Europe Has the Machine. It Lacks the Power
Why does the continent that controls the semiconductor supply chain’s most critical bottleneck remain hostage to those who switch the chips on?
“Having the machine is not the same as controlling the system.”
In Veldhoven, in the south of the Netherlands, there is an unmarked building where the most critical machine in today’s digital economy is made. It weighs around 180 tonnes and is transported in dozens of containers. Its purpose sounds simple: to etch circuits at the atomic scale onto the surface of a silicon wafer, the process that turns purified sand into a chip. It is called extreme ultraviolet lithography, and only one company in the world has mastered it: ASML. Without it, there can be no industrial-scale production of leading-edge semiconductors. Not in Taiwan, South Korea or the United States.
Europe controls this bottleneck. And yet it remains far from the centre of the power that bottleneck helps to create.
This is Europe’s strategic contradiction in 2026. The continent holds an indispensable industrial asset but does not control the layers where value and power are now concentrated: AI chip design, computing at scale, hyperscale cloud providers (AWS, Azure and Google Cloud are all American) and sovereign computing infrastructure. In a world where artificial intelligence is no longer just software but has become economic, military and geopolitical infrastructure, this is not a marginal industrial weakness. It is a structural vulnerability.
I. Europe’s paradox
Europe is not irrelevant to semiconductors. ASML ended 2025 with €32.7 billion in sales, an order book of €38.8 billion and €4.7 billion in R&D spending (ASML, Q4 2025 results). It remains the only company in the world capable of commercial-scale EUV lithography. Infineon, NXP and STMicroelectronics dominate segments such as power electronics and automotive. IMEC, in Leuven, is the world’s most advanced semiconductor research centre. The problem is not a lack of expertise. It is the relative position of that expertise in the new hierarchy of technological power.
The supply chain underpinning advanced AI is organised differently today. Design of the most critical accelerators (including graphics processing units, high-bandwidth memory and inference accelerators) remains concentrated in the United States. Leading-edge manufacturing is still anchored in Asia. Global-scale cloud is dominated by American providers. And the most aggressive demand for computing is being absorbed by laboratories and hyperscalers Europe does not control. When NVIDIA founder Jensen Huang announced at GTC 2026 that cumulative demand for the company’s two latest architectures would reach one trillion dollars by the end of 2027, demand at scale was still being driven by non-European players. Europe has an indispensable tool, but it does not command the system that tool makes possible.
The roots of this go back to decisions made in the 1990s. While the United States and Asia invested in advanced manufacturing, Europe chose to specialise in niches such as automotive, power and sensors, and to outsource leading-edge production. The result was a capable but peripheral ecosystem: strong in segments the market came to reward less, and absent from those it began to contest as geopolitical weapons.
The European Chips Act, adopted in 2023 with the ambition of mobilising €43 billion and doubling Europe’s share to 20% by 2030, has not reversed this trajectory. In Special Report 12/2025, the European Court of Auditors was unequivocal: the target is “very unlikely to be achieved”. The European Court of Auditors reports an EU share of 9.8% of the global microchip value chain in 2022. SEMI Europe estimates €69 billion mobilised by October 2025 across R&D and manufacturing. Under the FOAK instrument, it identifies a concentration in advanced front-end manufacturing and gaps in support for other parts of the chain, including design, equipment and materials. In September 2025, all 27 Member States signed a joint declaration calling for a more ambitious Chips Act 2.0. When everyone calls for a strategic overhaul this soon, that in itself is a political diagnosis.
II. The landscape changed faster than European policy
Demand for AI computing accelerated so quickly that many of the assumptions underlying Europe’s debate became obsolete. The global semiconductor market reached $795.6 billion in 2025, growing 26% in a single year; WSTS forecasts $975 billion for 2026, bringing it close to the trillion-dollar mark. WSTS’s autumn 2025 forecast projected 37% growth in the logic segment that year. This is not a normal sector cycle. It is a reordering of the digital world’s industrial hierarchy.
The acceleration is also visible in infrastructure. At GTC 2026, NVIDIA unveiled the Vera Rubin platform and said it could cut inference cost per token by up to an order of magnitude compared with the previous generation. In his keynote, Jensen Huang put the capacity improvement for a large data centre at 350 times the level of technology from two years earlier (NVIDIA Newsroom, 16 March 2026). Microsoft announced that Azure was among the first hyperscalers preparing large-scale Vera Rubin deployments. More important than the precise metric is its strategic meaning: the competitive frontier is no longer just chip manufacturing. It is integrating energy, cooling, networking, software and capital at the right scale and at the right time.
