DeepSeek’s groundbreaking achievement has captured global attention, seemingly training a competitive open-source model for hundreds of millions of dollars less than equivalent US foundational models, potentially with lower-powered chips (although this is still being evaluated) and certainly far less compute than expected. This success underscores a broader opportunity for Europe in the AI landscape.
“Europe is behind in the AI race – and cannot catch up.”
This misperception is commonly held and causes great consternation for Europe’s political leaders, as the US surges ahead with buoyant capital markets, rising valuations, job creation, and wage growth. Yet our early-stage venture capital firm invests in technology startups in both the US and Europe – and we take a different view. We think structural conditions mean that Europe will be highly competitive when it comes to building great applied AI businesses in the years to come.
DeepSeek’s achievement points to an inevitable future where functional models capable of supporting real-world applications won’t require the vast capex investments seen to date. While we continue to believe that high-end infrastructure will still be necessary for top-tier models, given the success of DeepSeek the performance bar for justifying such investments will rise, favouring efficient and innovative applications of AI.
Europe has been a global leader in the application of technology to numerous verticals including financial services (Revolut, Klarna), drug discovery (Exscientia, BenevolentAI), music (Spotify, Shazam), social commerce (Auto1, Vinted), and many other verticals, and has proved in past cycles to be able to build competitive advantages in these spaces by leveraging the latest technology trends.
We see this same pattern in AI, where AI will enable those existing verticals where Europe has a strategic advantage to pull even further ahead. In our portfolio, we see this with Incode, an AI-driven biometric authentication company where the core AI development team is based in Eastern Europe, enabling Incode to surpass legacy fintech companies in biometric authentication. We see this with German agentic AI startup Cognigy, which has rapidly become the Gartner Magic Quadrant leader in using AI for call centres, as well as with Hawk, a German startup deploying applied AI to improve AML and KYC procedures for banks, and with German startup Hypatos, which uses AI to process documents and automate procedures for accounting. Many of our recent application layer AI investments have been in Europe, and our overall mix of AI companies is roughly 50/50 between US and EU teams.
The reason Europe can leverage AI is because of the amazing combination of both engineers and go-to-market commercial leaders. We have seen this at DN Capital over many years with the founding team at companies like Auto1 and Shazam. And while a smaller venture ecosystem is typically seen as a disadvantage, it does carry a number of distinct advantages. For one, start-ups seeking great AI engineers do not have to readily compete with multi-million dollar cash and equity packages offered from headquarter operations at Meta, Google, Amazon, or Microsoft, or from foundational model players like OpenAI, X.ai and Anthropic. It also means that entrepreneurs are used to having to generate better returns with less capital than in the US, and so have in their DNA to run businesses more efficiently.
The announcement that China’s DeepSeek has been able to train its model at a fraction of the cost of similar performant US models also highlights Europe’s opportunity. European startups’ need for stricter capital discipline due to lower investment flows means they will view leveraging efficient open-source AI models as a significant advantage, whereas US founders awash with investment will see less relative benefit.
DeepSeek also demonstrates two key trends. First, the costs for building certain types of models will continue to fall dramatically, maybe even at the pace that we have seen for the mapping of the genome (a 1000x decline in cost in roughly a decade). Second, to justify the price points needed to generate a return on investment for top-end infrastructure, the level of performance coming out of cutting-edge models will have to accelerate at an even greater rate than in 2020-2024. These shifts greatly benefit the European tech ecosystem, where less expensive talent and disciplined investment models enable startups to thrive.
We remain convinced that building the next generation of foundational models will still require substantial compute and energy. Without significant changes to Europe’s energy policies including more nuclear investment and shifting LNG imports from the East to come from the States, as well as dramatic increases in geothermal and other clean energy sources, Europe’s higher costs relative to other geographics are likely to be a barrier at this foundational model layer.
However, at the application layer, Europe’s strength in applied AI is significantly less affected. While a challenge going forward could be if EU regulations on data privacy could complicate matters if companies must rely on energy-intensive native models stored within the EU, we believe collaborative discussions between the government and the venture industry can find solutions here over time.
We are confident Europe will be able to use AI to pull ahead in its other leading verticals. Fundamentally, DeepSeek shows that when entrepreneurs are capital-constrained or technology platform-constrained but have raw intellect and ambition, they can do amazing things.
Agentic AI is transforming customer experience & driving growth