Aditya V Kashyap, AI and Innovation Leader, driving enterprise transformations through trusted strategy, governance and bold leadership.
getty
In January 2019, at TSMC's Fab 14B in Tainan, a batch of photoresist from a longtime chemical supplier arrived with a trace polymer that was subtly off specification. By the time the defect was detected, as many as 30,000 wafers at the 12- and 16-nanometer nodes had been scrapped, customers, including Nvidia and MediaTek, had to be renegotiated with and the company had taken an estimated $550 million off a single quarter's revenue. There was no cyberattack, no earthquake, no geopolitical event. It was a chemistry error at a supplier most people outside the industry had never heard of.
That incident is the most honest lens on the semiconductor supply chain. Inside large-scale systems, you learn quickly that fragility rarely sits where the headlines look. It lives in the adjacencies.
What rarely gets discussed in boardrooms is why this concentration is so stubborn. The temptation is to read it as a policy failure that industrial subsidies can reverse. In practice, the binding constraint is not capital. It is tacit knowledge.
A process engineer who has spent 15 years learning the idiosyncrasies of a specific deposition tool at a specific fab carries value no blueprint encodes. Qualifying a new source of excimer gas can take three to 18 months, because each fab's process is tuned to its current supplier's exact impurity profile.
An ASML EUV machine integrates components from more than 800 suppliers and depends on Zeiss SMT mirrors, produced in Oberkochen, whose layered optical coatings took 15 years to develop and are flat to fractions of a nanometer. A fab is not a building. It is a living organism of recipes, calibrations and supplier relationships compressed into a radius of perhaps 50 kilometers.
This is why the fragility is not accidental. It is the signature of a system optimized for efficiency across four decades. Inventory buffers were engineered out. Second sources were consolidated. Geographies specialized. By 2022, the median chip inventory held by U.S. manufacturers had fallen from 40 days in 2019 to fewer than five. Anyone who has designed resilient distributed systems knows the trade. You cannot have this level of precision and this level of redundancy simultaneously. What gave was the slack.
The quiet war now underway is the slow and expensive attempt to put some of that slack back. Industrial policy in Washington D.C., Brussels, Tokyo and Seoul has moved beyond the old mercantilist instinct to subsidize national champions. It is now about purchasing optionality.
Export controls issued by the U.S. Bureau of Industry and Security in October 2022, and tightened in October 2023, December 2024 and January 2025, were not primarily designed to deny a single country a single capability. They were designed to slow the clock on a catch-up curve that would otherwise compress decades of ecosystem-building into years. The December 2025 reopening of limited H200 exports to China under case-by-case licensing shows the policy is calibrated rather than absolute.
The insight that almost never gets articulated publicly is this: Semiconductor sovereignty, as it is usually imagined, may be structurally unattainable. Every serious attempt to build a fully domestic stack, whether in Phoenix, Dresden or Hefei, produces the same quiet discovery. The fab can be built. The surrounding ecosystem cannot be willed into existence on the same timeline.
Photoresists still flow from Japan. Optics still come from Oberkochen. Specialty gases still arrive from Korea and, until 2022, from Ukraine. Reshoring the fab reshores the flag, not the supply chain. What it tends to produce is a new dependency in a different geography.
The more honest framing is that the next phase will not be defined by dominance. It will be defined by resilience. Who can keep producing when a single node degrades? Who has visibility two and three tiers deep into their supplier map? Who has quietly qualified parallel pathways for critical materials? Who understands that trailing-edge nodes, the unglamorous 28- and 40-nanometer processes running automotive electronics and industrial controls, are where systemic shocks most often transmit?
Only about 8% of planned capacity expansions between 2021 and 2023 targeted nodes above 28 nanometers, despite these processes representing more than half of global chip demand. That underinvestment, not Taiwan, is where the next automotive shock will originate.
None of this is a tech story. It is the architecture of how modern civilization coordinates itself. The cleanroom in Hsinchu, the lithography bay in Veldhoven, the optics shop in Oberkochen and the specialty-gas plant outside Ulsan are nodes in a single distributed machine that the global economy has come to assume as ambient.
It works quietly, almost invisibly, until it does not. Fab 14B reminded us of that in 2019 for $550 million. Mariupol reminded us in 2022 at a considerably higher price. The institutions that plan for this rather than around it will shape the geometry of the next industrial decade. The rest will continue to describe the machine while others redesign it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。