Christophe Fouquet, CEO of ASML, told TechCrunch in a fresh interview that no competitor is coming for the Dutch semiconductor equipment company’s position in extreme ultraviolet lithography, a statement of competitive confidence that is unusual in its directness but accurate in its substance: ASML manufactures the only machines capable of printing the smallest transistor features required for leading-edge AI chips, and the combination of physics, accumulated expertise, supply chain concentration, and capital requirements that would face any challenger makes this the most defensible monopoly in the technology industry and the narrowest bottleneck in the physical infrastructure that every AI model, data center, and startup compute budget ultimately depends on.

ASML’s EUV lithography machines produce light at 13.5-nanometer wavelengths generated by firing a laser at tin droplets 50,000 times per second, creating plasma that emits extreme ultraviolet light which is then directed through a series of mirrors polished to atomic smoothness onto silicon wafers coated with photosensitive material. The precision required at each step of this process is at the limits of what physics permits, and ASML has spent approximately 30 years and tens of billions of euros developing the manufacturing knowledge, supplier relationships, and operational expertise to produce these machines at a rate of roughly one to two units per week. Each EUV system contains approximately 100,000 components, requires a dedicated supply chain involving Carl Zeiss for the mirror optics and hundreds of specialised suppliers for other subsystems, and sells for approximately $200 million per unit for standard EUV and over $380 million for the next-generation High-NA EUV systems that will enable the next several generations of chip miniaturisation. TSMC, Intel, and Samsung collectively account for essentially all EUV purchases, and their ability to manufacture chips at 5-nanometer, 3-nanometer, and 2-nanometer process nodes, the geometries that Nvidia’s H100, H200, and B200 AI chips are built on, is directly dependent on ASML machine availability.

The moat comparison with Nvidia is worth making explicitly because it reveals something important about where value is concentrated in the AI supply chain. Nvidia’s competitive position rests on its GPU architecture, the CUDA software ecosystem that developers have written applications against for 15 years, and its networking hardware. These advantages are genuine and durable, but they are being challenged from multiple directions simultaneously: AMD’s MI300 series has demonstrated competitive performance on some AI training workloads, custom silicon from Google’s TPUs and Amazon’s Trainium are winning internal workloads that previously ran on Nvidia hardware, and the hyperscaler investment in proprietary AI chips represents a direct long-term challenge to Nvidia’s data center GPU dominance. None of these challengers will displace Nvidia quickly, but the challenge is visible and the competition is real. ASML faces no equivalent challenge because the barriers to building a competitive EUV system are not primarily financial. A company could commit $20 billion to building an ASML competitor and still not have a functioning EUV machine at the end of the program, because the knowledge required to make the machine work is distributed across ASML’s 44,000 employees, its supply chain relationships, and 30 years of operational learning that cannot be transferred by hiring engineers or acquiring patents. Fouquet’s “no one is coming for us” is not arrogance. It is a precise description of a competitive position that is protected by physics and accumulated human expertise rather than by capital alone.

The export control dimension has made ASML a geopolitical instrument in a way that amplifies its strategic importance beyond its commercial role. The Dutch government, under pressure from the United States, has restricted ASML from shipping EUV machines to China since 2019, and the restrictions have progressively tightened to include the Deep UV lithography machines that ASML had been shipping to Chinese customers for less advanced chip production. China’s semiconductor ambitions, embodied in SMIC, Hua Hong, and dozens of state-backed chip initiatives, are directly constrained by the absence of ASML equipment. The most advanced chips SMIC can produce without EUV access are at 7-nanometer equivalent geometries using multiple patterning techniques that add cost and time but cannot fully substitute for the resolution that EUV enables. China’s response has been to invest approximately $100 billion in domestic semiconductor equipment development through companies like SMEE, which is attempting to build domestic lithography equipment, but independent assessments suggest SMEE remains 10 to 15 years behind ASML’s current EUV capability and decades behind the High-NA EUV that ASML is now shipping. The export control regime has effectively frozen China’s access to leading-edge chip manufacturing at the current state, while the rest of the world’s foundry capacity continues to advance, creating a technology gap that widens with each new ASML machine generation that China cannot receive.

The foundry capex implications of ASML’s position affect every AI compute budget downstream. TSMC’s capital expenditure program for 2025 and 2026 is approximately $38 to $40 billion annually, and a substantial fraction of that investment goes to purchasing ASML equipment for its 2-nanometer and 1.4-nanometer process node buildout. When ASML’s production rate cannot keep pace with foundry demand, which has been the situation during periods of acute semiconductor shortage, the constraint propagates through the supply chain: TSMC cannot expand capacity as fast as its customers need, Nvidia cannot procure wafers at the volume its customers are ordering, and AI data center buildout proceeds more slowly than the hyperscaler capex commitments would otherwise allow. ASML has been increasing its EUV production rate and is investing in High-NA EUV manufacturing capacity, but the company’s production scale is constrained by its own supply chain, specifically the Carl Zeiss mirror assemblies that require years of lead time to procure and the limited number of qualified suppliers for other critical components. The delivery backlog for EUV systems is measured in years rather than months, and TSMC, Intel, and Samsung have pre-committed to machine orders years in advance to secure their place in the production queue.

For AI startups and investors, ASML’s position in the supply chain is the physical constraint that all other AI infrastructure analysis ultimately bottoms out at. The compute roadmaps that hyperscalers publish, the AI training run budgets that frontier labs announce, and the inference capacity projections that cloud providers use to justify their data center construction programs all assume that the chips required to execute those plans will be available on the timescales the plans require. That assumption is well-founded in the near term because TSMC, Samsung, and Intel have already ordered the ASML machines required to build the next two to three generations of AI chip capacity, and those orders are in various stages of manufacturing and delivery. The risk scenario that ASML’s centrality creates is not a sudden supply disruption but a gradual constraint: if AI demand grows faster than ASML can expand EUV production, the price and availability of leading-edge chip capacity becomes a binding constraint on AI development in a way that shifts competitive advantage toward the companies with the deepest foundry relationships and the largest advance purchase commitments. That is already the dynamic at the frontier: Nvidia’s competitive advantage is partly a reflection of its TSMC relationship and wafer allocation, which is itself a downstream expression of ASML’s machine production rate. Fouquet’s confidence is not just a CEO’s soundbite. It describes the company that sits closest to the physical foundation of the entire AI economy.

Also read: SubQ Claims a 12 Million Token Context Window on a Sub-Quadratic Architecture and the Claim Is Worth Taking Seriously Before Taking at Face Value • Meta Is Using AI to Infer Age From Height and Bone Structure and the Approach Will Define How Every Consumer Platform Handles Minor Detection • Google Released Gemma 4 With Multi-Token Prediction and the LocalLLaMA Reaction Tells You Exactly Why This Is More Than Another Model Drop