.

20:57John Thornhill:Who's in your team to develop this frontier model? We have a very strong team with people that have been doing this before. Our CTO, Luca Antiga, he was one of the initial contributors of the core of PyTorch, which is the framework behind the AI revolution. We have key people, for example, our AI cloud lead was leading high performance computing at ASML. Our domain experts as well that are leading the business and software solutions are coming from very reputable organizations. For example, for the financial services domain, we hired the head of investment AI at BlackRock and many other people from several tier one financial institutions joined him in the process of building that team.

21:44I can go on and on, but the team is very strong and coming from key positions in the industry globally to enable this market leadership.

21:52John Thornhill:And how much do you estimate it would cost to develop a frontier model nowadays? It's a minimum of 500 million euros and can go up as much as a billion euros for one version of it. Which raises the interesting issue of financing. you've been talking about wanting to raise about 10 billion dollars over the next couple of years or so in order to finance your ambitions how are you going to do that so first of all by being frugal and very diligent we believe that the future is old so anticipating time and trying to make these investments ahead of time can be an unprofitable decision. So the way we're doing it is by getting as much infrastructure and investment as we need for the goals that we have.

22:43Our goals right now are to continue build leadership in specific domains where mission-critical use cases are core. We mentioned finance, government, advanced engineering, defense, and we're getting traction from national security. So in these domains, we're integrating everything from the chip level up to the software interface. And we're even bringing domain experts to support that build out. So it's not just AI engineers, but for example, in finance is quants and people that were on the other side of the table as clients before and that understand exactly how to build these use cases. So that is not very capital intensive.