"The compound interest effect of AI is not understood."

Milos Rusic and Malte Pietsch founded the start-up Deepset. The former TU Munich students deliberately chose Berlin as their location.
Deepset is considered one of the most promising Berlin start-ups in the field of artificial intelligence. Last year, investors, including the investment arm of Google's parent company Alphabet, invested 30 million dollars in the expert for large language models. The founders themselves are amazed at the momentum that the technology has generated in recent months.

Berliner Wirtschaft: You met during your studies at the Technical University of Munich. Why didn't you establish your company there, but in Berlin instead?
Milos Rusic: We actually discussed this question intensively. Malte was already living in Berlin at the end of 2017, when Deepset was set up, and I was living in Munich. The decisive factor for us was that the market for talent here in Berlin is simply more convincing. Above all, it's easier to find international talent here. Berlin is particularly attractive for English-speaking people because the city is simply more international.
Malte Pietsch: There was also already a small AI scene in Berlin. There weren't that many start-ups dealing with artificial intelligence back then. As a technology-oriented company, it was important for us to be in a location where we could exchange ideas with companies that do similar things to us.

What other arguments were in favor of Berlin and which were in favor of Munich?
Milos Rusic: Well, a strong argument in favor of Munich is TUM, the Technical University of Munich. It is a strong organization where technologies are very well supported. Munich is also much stronger as an industrial location than Berlin. In this respect, there is also more industrial research and industrial applications there. That's how Siemens became a customer of ours very early on.
Malte Pietsch: In Berlin, on the other hand, there was a strong AI open source community. This is good for us because we are convinced that there will be technology standards in our market and that no one will be able to avoid open source. Open source builds trust, creates independence and appeals to the masses of developers. We have already seen in other fields of technology that open source ultimately prevails.

Your location discussion took place more than six years ago. Has anything changed since then?
Malte Pietsch: Paris has developed enormously since then. We also looked at London back then, but didn't even think about Paris. I think Paris has developed so much because the government has really supported the industry and invested a lot. In addition, much more has happened in the San Francisco Bay Area with AI than was expected. The location is so relevant that we have to have people on site and we now do. There's a lot going on in that area right now.

And Berlin?
Malte Pietsch: I see Berlin at a crossroads right now. There is still a large ecosystem here, many companies working with AI, and the AI Campus in Gesundbrunnen. Things are already happening, but not at the same speed as in San Francisco or Paris.
Milos Rusic: Yes, definitely, Berlin is a good location for AI. But it could develop even better. You can tell that the ecosystem is still very much shaped by the tens and by e-commerce start-up founders. These founders are now involved as investors and have a major influence on the venture capital scene. That has also worked very well. But AI is something completely new. The business models have a completely different dynamic and the market is much more global.

What is happening in Paris or San Francisco that is not happening in Berlin? Or what does Berlin need to do to become as successful in AI as it is in e-commerce and the fintech scene?
Milos Rusic: I don't see a Berlin problem. I believe that any new technology sector within Germany will most likely cluster in Berlin. Berlin is the only true international cosmopolitan city in Germany where people come from all over the world. Berlin will always have the advantage that the talent comes here. There is more of a German problem.

What is that?
Milos Rusic: The following also applies to the AI sector: start-ups thrive on their technologies being used, only then they can develop further. In the USA, the will to try out and use new technologies is more pronounced. This is how new, leading industries are created there. In Germany, organizations are generally more cautious and less willing to take risks when it comes to using new technologies. If we really want to stay ahead, we need more demand - especially from the public sector.
Malte Pietsch: Once again, we are very strong in research in Germany. But we are not putting the results into practice. There really are a lot of research programs into which a lot of money is flowing. We have looked at a lot of them. But this interface between research and industry, which is actually so important, is so bureaucratic that we say: it's not worth it.

Why not?
Malte Pietsch: We would have to invest too much time in all the paperwork. We can make better use of this time by further developing our technology. In countries like the USA or France in particular, funding is much less complicated, as we have seen.

Is the importance of AI in Germany misjudged?
Milos Rusic: No, nobody doubts that AI is relevant anymore. Our impression is also that politics in both Germany and Europe understand that we need to work together to bring up our own champions. But it is uncertain which instruments can be used to achieve this. In my opinion, public procurement can be a very effective tool. It creates demand. The public administration would enter into contractual relationships with young companies, which can also be terminated again if nothing comes of it.
Malte Pietsch: But I'm not sure to what extent there is an understanding of how quickly this technology is now developing and how quickly it is changing the world. To be honest, I have to say that it's often astonishing to us how things are progressing. We've been doing Deepset for six years, but what's been happening in AI since the beginning of 2023 is insane, and it's continuing at this pace. This dynamic is probably difficult to perceive from the outside.
Milos Rusic: I don't think people understand the compound interest effect of AI. In the beginning, everything grows very slowly, and that suggests that you have a lot of time. But all of a sudden, the shallow growth turns into exponential growth. You can't let this moment pass, otherwise you won't be able to keep up. Perhaps this risk exists in Germany because we are a little too patient.

Perhaps this is a problem for SMEs in particular. You have acquired large corporations such as Airbus as customers for your product “Haystack”, which can compile documents and specific content from large amounts of data or set up chatbots, for example. Are your solutions only suitable for large companies?
Malte Pietsch: SMEs can also use AI. We also have smaller customers who only have a very small IT team.

What else do you want to do in Berlin?
Malte Pietsch: One of our next goals is to relocate. We will be moving in the fall and are already creating new premises in Mitte. That will be our headquarters. Our people who live in Berlin can meet and work there, and we also want those who are based at other locations to come by from time to time. We also want to make an active contribution to bringing the AI community here in Berlin closer together.
Milos Rusic: Deepset is not intended to work purely remotely. We want to be hybrid. There will be locations where people meet. Natural personal interaction is very important from time to time. However, Berlin should not just be our headquarters, but should always remain our largest location.

This interview was first published on Berliner Wirtschaft in German.