At Rise of AI Summit 2020 in Berlin, Adrian Locher, co-founder and CEO of Merantix, argues that Germany and Europe already have the ingredients for AI success but consistently fail to assemble them. He says Germany has more AI researchers per capita than the US and China combined, strong hardware and industrial players, and access to top talent, yet fails to transfer research into software companies. Rather than presenting a strategy, Locher describes what it takes to actually build AI products: handling messy edge-case data, keeping pace with fast-moving research, coping with long B2B sales cycles, and accepting that full automation is unrealistic. He details how Merantix, founded in 2016 as a venture studio, has built five companies, recruits from over 50 applicants per hire, runs internal knowledge-sharing formats, raised a 25 million fund backed by US institutional investors, and is opening a 5,000 square meter Berlin campus for over 450 people in April 2021. In the Q&A he stresses that teams, not technology or markets, determine whether AI ventures succeed or fail.
Key takeaways
- Germany and Europe have more AI researchers per capita than the US and China combined, but fail to transfer research into commercial products.
- Merantix was founded in 2016 as a venture studio and has built five companies, targeting ten more within four years.
- Full automation is unrealistic in practice: Merantix's medical imaging company automates about half of cases and still routes the rest to a doctor.
- Merantix raised a 25 million fund backed by major US institutional investors after self-funding its first three companies.
- Merantix is opening a 5,000 square meter campus in Berlin in April 2021 for over 450 people across 20 to 30 teams.
Detailed notes
Germany has the ingredients but fails at transfer
Locher opens by citing numbers he calls encouraging: on a per capita basis, Germany, and Europe more broadly, has more AI researchers than the US and China combined, top research institutions, strong hardware and manufacturing industry, and access to excellent talent. The recurring problem, he says, is transfer: companies with strong hardware fail to become strong software players, and research does not translate into investment or commercial products at the pace seen elsewhere.
Why Merantix builds rather than plans
Merantix, founded in 2016 as what Locher calls the world's first AI venture studio, has built five companies in medical imaging, biotech, data infrastructure and business intelligence, with two or three more planned for the following year and a target of ten new companies within four years. Locher frames the studio's approach as prioritizing execution over strategizing: assembling teams, validating ideas, creating traction and building products with real market value.
The practical problems of building AI products
Locher lists recurring obstacles: AI applications need large volumes of high-quality data, and the edge cases that make products work well usually require data that does not already exist and must be created. Research moves so fast that findings from two or three years ago are often obsolete, citing natural language processing's rapid progress after vision was largely solved around the ImageNet moment roughly eight years earlier. B2B sales cycles of six to twelve months create a mismatch with this pace. He also argues machine learning models are a general-purpose technology whose infrastructure is being rapidly commoditized, as shown by TensorFlow and PyTorch replacing the need to build custom training infrastructure, so the technology itself does not provide a durable moat. Full automation is unrealistic: Merantix's medical imaging company, working on breast cancer screening, automates about half of cases while a doctor still reviews the rest, and self-driving cars similarly remain far from level five autonomy.
Assembling the ecosystem at Merantix
Locher's central message is that success requires assembling the right ecosystem rather than more planning. At Merantix this means recruiting selectively, drawing dozens of applications daily and hiring roughly one in fifty candidates for both technical skill and social fit, since AI problems are multidisciplinary and require diverse teams that can collaborate. The studio runs weekly paper-reading groups and deep dives to circulate know-how, was a founding partner of the German AI Association, funded its first three companies from its own balance sheet, and this year raised a 25 million fund backed by major US institutional investors. It runs several EU-funded projects with partners including Fraunhofer and Max Planck, and its portfolio companies share data, contracts and knowledge under an open research agreement. Merantix is opening a 5,000 square meter campus near Nordbahnhof in April 2021, housing over 450 people across 20 to 30 teams including startups, corporate ventures, corporate machine learning teams, academia and investors.
Q&A highlights
Asked how companies without Merantix's reach can attract talent and capital, Locher says it starts with the initial team and a hiring bar that never drops. Asked why Berlin over Zurich or Silicon Valley, he says Berlin attracts people from all over the world and offers a much lower cost of living, with team members from over 30 nationalities and 80 percent having relocated to join. Pushed on whether putting ingredients together is itself the hard problem, he agrees it is incredibly hard but says Europe has every factor needed and must be bolder. Asked for one policy recommendation for Germany's AI future, he says he would make AI a national priority, invest in research and its transfer and application, invest in early technology education, set up varied investment vehicles, and let government agencies fail visibly and make that acceptable. On why Merantix hires for social skills as well as talent, he says in his experience it is always the team, not the technology or the market, that fails first when a company struggles.
Quotes
We have all the ingredients and we failed to put them together.
Adrian Locher
Smart people want to work with other smart people and they want to have a nice environment, they don't need too many perks.
Adrian Locher
It's not technology, it's not the market, it's not the product that fails first, it's always the team that at some point starts to break.
Adrian Locher
Questions answered in this talk
What is the main obstacle for Germany and Europe in turning AI research into successful companies?
According to Adrian Locher at Rise of AI Summit 2020, Germany and Europe have more AI researchers per capita than the US and China combined, strong hardware and industry players, and access to top talent, but consistently fail to transfer this into strong software companies and commercial products.
Why can AI products not reach full automation?
Adrian Locher explains at Rise of AI Summit 2020 that even a successful product like Merantix's medical imaging company automates only about half of cases, with the rest routed to a doctor, because edge cases make full automation unrealistic, similar to how self-driving cars remain far from level five autonomy.
Why did Merantix choose Berlin over Zurich or Silicon Valley?
According to Adrian Locher, Merantix chose Berlin because it can attract talent from all over the world and has a much lower cost of living than San Francisco, and the company now has team members from more than 30 nations, with about 80 percent having relocated to Berlin to join.