Late at night, scrolling through Coatue Management’s latest 13F filing, I almost rubbed my eyes in disbelief. The hedge fund famed for its surgical precision in tech growth bets had dramatically increased its stake in ASML while quietly trimming NVIDIA. This wasn’t just portfolio rebalancing—it was a strategic pivot whispered in footnotes. Philippe Laffont’s team, long hailed as apex predators in the AI gold rush, are now shifting capital from the chip designer to the machine maker. And that signals a truth the market has collectively overlooked: the AI feast is migrating from “chip design” to “manufacturing bedrock.”
For two years, Wall Street deified NVIDIA. $81.6 billion in capex, $150 billion committed to TSMC’s 3nm wafers, datacenter GPUs sold out for quarters—everything painted a myth of infinite compute expansion. But myths have limits. As TSMC quietly locks down 3nm capacity, Moore’s Law hits atomic walls, and only one company on Earth can ship EUV lithography tools, investors are waking up: the real bottleneck isn’t algorithms—it’s the beam of light etching circuits onto silicon.
ASML isn’t just another equipment vendor. It’s the gatekeeper of modern semiconductor civilization. Without its EUV systems, TSMC, Samsung, and Intel can’t produce 5nm chips, let alone 3nm or the next-gen High-NA EUV. In an era of AI-driven demand explosion, this near-monopoly has transformed ASML into the rarest kind of infrastructure asset. Coatue’s move reveals where smart money is re-pricing certainty: better to bet on whether ASML can keep delivering $200 million machines than whether NVIDIA can sustain Blackwell Ultra hype cycles.
Tellingly, European market sentiment recently found support precisely from ASML and Infineon. No coincidence. Infineon, the power semiconductor titan, anchors the electrification wave in EVs and industrial automation; ASML embodies the lifeline of advanced nodes. Together, they signal a broader shift: as AI froth risks peaking, capital flows toward assets with real capacity, technological moats, and irreplaceability.
Consider the valuation gap. NVIDIA trades above $2.2 trillion; ASML hovers around $300 billion. One rests on software ecosystems, CUDA lock-in, and market expectations; the other on physical reality—machines requiring tens of thousands of engineers, decade-long R&D cycles, and Dutch government approvals. When euphoria fades, which asset proves more antifragile? The answer writes itself.
Geopolitics adds another layer. U.S. export controls tighten by the month, yet ASML’s DUV tools remain mainland China’s only path to fab expansion. Even without EUV access, Chinese foundries are hoarding NXT:2000i systems. This quietly diversifies ASML’s revenue base—no longer solely dependent on TSMC or Intel, but woven into every crevice of the global supply chain. That decentralized resilience is something NVIDIA simply cannot replicate.
Meanwhile, NVIDIA faces the curse of success. Every major customer—Microsoft, Amazon, Google—is secretly building in-house AI chips to escape CUDA dependence. The so-called “anti-NVIDIA alliance” of AMD, ARM, and Marvell may not be formal, but fragmentation in inference chips is already real. When compute shifts from scarce to abundant, NVIDIA’s pricing power evaporates. Who will remember Blackwell then?
I judge the next 18 months as the inflection point in semiconductor investing logic. Capital will stop paying for “AI narratives” and start paying for “manufacturing capability.” Equipment giants like ASML, Applied Materials, and Tokyo Electron will become the core assets of the new cycle. If NVIDIA fails to prove its ecosystem is truly unassailable, a painful valuation reset awaits.
So here’s the final question: while everyone chases the flame of AI compute, who remembers the match that lit it? Perhaps real power no longer resides in GPUs—but in that silent EUV scanner standing in a cleanroom, shaping the future one photon at a time.