On July 2, Jochen Hanebeck inaugurated the world's largest power semiconductor factory in Dresden: a five billion euro investment, approximately 1,000 new jobs, completed three months ahead of schedule. One day earlier, Germany had submitted 14 semiconductor projects to the EU Commission, the first EU member to do so. Behind these announcements lies a strategic paradox: Europe has yet to produce a competitive answer to Nvidia for the chips that train AI models.
What Dresden is manufacturing
Infineon's Smart Power Fab is a 300-millimeter wafer fabrication plant, claimed to be Europe's most modern and flexible of its kind. It produces power semiconductors: chips that do not perform calculations but control electrical current. They regulate battery output in electric vehicles, convert alternating current from wind and solar installations, and supply regulated power to servers in AI data centers. The facility operates without natural gas; approximately 90 percent of water used is recycled internally.
Of the five billion euros, roughly one billion comes from EU Chips Act funds, the first major project launched under this program in Germany. The data center business, supplying power semiconductors for AI server farms, grew at Infineon from 250 million euros in fiscal year 2024 to approximately 700 million euros in 2025. For 2026, the company projects 1.5 billion euros, rising to 2.5 billion euros in 2027. This segment currently represents about ten percent of total revenue and is growing faster than all other segments.
Managing power, not training AI
Infineon's strategic label for this market is "Powering AI." This is not metaphorical but descriptive: Infineon supplies chips that regulate electricity in data centers, not the chips on which AI models actually compute. No European company manufactures Nvidia-grade high-performance processors for large language model training. A modern AI data center consumes 100 to 500 megawatts. A large portion goes to power conversion and regulation, exactly where Infineon sits in the value chain.
Joachim Ragnitz from the Ifo Institute Dresden regards government subsidies as a "second-best solution." The crucial question, according to Ragnitz, is how Infineon will maintain production economically after subsidies expire. A structural gap is evident in a different metric: According to expert estimates, the United States installs four times more new data center capacity annually than Germany possesses in total. The gap is widening.
The IPCEI package that the Federal Ministry of Economic Affairs submitted to the EU Commission on July 1 targets a different part of this gap. The program, "Important Project of Common European Interest: Advanced Semiconductor Technologies" (IPCEI AST), comprises 35 German projects receiving 3.8 billion euros in federal funding with total investment volume of nearly ten billion euros. Coordinated jointly with the Netherlands and France, 17 EU member states participate. The Federal Ministry explicitly names "cutting-edge AI chips that significantly surpass current technology" as a project goal. This would be the first time Europe builds its own competitive base in this category.
Where Europe is actually irreplaceable
No three-nanometer chip in the world is manufactured without German components. Carl Zeiss SMT in Oberkochen supplies optics for extreme ultraviolet (EUV) lithography machines; Trumpf in Ditzingen supplies laser light sources for the same process. In both subsectors, these companies hold approximately 100 percent of global market share: TSMC in Taiwan, Samsung in South Korea, and Intel in the United States depend on these suppliers. If Zeiss or Trumpf fail, the world's high-performance chip production lines halt.
A second European avenue runs through alternative chip architectures. The Fraunhofer Institute for Applied Solid State Physics and the Fraunhofer Institute for Integrated Circuits are establishing the chip design center "FIZ Chip AI" in Heilbronn, funded by the Dieter Schwarz Foundation. The focus is neuromorphic processors operating with temporal signal pulses rather than continuous voltage levels. Fraunhofer IIS developed the SENNA chip as an initial reference architecture: 1,024 artificial neurons on less than eleven square millimeters, response times under 20 nanoseconds. These so-called Spiking Neural Networks are particularly energy-efficient and suited for AI applications directly on edge devices and in industry. They are not designed for training large language models with billions of matrix operations—that is a different task.
Dutch startup Axelera AI pursues a third avenue: commercial inference chips executing trained AI models in practice. In February 2026, the company raised 250 million U.S. dollars, led by Innovation Industries with BlackRock joining as a new investor, bringing total funding above 450 million dollars. CEO Fabrizio Del Maffeo announced the capital would fund mass production of the "Europa" chip, designed for energy-efficient industrial AI applications. Axelera targets the inference market, not the training chip sector.
Awaiting IPCEI approval in Brussels
The EU Commission must review Germany's 14 IPCEI AST prenotifications for state aid compliance and approval before funds flow. The predecessor IPCEI Microelectronics program required approximately one year from notification to approval, followed by another twelve to eighteen months before production launch. For IPCEI AST, this realistically means production beginning no earlier than 2028.
The question that Infineon, Fraunhofer, and Axelera collectively cannot answer is: Who in Europe builds the chips training the next generation of AI models? Nvidia currently controls over 80 percent of the AI training processor market, with AMD a distant follower. A European supplier is entirely absent in this category. IPCEI's goal offers hope this changes by decade's end. The true scale of this chip ambition within the 35 projects will only emerge when Brussels examines the applications.