ASML’s absence from The Motley Fool’s top 10 stock picks on June 12, 2026, does not diminish its pivotal role in the AI semiconductor ecosystem. On the contrary, the Dutch equipment giant sits at the epicenter of a global bottleneck: it alone manufactures extreme ultraviolet (EUV) lithography systems capable of patterning chips at 3nm and below—the very nodes powering NVIDIA’s Blackwell Ultra and Intel’s Gaudi 4 AI accelerators. In the triangle formed by chip design (NVIDIA), foundry capacity (Intel/TSMC), and manufacturing tools (ASML), ASML is not a supporting actor but the metronome setting the pace for the entire AI hardware race.
NVIDIA’s market cap has surged past $3 trillion, driven by insatiable demand for AI chips that pushes TSMC, Samsung, and Intel to expand advanced-node capacity. But expansion hinges on tool availability. ASML shipped just 62 EUV systems in 2025, including only five High-NA EUVs—the next-generation machines essential for sub-2nm logic. Even by 2027, ASML’s annual High-NA EUV output is projected to reach only 35 units. Each costs over $350 million and takes more than 18 months to deliver. This means NVIDIA cannot simply pay a premium to accelerate production of its GB200 NVL72 systems; physical constraints bind even the most capital-rich players.
Intel is attempting to break this bottleneck. Its Intel Foundry Services (IFS) strategy aims to become a U.S.-based alternative for AI chip manufacturing. In 2025, Intel secured a deal to produce select NVIDIA inference chips on its 18A node—a move designed not only to capture share from TSMC but also to qualify for priority access to U.S. CHIPS Act-funded equipment subsidies. Yet Intel’s 18A process still relies on ASML’s EUV tools, particularly the High-NA variant. Though Intel has pre-reserved multiple High-NA EUVs and plans to deploy them at its Arizona Fab 52 by late 2026, yield ramp remains at least two quarters behind TSMC’s N2P node. I judge that Intel will struggle to become NVIDIA’s primary supplier for high-end training chips before 2027 but may carve out a niche in inference and edge AI segments.
Geopolitics intensifies this bottleneck. U.S. export controls explicitly restrict sales of High-NA EUVs to China, and while conventional EUVs can still be sold to Taiwan, China and South Korea, the path to cutting-edge nodes is effectively blocked. This forces Samsung and SK Hynix to slow co-development of HBM4E and AI SoCs, while TSMC redirects High-NA capacity toward U.S. clients like NVIDIA and AMD. This “capacity reallocation” is policy-driven, not market-driven. Despite being a European company, ASML operates within a tightening web of U.S.-led tech restrictions, making its commercial decisions increasingly political.
Compounding the challenge, AI chip scaling is nearing physical limits. NVIDIA’s Blackwell uses TSMC’s 4NP process with transistor density exceeding 300 million per square millimeter. Achieving zettaFLOP-scale AI systems by 2028 demands 2nm or even 1.4nm nodes—entirely dependent on High-NA EUV adoption. Only three foundries—TSMC, Intel, and Samsung—are positioned to deploy these tools. With ASML producing just 35 High-NA EUVs annually, the next three years of advanced logic capacity will concentrate overwhelmingly in these three fabs. NVIDIA’s supply chain is far less diversified than it appears.
ASML’s omission from investment lists likely reflects short-term concerns over its rich valuation—its P/E ratio has hovered above 40 for years. Yet from an industrial perspective, it remains the most irreplaceable node in AI infrastructure. Without ASML, there are no sub-3nm AI chips; without those chips, the scaling laws underpinning large models hit a hard ceiling post-2027. The real risk isn’t whether ASML is a “buy”—it’s whether the AI industry has overestimated manufacturing elasticity.
When Jensen Huang declares “AI is the new electricity,” he overlooks a critical truth: power plants are built far slower than demand grows. ASML is the slowest, most expensive, and most essential power plant of all. The question is whether the global AI economy can tolerate its construction schedule.