Europe produces less than 10% of global semiconductors and depends on extra-European capacities for the manufacturing and packaging of many advanced chips, including those used in AI. The Chips Act provides for over 43 billion euros in public and policy investments until 2030, with expected effects on private investment; announcements of associated projects subsequently exceeded this amount. The future TSMC site in Dresden will produce chips at 12 to 28 nanometers starting in 2027: solid components for automotive and industry, significantly lagging behind state-of-the-art logic processes of 2 to 3 nm, where architecture and packaging also matter. This gap results from industrial and political factors whose implications deserve examination without fatalism or illusion.

The Essentials

  • Europe controls industrial semiconductors and remains dependent for advanced AI chips. Its policy aims to reduce this dependence; it has not yet succeeded.
  • The EU accounts for 8.8% of global revenue from the value chain in 2025. The EU does not yet have a large open commercial foundry comparable to TSMC for the most advanced nodes; however, it does have advanced Intel 4/Intel 3 production in Leixlip, Ireland. A significant share of leading advanced AI chip designers is American, but advanced designs do not come exclusively from the United States; manufacturing is concentrated in Taiwan and South Korea. The only 3 nm class capacity on its soil, the Intel plant in Leixlip, Ireland, is primarily intended for Intel products, but Intel also plans to provide capacity there to Intel Foundry customers (COM(2026)504; Council EU 10094/2026).
  • TSMC Dresden, operational in 2027 in 12-28 nm technology, strengthens the European automotive and industrial sector but does not fill the gap in advanced AI chips.
  • The tension between technological sovereignty and the cost of complete decoupling structures the debate. Entering advanced nodes requires very substantial investments. No primary source establishes that member states refuse to commit to them.
  • By 2035, the question of whether Europe can maintain this intermediate position without being gradually excluded from AI value chains remains open.

Europe Weighs Less than 10% of the Global Market, and the Trend Is Declining

In 2000, Europe produced a significant share of global semiconductors. In 2025, this share has fallen below 10%. In the baseline scenario, the Commission projects approximately 9.6% for the global semiconductor segment in 2030; IDM and fabless shares are presented separately and should not be added or confused. The political objective remains 20%. The financing mobilized is not necessarily sufficient to reverse a long-term trajectory.

This decline says something precise about the nature of the sector. Manufacturing competitive semiconductors requires colossal cumulative investments, a workforce of engineers trained on the latest equipment, and ecosystems of subcontractors that cannot be reconstituted in a few years. TSMC, Samsung, and Intel have invested massively for several decades to reach the 3 nm node. Europe must still mobilize cumulative investments in advanced nodes, and the necessary financing capacities remain to be assembled.

Proposed in 2022, the European Chips Act was adopted in September 2023; it aims at the resilience of the entire ecosystem, including chips intended for the automotive and industrial sectors, as well as advanced technologies and capacities. The first objective is on track. The objective of advanced capacities continues to be pursued, but the necessary capacities and financing remain to be built. It is in this gap between initial ambition and likely outcome that the entire strategic question lies.

TSMC Dresden: A Real Advance for Industry, Insufficient for AI

The Dresden site is presented as the symbol of European industrial renewal, and it is, within a precise scope. TSMC is investing there with Infineon, NXP, and Bosch in a joint structure called European Semiconductor Manufacturing Company. Production will begin in 2027 using 12 to 28 nanometer processes, technologies mastered for several years but that exactly meet the needs of electric vehicles, industrial control systems, and embedded electronics. For these uses, node fineness matters less than reliability and availability.

The EU imports a significant share of certain components and depends notably on Asian facilities for several manufacturing and assembly stages; the semiconductor crisis of 2021-2022 severely disrupted and sometimes interrupted European automotive production, without demonstrating a general and permanent paralysis of all factories. TSMC Dresden reduces part of European dependence for 12 to 28 nm automotive and industrial chips, but not for advanced AI chips under 5 nm. Document 10094/26 ADD 6 confirms European vulnerabilities and the need for investments, but does not provide data establishing significant improvement specifically for 12-28 nm nodes.

But there is a chasm between 12 nm and 3 nm. Recent accelerators used for large models, such as the NVIDIA H100, are often manufactured on nodes under 5 nm, but the NVIDIA A100 is manufactured at 7 nm. Advanced logic capacities are highly concentrated at TSMC, Samsung, and Intel, primarily in Taiwan, South Korea, and the United States. Since 2026, the EU has supported an advanced pilot line at imec, oriented toward technologies beyond 2 nm; this line remains a pilot, without mass commercial production.

