The words came out of Taiwan Semiconductor Manufacturing Co. in a form the industry rarely hears: an admission of limits. Asked how much more demand the world’s most important foundry could absorb, TSMC said, in effect, that it could only support so much. The company added that ramping its Arizona factories to full production could take a very long time. For a company that has spent two decades building every advanced chip on earth and never complaining about the volume, the statement was as close to a confession as the industry gets.
The admission arrived in the same week that ASML’s market value crossed $674 billion, making the Dutch company the most valuable corporation in European history. The two facts belong together. TSMC’s capacity is the bottleneck of the AI era — every advanced accelerator, every flagship smartphone processor, every data-center GPU must pass through its fabs. And every one of those fabs depends on ASML, the only company in the world that makes the extreme ultraviolet lithography machines required to print the most advanced chips. TSMC’s ceiling is, in a real sense, ASML’s floor.
The supply chain has a simple geometry. NVIDIA and its rivals design the chips. TSMC manufactures them, but only as many as its fabs can physically produce. ASML sells the machines that TSMC needs to add capacity, and those machines take years to build, ship and install. When demand grows at the pace the AI buildout demands, the constraint is not design talent or even money — it is the number of lithography systems in existence, the cleanrooms around them, and the time it takes to bring a new fab from groundbreaking to production.
TSMC’s Arizona experience illustrates the problem. The company’s first U.S. fab is operational and producing at yields comparable to its flagship Taiwan plants, but the broader expansion — a planned footprint of a dozen fabs and packaging facilities — will take years, and the company has acknowledged that construction-worker shortages and the sheer complexity of the process slow everything down. The second fab is still moving in equipment. The third is under construction. Each one is a multi-year project, and none of them arrive fast enough to satisfy the current demand curve.
The market has priced the imbalance accordingly. ASML’s rise to the top of the European market is a direct consequence of its monopoly position: every AI chip production plan on earth runs through its equipment, and the company’s order book is a proxy for the industry’s ambitions. The valuation says that the power in the chip supply chain has migrated upward, to the company that controls the machines rather than the companies that merely design what the machines produce.
The same week offered a reminder of how fragile that chain is. Broadcom’s earnings, though record-breaking, failed to meet the market’s inflated expectations, and the resulting selloff erased more than $1 trillion in chip-sector value. The crash was a valuation event, not a demand event — Broadcom’s own bookings reached three times its quarterly shipments — but it exposed the nervousness underneath: investors are not sure whether the industry’s capacity can ever catch up with its ambitions, or whether the ambitions themselves are the bubble.
TSMC’s admission cuts to the heart of that question. The company is not saying demand is weak; it is saying the opposite. It is saying that demand is so strong, and capacity so hard to expand, that the industry will remain supply-constrained for years. For chip buyers, that means scarcity pricing and allocation — TSMC deciding which customers get wafers. For investors, it means the equipment makers that can actually expand capacity — ASML above all — capture an ever-larger share of the industry’s value.
The geopolitical layer complicates the picture further. The United States has pushed TSMC to build in Arizona, and the company has responded with the largest foreign investment in its history, but the pace is set by physics, not policy. Every fab takes years, every machine takes years, and no amount of political will shortens the cycle. The export controls that Washington has imposed on advanced chips have their own effect: they divert demand, push China’s chipmakers onto older equipment, and concentrate even more of the world’s most advanced capacity in Taiwan’s existing fabs.
For ASML, the strategic question is how long the monopoly lasts. The company’s EUV machines are the product of decades of research and a near-impossible combination of optics, vacuums and precision engineering, and no credible rival has emerged. Its valuation assumes the position endures — and the week’s events support that assumption: as long as AI demand grows and TSMC cannot build fast enough, the company that makes the machines will keep collecting the rent.
TSMC’s rare admission and ASML’s historic valuation are two views of the same fact: the AI buildout has run into the physical world. Software can scale instantly; fabs cannot. The companies that own the physical gateways — the lithography systems, the cleanrooms, the power plants — hold the pricing power that the chip designers no longer have. Europe’s new market leader is a quiet machine-maker in the Netherlands, and the reason it now tops the continent is that the entire world’s most valuable industry cannot function without it.