At the same time, artificial intelligence has become infrastructure for state power. Anthropic, which held a $200 million Pentagon contract to integrate its model into classified networks, refused to remove restrictions on domestic surveillance and autonomous weapons. The Pentagon responded by designating it a “supply-chain risk”, applying to an American company for the first time a label previously reserved for entities linked to foreign adversaries. OpenAI signed an alternative agreement hours later, describing equivalent safeguards based on different legal mechanisms (NPR, CNN and Fortune, February and March 2026). Across the Pacific, China has not slowed down: DeepSeek matched the performance of the best American models at a fraction of the cost, while Huawei is expanding its Ascend line and Bernstein Research projects it will reach around 50% of China’s accelerator market in 2026. Export controls slowed China down, but did not stop it.
Europe is present in critical parts of this equation. It has equipment, research and engineering. What it lacks is the ability to turn these scattered assets into effective computing sovereignty. That gap between technical excellence and systemic power is Europe’s real problem.
III. Europe’s mistake is not technological. It is strategic.
Europe cannot replicate TSMC, build an NVIDIA or compete with China’s state-coordinated scale. Building a strategy on that illusion would waste time and resources. But a credible European path exists, provided the continent is willing to compete asymmetrically.
The first lever is to treat ASML not merely as an industrial champion but as a European strategic asset. ASML has more than 16,000 people in R&D and engineering; no country manufactures semiconductors below 7 nanometres without its machines. Yet the political framework governing this advantage remains fragmented. Export controls on semiconductor equipment to China are negotiated bilaterally by the Netherlands, the United States and Japan, outside the European Union’s institutional framework. The reason is structural: national security remains the exclusive competence of each Member State. This means the export policy for Europe’s most important technological asset is set in The Hague and Washington, not Brussels. If ASML is an instrument of European power, the policy governing it must be European.
The second lever is to accept that the future of semiconductors will not be decided solely by the race to the smallest node. It will also be shaped by systems integration, three-dimensional stacking, chiplets, advanced packaging and design software. IMEC leads global research in this field; CEA-Leti’s RESOLVE initiative brings European centres together across 15 technology lines. Europe has very few fabless companies, that is, companies that design chips without manufacturing them. Axelera AI, which secured €61.6 million from the EuroHPC DARE programme in 2025, is one of them. Without autonomy in design, including in electronic design automation software now dominated by three American companies, sovereignty will always be incomplete. The real test is not to manufacture the world’s fastest chip. It is to occupy positions without which no one can scale safely and predictably.
The third lever is physical infrastructure. Digital sovereignty depends on more than software or regulation. It depends on energy, location, connectivity and jurisdiction. Gaia-X fell short of its initial ambition, and European data centres continue to run on American platforms. The legal point is concrete: the CLOUD Act can give access to data held by providers subject to US jurisdiction, even when that data is stored outside the United States. FISA Section 702 operates under a different legal framework, concerning the collection of intelligence on non-US persons; nevertheless, the combination of European physical location and dependence on providers outside the EU leaves a real area of exposure. In June 2025, Microsoft France’s chief legal officer told the French Senate that he could not guarantee that French data would not be seized by US authorities. The strategic lesson is simple: data residency is not the same as data sovereignty.
The alternative is a federated network of European data centres powered by renewable energy and operating exclusively under EU law. The physical infrastructure for this is beginning to emerge: the Start Campus in Sines, with 1.2 gigawatts of capacity and 100% renewable energy, shows that it is possible to build at scale in Europe. Around 99% of intercontinental data traffic travels over submarine fibre-optic cables, and Sines is already a landing point for EllaLink, Equiano and 2Africa, with four new systems planned by 2026. Portugal met 71% of its electricity consumption from renewable sources in 2024, according to REN, one of Europe’s strongest performances and a rare strategic asset on the energy map of advanced economies, with a regulatory horizon spanning 2026 to 2029. The Nordic countries offer similar conditions. Europe has the physical assets. It lacks the political framework to turn them into sovereignty.
Finally, none of this works without people. Europe trains excellent engineers and then loses them. TSMC and Samsung actively recruit at European universities. Europe needs scholarships tied to staying on the continent, faster visas for engineers from outside the EU and tax conditions that make staying more attractive than crossing the Atlantic. If Europe cannot compete on factories, it must compete on talent. And to compete for talent, it must offer an ecosystem where research becomes companies, not just publications.