For further discussion of competitive dynamics in global semiconductors, the semiconductor battle against a backdrop of massive public subsidies offers a useful overview of ongoing investments in the United States and Asia.

A Strategic Choice, Not a Default Failure

The tension between technological sovereignty and the cost of decoupling is evident in the analysis that Mathieu Duchâtel develops on Sino-American bipolarity and its implications for Europe. In his work on Xi Jinping’s ambitions and the return of bloc rivalry, Duchâtel emphasizes that Europe is constrained to define its own conception of economic security, between a complete decoupling that is too costly and maintained dependence that is too risky. TSMC Dresden illustrates this tension in a concrete industrial decision: the site strengthens European capacities for mature nodes, while the cost of decoupling in the most advanced nodes remains difficult to establish.

This positioning has a logic. The 12-28 nm nodes remain important for many automotive and industrial applications. Securing them on European soil addresses a real and immediate need. Developing advanced nodes requires considerable investments, talent, and an ecosystem, but policy options are not necessarily limited to head-to-head rivalry with existing leaders.

The complementary reading by Thomas Philippon on market concentration and conditions for industrial competition sheds light on another angle. Philippon has shown that American markets have become structurally less competitive than European markets in many sectors, which paradoxically creates rents that entrants could contest with the right tools. The concentration of advanced capacities increases systemic risk; it can justify diversification investments, but does not mean that alternative capacities can be created quickly. However, the financial, technological, industrial, and human barriers to entry in advanced nodes are extremely high, and the concentration argument alone is insufficient to justify entry.

The issue, therefore, is not whether Europe was right not to target the 3 nm node now. The issue is whether the intermediate position it is consolidating leaves it sufficient leverage to avoid marginalization in the AI value chain over the next decade.

Dependence for Advanced AI Is Not Symmetric Between Allies

Two types of dependence must be distinguished. Dependence on democratic allies such as the United States, TSMC Taiwan, or South Korea involves manageable political risks within an Atlantic cooperation framework. Dependence on actors subject to increasing geopolitical pressure, notably Taiwan at the center of Sino-American rivalry, is of a different nature.

TSMC holds a preponderant position in global production of the most advanced nodes. A major and lasting disruption of Taiwanese production would probably have strong global consequences for advanced AI chip supply, and Europe would be highly exposed to it, with very limited local substitution capacities while non-EU alternatives would themselves remain constrained. The United States responded to this risk with the CHIPS and Science Act, which massively subsidizes the construction of advanced fabs on its soil. The EU has the Chips Act and a proposed Chips Act 2.0 aimed notably at advanced chips, but its level of financing and industrial maturity differ from the American program.

This asymmetry in managing geopolitical risk is documented in COM(2026)504: the Commission emphasizes the fragility of the European ecosystem, vulnerabilities of the user industry, and dependencies on third countries; it does not formulate this conclusion with the expression “systemic risk.” The document identifies strategic partnerships with allies as a means of reducing risk, which can reduce or diversify certain dependencies without guaranteeing complete autonomy by themselves. It is an honest but limited response, which transfers part of risk management toward diplomacy and international cooperation rather than toward its own industrial capacity.

The question of digital sovereignty is all the more critical because it touches on encryption and security infrastructures: whoever adopts post-quantum encryption first will control information shows that cascading technological dependence also concerns the software and cryptographic layers that rely on these same chips.

Design and Equipment: Europe’s Industrial Scope

Advanced manufacturing dependence masks a more nuanced reality in two adjacent segments. Europe is the global leader in lithography equipment: ASML, based in Veldhoven, Netherlands, is the sole supplier of EUV lithography systems, used in the manufacturing of the most advanced chips, including sub-5 nm generations. Without ASML’s EUV machines, manufacturing state-of-the-art logic chips for TSMC, Samsung, and Intel would be severely compromised. This position constitutes geopolitical leverage: since September 1, 2023, the Netherlands has strengthened export controls on advanced equipment, notably certain DUV systems; Chinese access to EUV systems was already blocked by licensing restrictions.

Chip design is also a European strength. ARM, whose architectures equip virtually all mobile processors and a growing share of AI servers, is a British company now listed in New York but whose engineering remains largely European. Infineon, NXP, and STMicroelectronics are among the world’s most performing specialized chip designers for embedded applications.

