The next constraint on AI expansion may not be money, models or demand. VanEck product manager Nick Frasse says the industry’s pace is being set by the slow, stubborn business of building chip factories, data centres and the machinery inside them. ASML, TSMC and memory suppliers sit at the centre of that bottleneck.
ASML Watch analysis
What happened
In a conversation with Phil Rosen, reported by TradingView and Benzinga, Frasse described data centres as “token factories” whose workloads are likely to grow as AI systems handle broader, more agentic tasks. His argument is that physical infrastructure cannot expand at software speed.
Frasse said building a semiconductor fabrication plant or a data centre typically takes two to three years. ASML is especially significant because it is the only company making the extreme ultraviolet lithography systems used to manufacture the most advanced chips. TSMC turns chip designs into silicon, while Micron and SK Hynix are increasingly important as high-bandwidth memory becomes a constraint on AI accelerators.
The supply chain is therefore less a neat race between rival chip brands than a relay in which every runner depends on the next one having enough capacity. Glamorous? No. Decisive? Increasingly so.
Why it matters
Frasse’s thesis is that construction times create an “inherent governor” on the AI boom. If new capacity takes years to arrive, demand cannot instantly produce an unlimited wave of usable compute, even when investors are throwing capital at the problem.
That could restrain overbuilding, but it also means shortages may persist. A chip designer can have orders, a cloud provider can have customers and an AI lab can have a promising model, yet the whole stack still waits for specialist equipment, packaging capacity and memory plants to catch up.
Our read
This is a useful corrective to AI coverage that treats compute as something summoned by purchase order. The physical layer is slower, more concentrated and rather less impressed by investor enthusiasm.
Frasse works for VanEck’s semiconductor ETF, so his view is also an investment perspective, not a neutral forecast. Still, the underlying timelines are the practical point. Watch the factories and the equipment queues, not only the model launch calendar.
What to watch
- Whether ASML and its specialist suppliers can expand equipment output fast enough for leading-edge demand.
- Whether TSMC’s advanced packaging capacity becomes a larger constraint than wafer production.
- Whether Micron and SK Hynix add high-bandwidth memory capacity before AI deployments hit a harder ceiling.
- Whether multi-year customer commitments soften the semiconductor cycle or merely postpone its next correction.
Discussion spark: If chip factories and specialist equipment take years to expand, does that physical bottleneck make the AI boom more sustainable, or simply make shortages and high prices last longer?
Sources and evidence
Independent WittyWires tracker for public updates about ASML. Not affiliated with or endorsed by ASML; this is not an official account.