IV. This is where Portugal comes in
The question is no longer just what Europe has lost. It is where it can still win. This is where Portugal comes in: a country without a leading-edge semiconductor industry, but with a rare combination of assets that could become strategically important to Europe in the right context.
Sines is the first of these assets. Start Campus is developing a site with 1.2 gigawatts of capacity, 100% renewable energy, seawater cooling and LEED Gold certification. Microsoft’s announced $10 billion investment in AI infrastructure in Sines strengthened the site’s international credibility. In a Europe where the biggest constraint on new data centres is often energy and permitting, this is no detail. It is strategic geography.
Connectivity is the second asset. Portugal is one of the few European countries connected directly by submarine cable to South America, Africa and North America, and to Asia through multilateral systems that land on its territory. Google’s transatlantic Nuvem cable, expected to enter service in 2026, will strengthen this position. A country combining energy, land, regulatory stability and intercontinental connectivity can move from the periphery to becoming a platform.
The third asset is the possibility of positioning Portugal as a European computing platform under EU jurisdiction. At a time when data residency is not equivalent to data sovereignty, offering processing capacity under European law, powered by clean energy and connected by fibre to three continents, is a position few countries can match.
The fourth asset is the opportunity to attract research, not just property investment. Ireland did not become relevant in the 1990s by hosting factories. It became relevant by hosting decisions. Portugal can follow the same path: turn Sines from a container port into a port of intelligence. IST, FCUL and the University of Porto produce computer scientists of recognised quality, but most emigrate before the age of thirty. The difference between being a data warehouse and a centre of intelligence lies in the ability to retain talent, generate intellectual property and create companies that design technology rather than merely house it.
Portugal can no longer be described as a country without a national AI agenda. In January 2026, the Government launched the National Artificial Intelligence Agenda (ANIA), with more than €400 million through 2030. The problem, then, is no longer the absence of a plan. The question is whether that agenda will be delivered with enough scale, continuity and ambition to connect talent, infrastructure, applied research and private investment in a coherent strategy. There is a vast distance between having an agenda and holding a position. A country can host data centres and remain peripheral. It stops being peripheral only when it turns infrastructure into bargaining power, talent into intellectual property and location into centrality.
V. The sovereignty being decided now
Europe has a habit of producing good diagnoses and late responses. In semiconductors, that is particularly dangerous because the world is not waiting for the next European consultation to invest, build and scale. While Brussels prepares Chips Act 2.0, Jensen Huang has already announced the next architecture, Feynman, expected in 2028. While the 27 consult, Beijing scales up. While Europe debates, the United States militarises.
The window to build European computing sovereignty is not a decade. It is the next three to five years. After that, investment decisions already made by others will determine who controls AI infrastructure and who is left paying to use it.
Europe is not starting from zero. It starts with an asset no one else has. What it lacks is not basic technological capability; it is the political will to turn industrial advantage into strategic sovereignty. The machine already exists. What is still missing is the decision to govern the world it helps build.
Methodological note
This article draws on primary and institutional sources: ASML, Q4 2025 results (28 January 2026); WSTS, final 2025 data and Autumn 2025 Forecast; European Court of Auditors, Special Report 12/2025; SEMI Europe, Chips Act Report (November 2025); NVIDIA Newsroom and GTC 2026 keynote (16 March 2026); NPR, CNN, Fortune and MIT Technology Review coverage of the Anthropic-Pentagon case; Start Campus and Microsoft (Web Summit, November 2025); Bernstein Research projections for Huawei’s market share; REN data on renewable electricity consumption in Portugal; and Council of Ministers Resolution No. 2/2026 (ANIA). The cited 9.8% share concerns the global value chain in 2022, according to the European Court of Auditors. Vera Rubin performance figures are manufacturer projections and remain subject to confirmation in production.
Original bibliographic details (Portuguese)
Ensaio de Miguel Sousa Soares, publicado originalmente no LinkedIn a 26 de março de 2026. Esta página é o arquivo pessoal do autor.
The Nexperia Case: The Day Europe Woke Up to Economic Warfare
Miguel Sousa Soares is an economist and analyst of geoeconomics and artificial intelligence, with more than twenty years’ experience in strategy and corporate finance. He writes for Visão, Jornal de Notícias, Expresso and Correio da Manhã. He comments on artificial intelligence on NOW on Fridays.