Europe, AI, and the game on the field | Part I

One of my favorite founders once described himself as a “micro pessimist but macro optimist.” He meant that while rigorous scrutiny can overshadow the short-term, the pursuit of transformative opportunities is inherently optimistic. I like that.

In this first post, I'll explore the landscape for AI-native B2B startups in Europe. I'll explain why Europe may not be ideal for building large foundational models or horizontal applications, and outline where European founders can leverage distinct advantages.

 "Our job is not to see the future, it's to see the present very clearly." — Matt Cohler

Less Capital, Less Power, More Regulation: Europe’s Frontier AI Formula  

On November 30th, 2022, ChatGPT was introduced to the public, challenging the Turing test and ushering in a new era of generative artificial intelligence. The technology is still in its early stages, and we have yet to fully understand its trajectory. 

There are indicators that upcoming foundational model releases will significantly improve advanced reasoning, mathematics, and multimodal capabilities. Some insiders estimate that artificial general intelligence may be achieved as soon as 2027.

Plainly put, the global future of large foundational models will not be written by European companies. Despite a talent pool comparable to the U.S., structural impediments make competing at the frontier very challenging. These include a stark disparity in available capital (U.S. 10:1 EU), data center capacity (U.S. 2:1 EU), and higher energy costs (U.S. 1:1.7 EU). Efforts to bridge this gap have been insufficient. In a race against a rocket car, upgrading from a bicycle to a scooter won’t put you ahead.

Meanwhile, European initiatives in and around AI have been predominantly focused on regulation and policy, exemplified by frameworks like the EU AI Act, GDPR, and prohibitive energy policies. Perhaps this reflects an implicit acknowledgment that if Europe is unlikely to push the AI frontier, it may seek to shield its markets through protective policies. For example, Meta withheld its multimodal models and Apple delayed its Apple Intelligence features from European release due to concerns over "unpredictable" regulation.

Presently, the GDP of the EU is around 78% of that of the U.S. (rising to about 91% when including the UK and Switzerland). It remains to be seen whether Europe’s regulatory approach will help narrow or widen this gap.

Horizontal AI-applications: No home court advantage for European founders

Given these structural challenges and Europe's regulatory focus, European AI-native startups face an uphill battle in both foundational models and applications with a low degree of specialisation. Particularly vulnerable are startups whose core IP revolves around generic RAG (retrieval augmented generation) and prompt engineering. These opportunities are being tackled directly by incumbents who are anything but asleep at the wheel. 

VCs often ask if the startup can figure out the distribution before incumbents can replicate its innovation. In this case, the challenge is twofold: companies like OpenAI, Microsoft, and Alphabet own the foundational models and build applications themselves, while enterprise giants like Salesforce can deploy those models through API calls, just as easily as any startup. Furthermore, it is possible that foundational model companies may choose to limit access to future frontier models and/or make the API calls for the most powerful models prohibitively expensive.

OpenAI in particular may leverage improvements in its upcoming models to push further into the application layer and create a lock-in effect. A case in point: OpenAI recently signed a c.$100 million contract with T-Mobile U.S. to co-develop advanced customer support chatbots. OpenAI has also established multiple European offices (London, Paris, Munich, Dublin, Brussels and Zurich).

The only way for startups to win in emerging horizontal categories is to achieve "escape velocity", gaining substantial market share before incumbents catch up. Perplexity, an emerging contender in AI-powered search, shows promise but hasn't yet secured the status of a standalone winner. Generating this kind of global momentum is more difficult from Europe due to slower customer adoption and higher regulatory hurdles. It is not impossible, just harder than elsewhere.

So, are AI-focused European startups facing a dead end? Not exactly.

McKinsey projects that Generative AI could boost Europe’s economy by over $500 billion in the next five years; 3 to 10 times the size of today’s European software market, depending on the definitions used. While it’s not a perfect one-to-one comparison, it signals tremendous growth potential. Even though incumbents will capture a significant portion of this expanding market, there will be emerging opportunities for which European startups are uniquely positioned to win. These opportunities won't be small or insignificant; they'll represent markets multiple times larger than Europe's current software sector and could fundamentally transform how Europeans work and live.

In my next post, I'll dive into the specific areas where European founders have a unique edge, leveraging factors like regulation, demographic shifts, and Europe's existing leadership in key industries. These advantages may prove decisive in the next chapter of AI innovation. See you soon!