For surface technology, this is not a distant digital issue. Coating processes are determined by a multitude of interacting parameters. Material, pretreatment, temperature, chemistry, film thickness, application, equipment condition, and numerous other influencing factors determine the outcome. Despite modern simulation and process monitoring, time-consuming empirical development work therefore remains indispensable in many areas.
This is precisely where AI opens up a new dimension.
The Fraunhofer IPA is currently combining physical coating simulation with machine learning to design spray coating processes more quickly and precisely. In the field of electrostatic coating, the institute sees the potential for up to a 40 percent reduction in paint consumption. Fraunhofer IFAM, in turn, is working to link process parameters and sensor signals using self-learning methods so that surface processes can detect deviations and self-adjust. What today still requires test series, expert knowledge, and a great deal of time could thus be calculated, predicted, and optimized to a much greater extent in the future. And that is precisely why the question of who has the most powerful AI is anything but an academic one for a technology-oriented industry.
$286 billion in one year
Europe has recognized the problem. Up to seven so-called AI “gigafactories” are set to be built. The EU aims to mobilize more than 30 billion euros in public and private investment for this purpose. Each of these facilities is expected to have more than 100,000 modern AI processors and enable the development of particularly powerful models. When viewed in an international context, however, these seemingly large sums are quickly put into perspective.
According to the Stanford AI Index, $285.9 billion in private capital flowed into American AI companies in 2025 alone—more than 23 times as much as in China. However, China’s actual investments are likely to be significantly higher, as state-directed funding is only partially reflected in these figures. The figures are therefore not directly comparable overall. Nevertheless, they illustrate the scale of the race. AI, however, requires not only capital but also enormous amounts of electricity.
AI Meets Europe’s Energy Problem
According to calculations by the International Energy Agency, grid issues could delay the connection of around 20 percent of the global data center capacity planned by 2030. At the same time, energy remains expensive in Europe. According to IEA calculations, energy-intensive industrial companies in the EU will pay, on average, more than twice as much for electricity in 2025 as comparable companies in the U.S. and about 50 percent more than in China.
As a result, the AI industry in Europe faces the same location challenges that the chemical, metal processing, and surface technology sectors have been discussing for years: energy prices, grid capacity, permits, and infrastructure. But even capital and energy describe only part of the problem.
What we call AI today is just the beginning
ChatGPT, Claude, Gemini, and DeepSeek are already astonishingly powerful. It would be naive to assume that we have already reached a plateau in development in any form. Rather, the Stanford AI Index 2026 shows just how rapidly the capabilities of leading systems are still evolving. On the particularly challenging “Humanity’s Last Exam” benchmark, Frontier models improved by about 30 percentage points within just one year. Tests that were actually supposed to remain challenging for years are, in some cases, already reaching their limits after just a few months. This is particularly crucial for industrial applications.
If AI is better able to analyze technical literature, plan test series, support simulations, identify relationships between process parameters, develop materials, or optimize designs in the future, the performance of the model used will directly become a productivity factor.
A better model then doesn’t just save a few minutes when writing an email. It can accelerate development processes. And this advantage applies simultaneously to thousands of developers, researchers, and engineers. Those who work with the more powerful tool can understand, develop, and optimize faster.
Usage Is Not the Same as Control
One could argue that Europe does not need to develop the best models itself. After all, companies today can simply use American—and, in some cases, Chinese—systems. This assumes that the most powerful models will remain available in the future. In June 2026, the U.S. government imposed export restrictions on two new Anthropic models for foreign nationals. Because Anthropic could not reliably verify the nationality of its users in real time, the models were initially shut down completely. The restriction was lifted again a short time later.
So this was not a permanent isolation of Europe. But the incident shows how quickly access to a particularly powerful AI system can become the subject of strategic decisions. And China is no longer a technological laggard. According to Stanford, American and Chinese models have swapped places at the top of the performance rankings several times since early 2025. In March 2026, the best American model was only a few percentage points ahead of the best Chinese one.
As a result, the development of the currently most powerful systems is increasingly concentrated in two major technology hubs.
What does this mean for surface technology?
European industry, in particular, still possesses something exceptionally valuable: process and application knowledge built up over decades. This is especially true for technologically demanding niches. An experienced coater knows that a process cannot be described solely by a few setpoints. He understands interactions, boundary conditions, material characteristics, and failure patterns that have emerged from experience over years or decades. This knowledge is a competitive advantage. However, ever-improving AI could make precisely these areas increasingly accessible at a faster pace.
If powerful models based on experimental data, simulations, and production data can identify correlations, calculate process windows, and propose new solutions, this may shorten the time required to build up experiential knowledge or get coating lines up and running. This would be both an opportunity and a risk for European industry. The opportunity lies in using AI to make existing know-how significantly more productive. The risk arises if competitors consistently work with more powerful systems. In that case, AI—of all things—could help to more quickly narrow technological gaps that previously took decades to build.
The real race has only just begun
The call—often made by policymakers—that Europe must not fall behind in AI implies that we are still among the leading nations in this field. In terms of investment, computing infrastructure, and cutting-edge models, there is already a significant gap between Europe and the U.S. At the same time, China has made significant technological strides. Europe is now attempting to build its own infrastructure—under conditions in which capital, energy prices, and networks already pose locational disadvantages.
There is little reason to believe that AI will not increasingly become a central tool for research, materials development, design, and industrial process optimization. Consequently, the quality of these systems will have an ever-greater impact on how quickly an economy can develop new technologies and translate them into value creation.
Thus, dependence on foreign AI means dependence on a tool used to develop the technologies of tomorrow. For Europe’s industry, the question in the not-too-distant future could therefore be how much competitiveness other nations are willing to grant it. It would probably be better not to let it come to that.