Why Europe’s hidden AI advantage lies at the application layer

The adoption of AI will define the business and geopolitical challenges of this decade. First movers are gaining compounding advantages, widening the gap before other companies can react.

But is there an AI bubble? The answer hinges on one key question: how quickly can AI deliver real value through applications? To understand that, consider Nvidia’s strategy.

Jensen Huang uses a five-layer cake metaphor to describe today’s AI landscape: energy, chips, data centers, models, and applications. He argues that the application layer is the most important, yet still largely missing layer. Nvidia is currently making significant investments to support companies building this layer. Because how quickly this layer can deliver value will determine whether the bubble bursts or whether it fuels the most profound economic transformation ever.

The real question is whether the application layer can mature fast enough to create lasting economic value.

Energy is energy. AI chips are still largely application independent. Although specialized inference chips are emerging, GPUs remain the de facto general-purpose solution. Data centers can support a variety of workloads. The underlying model can be specialized, but it is very good, and sometimes unfortunately good, at performing general purpose tasks. Training a super-specialized base model with much less data rarely yields results.

As a result, the application layer is much more fragmented than other layers. And even with abundant capital and world-class researchers, it’s much harder to succeed by brute force. The application layer of AI cake is not uniform. It will be a patchwork of flavors.

Understanding your users this deeply can lead to important insights. The goal of AI in the workplace is not just to maximize what machines can do. It must be about enhancing what humans can achieve. In complex engineering environments, AI must not only accelerate execution, but also help experts consider more options, understand complex trade-offs, and make better-informed decisions.

However, most organizations struggle with this change. Although adoption has been widespread, meaningful impact remains limited, largely due to poor integration with human workflows.

Consider product engineering and design. AI systems will be able to handle an increasingly large part of the design process for complex systems such as cars, spacecraft, mobile phones, and nuclear power plants. This means dramatically increased throughput for engineering teams, an enhanced ability to explore new ideas and concepts, and an unprecedented acceleration of progress.

What is currently missing is a proper interface between humans and AI. We need to redefine the role of human engineers and develop appropriate interaction patterns for them to work effectively with AI systems. For engineers, these experiences are highly context-dependent.

This challenge is more complex than many observers assume. While anthropomorphic paradigms like “virtual employees” may be appealing, AI systems make different mistakes and react at different speeds than humans. True productivity comes from collaboration between humans and AI, where both bring their strengths to the table.

Discussions often focus on model competitiveness and large-scale infrastructure, with Europe facing well-documented gaps. But more decisive factors are emerging. It’s the quality of the human-AI interface at the application layer. This remains a widely unsolved problem that requires nuance, creativity, and deep understanding of the domain to solve.

Europe is undervalued for its expertise in this very area. Years of developing advanced software for complex technology domains has led to advanced capabilities in human-centric interfaces for high-stakes environments where AI needs to support, rather than override, expert judgment. In the field of product engineering and computer-aided design, two European companies, Dassault Systèmes and Siemens, were among the world leaders of the last great wave of digitalization. European companies are now once again gaining strong positions in several areas of engineering AI.

At the application layer, the usual handicap for European companies, which often have little start-up capital, may become less important. This segment is more fragmented, less capital-intensive, and less amenable to aggressive investment, which could provide exactly the kind of opportunities that European companies can compete for and win.

However, this opportunity does not come automatically. Europe needs to bring specialized AI-powered applications to market quickly, learn from real-world adoption patterns, and iterate constantly. Application layer winners are not just the companies with the largest models or the deepest infrastructure. They are the ones who best understand users, most effectively integrate them into real-world workflows, and design systems that truly extend human capabilities.

This requires a clear commitment to invest in human-AI collaboration, keep humans accountable, and reward better decision-making, not just faster execution.

So, will an AI bubble occur? Perhaps, after all, there is speculation, excess capital and exaggerated expectations. But the more important question is whether the application layer can transform this moment into lasting value. If that were possible, what looks like a bubble today might be remembered as a surge in investment that enabled a fundamental transformation of the economy.

If value is ultimately created there, Europe will have an important role to play.