The phone in your hand, the car on your driveway, and the AI that answered your last question all run on chips from one island, and if that island ever goes offline, the recovery will be measured in years, not weeks.

In June 2026, tanks moved through the streets of Taoyuan, a Taiwanese city wrapped around one of the busiest airports on Earth. The army described the movement as a readiness drill. In the same few days, China sent its newest aircraft carrier through the Taiwan Strait, and Britain, France and Germany issued a rare joint warning about Chinese vessels crowding foreign ships operating on the Pacific side of the island. For Taiwan, on the accumulated evidence of the past several years, that pattern of activity is close to an ordinary month.

What is not ordinary is the specific object sitting at the far end of the supply chain the island supports. The chips that make modern artificial intelligence possible at all come overwhelmingly from a small number of buildings inside a single country. There is no second source anywhere on Earth for the most advanced processes, no back-up island, and no plan that would restore operations in less than several years if the primary site went offline.

One company, twelve thousand designs, one island

According to Taiwan Semiconductor Manufacturing Company’s 2025 annual report, published to shareholders in April 2026, the company manufactured 12,682 different products for 534 customers during the calendar year, using 305 distinct process technologies. Its consolidated revenue reached NT$3,809 billion, roughly US$122 billion, up 35.9 per cent year-on-year. The company shipped 15 million 12-inch equivalent wafers from an annual production capacity of approximately 17 million.

The 534 customer list includes almost every company whose name appears on a chip specification the modern economy runs on. Apple, Nvidia, AMD, Qualcomm, Broadcom and MediaTek all sit inside it, alongside a long list of automotive semiconductor firms, defence contractors, and specialist chip designers. The 12,682 product lines cover the silicon inside phones, cars, medical equipment, banking cards, industrial controllers, and the accelerator hardware training the current generation of large language models.

What connects them is the same handful of fabrication plants clustered along the western coast of a single Taiwanese island 130 kilometres from the coast of mainland China.

► ► Watch Silicon Canals: Everything You Own Depends on This One Island — how the world quietly built its brain on the most contested shoreline on Earth, and why every attempt to move it is running into the specific problem that the frontier stays where the frontier is.

The frontier has stayed in Taiwan

The concentration has been tightening rather than easing. TSMC’s 2025 annual report discloses that its Advanced Technologies category, which it defines as processes at 7 nanometres and beyond, accounted for 74 per cent of the company’s total wafer revenue during 2025, up from 69 per cent in 2024. Its 3-nanometre process alone accounted for 24 per cent of that revenue. Its 2-nanometre process entered high-volume manufacturing at the end of 2025, with a fast ramp expected through 2026, and additional 2nm capacity is being built at multiple sites inside Taiwan.

Three-quarters of TSMC’s wafer revenue now comes from processes that no competitor on Earth is manufacturing at comparable volume or yield. The commercial frontier has become the entire business, and the frontier is being expanded, not diversified.

According to a Center for Strategic and International Studies analysis titled “Silicon Island: Assessing Taiwan’s Importance to U.S. Economic Growth and Security”, Taiwan’s chip industry is embedded in supply chains that are fundamental to consumer electronics, automotive manufacturing, military systems, and dual-use applications including artificial intelligence. The report notes that Taiwan’s concentration of people, expertise, infrastructure and equipment cannot be replicated at scale in the short term. That single sentence is the crux of the entire policy problem.

Why the island is more fragile than the numbers suggest

A fabrication plant is not a sealed structure that can be protected by soldiers. Producing chips at the current commercial frontier requires an ecosystem of inputs, most of which have nothing to do with silicon at all. It requires electricity supply so stable that a single dip in current can ruin production batches worth several million dollars. It requires water pure enough to rinse the surface of a wafer, delivered continuously at swimming-pool volumes. It requires the specific lithography machines built by ASML in the Netherlands, and the engineers with the accumulated experience to keep those machines operating at yield. It requires ports for the incoming equipment, airports for the specialist personnel, and the accumulated production knowledge that has taken decades to develop inside the local workforce.

In 2021, Taiwan ran short of one of those inputs. Not chips. Water. A severe drought lowered the island’s reservoirs, and the government made a choice it did not describe publicly at the time in those terms. It directed the remaining water supply toward the fabrication plants rather than toward the agricultural sector. Farming could wait a season. The world’s most advanced chip production apparently could not.

The genuine fragility of the arrangement is not that any single element could fail. It is that the arrangement depends on the simultaneous availability of many elements, each of which has a single point of failure. Any one of them, disrupted for long enough, would slow or stop the most advanced corner of the modern economy.

The de-risking plan is real, and it is not enough

Every major government exposed to this dependence reached the same conclusion around the same time, which was that production had to be spread outside Taiwan. TSMC’s 2025 annual report confirms that its first Arizona fab reached high-volume production at the end of 2024, its second Arizona fab has been pulled forward to the second half of 2027, and construction on a third began during 2025. In Kumamoto, Japan, the first JASM fab reached volume production at the end of 2024, with a second under construction. In Dresden, Germany, a specialty fab broke ground during 2024.

Those developments represent tens of billions of dollars of committed capital and are producing real chips at real yields. What they are not producing is the frontier. TSMC’s same annual report confirms that multiple phases of its 2-nanometre expansion are being built at home in Hsinchu and Kaohsiung alongside its leading-edge packaging capacity. The Arizona, Kumamoto and Dresden plants are running at process nodes that are one to several generations behind the primary Taiwanese frontier.

The unresolved policy problem sits inside that specific gap. What the de-risking programme is building outside Taiwan is functional back-up capacity for the previous generation of production. What it is not doing, and on current construction timelines cannot do within the next several years, is replicate the ecosystem concentration that keeps the frontier operating.

The slow blockade

The pressure on Taiwan does not require an invasion to be operationally significant. Naval exercises rehearsing an operational sealing of the island, coast guard vessels testing which state controls Taiwan’s surrounding waters, and continuous air patrols probing the island’s response capacity are all sufficient to test the operational continuity of the fabs without any missile being fired.

According to recent Silicon Canals coverage of the same beat, analysts describe the specific fear this pattern names as a slow blockade. Not a sudden invasion with a clear beginning, but a patient tightening. A test of how long the operational continuity of the world’s most important island can be sustained under pressure that never quite reaches the threshold of an act of war. In June 2026, Wall Street Journal reporting documented Chinese Coast Guard vessels issuing direct commands to foreign commercial ships operating on Taiwan’s Pacific side during a five-day operation, drawing formal rebukes from the United States and other Western governments.

If the answer to how long the arrangement can absorb that pressure is not indefinitely, then the recovery timeline for the modern economy after any prolonged interruption of Taiwanese production has to be measured in the same units as the construction timeline for the alternative fabs being built abroad. On the current publicly announced schedules, that is not weeks. It is years.

During that recovery window, the specific chips required to train, deploy and iterate the current generation of AI models would not be manufactured at comparable volume anywhere. The consumer electronics, automotive and defence sectors that depend on those chips would experience shortages of a category not seen in the modern industrial economy.

The world spent four decades quietly consolidating its most economically consequential industry onto a single island, one individually reasonable commercial decision at a time. On the accumulated evidence, no one designed the arrangement, no one voted for it, and no one has yet identified a way to unwind it that operates on the same timescale as the pressures the arrangement is now being subjected to.