This portfolio of competencies changes the equation. Europe is not simply a passive consumer of technology: it is an indispensable link in the global supply chain, upstream of manufacturing. ASML produces approximately 200 EUV machines per year, and each machine is worth 200 million euros. Restricting or conditioning access to this equipment is a tool of industrial and geopolitical policy that Brussels is only beginning to fully mobilize.

Division of Labor by 2035

The question by 2035 is precise: can Europe maintain a viable industrial position in semiconductors by controlling the mature segment and remaining an importer of advanced AI chips, without being gradually excluded from the value chains where AI is integrated.

In the BAU scenario, Europe increases its capacity but remains primarily specialized in >28 nm nodes and mature technologies; capacity ≤28 nm remains limited. TSMC/ESMC Dresden must provide 28/22 and 16/12 nm capacities; STMicroelectronics projects in France and Italy strengthen mainly FD-SOI and SiC segments. Intel had announced then postponed its Magdeburg project; according to the Commission, it was subsequently withdrawn, while Intel’s European investments retained in the BAU scenario notably concern Leixlip, Ireland. New capacities can cover some of the automotive, industrial, and energy needs, but they do not guarantee reliable and complete domestic supply, notably for defense and chains dependent on packaging or imported components.

European dependence on advanced AI chips stems primarily from the absence of open European capacities for advanced design and manufacturing, facing suppliers and capacities that are predominantly American, Taiwanese, and South Korean. This scenario is satisfactory if Sino-American rivalry remains contained and Taiwan remains accessible. It becomes precarious if one of these conditions changes.

A second scenario assumes that the AI race accelerates demand for advanced chips to the point of changing the structure of industrial value-added itself. In that case, companies that do not control their AI accelerator supply could find themselves disadvantaged in financial services, healthcare, energy, and logistics. AI becomes an infrastructure for general productivity, beyond military or prestige applications alone. European dependence on advanced AI chips could then increase competitive constraints, and the choice not to target the 3 nm node could become more costly.

A third scenario, more uncertain but worth monitoring, is that of a technological breakthrough that reshuffles the deck. Neuromorphic architectures, photonic chips, or in-memory computing processors could reduce the advantage of sub-5 nm nodes for certain AI applications. Europe is investing in several of these directions, notably through Horizon Europe programs and the EIC microelectronics initiative. If one of these technologies reaches commercial maturity before 2035, the current hierarchy between advanced nodes and mature nodes could be partially disrupted. It is a long-term technological bet, not a certainty, and formulating it as such is the only possible intellectual honesty.

Signals to follow to distinguish these trajectories are identifiable. The evolution of the share of advanced chips in the value-added of European industrial sectors is the first indicator. The dynamics of R&D investments by European companies in alternative architectures is a second. The robustness of the transatlantic partnership on export controls of critical equipment, ASML foremost, constitutes a third. And Europe’s capacity to train sufficient microelectronics engineers, an often underestimated bottleneck, remains a fundamental determinant.

What comparative analysis of industrial capture strategies shows, notably through the lens of sectors like solar or batteries where China has followed a similar logic of progressive upgrading, is that intermediate positions are rarely stable over ten years. They either rise or fall.

Partial Sovereignty as Explicit Doctrine and Its Limits

The EU combines international partnerships with the objective of developing its own industrial capacities, including in advanced technologies. Producing mature chips on its soil, relying on ASML as leverage in the global supply chain, and developing industrial capacities for advanced nodes while resorting to diplomacy and partnerships. This orientation depends on available financing and the capacities of the European industrial ecosystem.

This choice deserves periodic reassessment. The European Chips Act will be subject to review before 2028. TSMC Dresden’s production data, early returns on the integration of European chips in automotive and industrial chains, and the evolution of AI demand in civil sectors will then provide a clearer picture of what partial sovereignty actually enables. The doctrine will confront results, and necessary adjustments can be decided on the basis of actual data rather than projections.

Sources

  1. European Commission COM(2026) 504 – European Chips Act Review (cite without link: European Commission, COM(2026) 504)
  2. Council of the EU, document 10094/2026 COMPET.1
  3. Deloitte, Semiconductor Outlook 2026 (Deloitte TMT Predictions 2026, no verified stable URL link)
  4. GlobX, European Chips Act Analysis, May 2026 (no verified stable URL link)
  5. The European Perspective, May 2026 (no verified stable URL link)
  6. Mathieu Duchâtel, China: Xi Jinping’s Ambition and the Return of Bipolarity, in Geopolitics and Geoeconomics of the Contemporary World, https://shs.cairn.info/geopolitique-et-geoeconomie-du-monde-contemporain–9782348070037-page-546?lang=fr