Taiwan's semiconductor industry traces back to the Industrial Technology Research Institute (ITRI), a government-backed research body established in 1973. ITRI's own history describes a 1974 meeting of government and industry figures, including RCA engineer Wen-Yuan Pan, that set out a plan for integrated circuits. In 1976 ITRI signed a technology transfer and licensing contract with RCA, and in 1977 it opened Taiwan's first IC production line, which ITRI says reached a 70% yield within six months. [1]

ITRI then turned research into companies. In 1980 it created United Microelectronics Corporation (UMC), transferring its 4-inch wafer technology and R&D team; UMC was among the first tenants of the Hsinchu Science Park and in 1985 became Taiwan's first listed semiconductor company. In 1984 ITRI launched a very large-scale integration (VLSI) project to develop its own technology, which it describes as a response to barriers to acquiring foreign technology. [2][1]

In 1985 Morris Chang was invited to return to Taiwan as ITRI's president to lead development of 6-inch wafer technology. In 1987 ITRI spun off Taiwan Semiconductor Manufacturing Company (TSMC), transferring fabs, equipment, technology and 98 staff. ITRI describes TSMC as introducing the world's first dedicated IC contract manufacturing model, in which the company manufactures chips designed by its customers instead of selling its own. [2][1]

This foundry model underpins the modern fabless industry. SIA and BCG describe foundries as serving fabless designers and integrated manufacturers alike, spreading the very high capital cost of fabs, which they estimated at up to 20 billion dollars for an advanced logic fab, across many customers. TSMC reports that in 2025 it made products for 534 customers using 305 process technologies. [3][6]

Taiwan's industry developed as a dense cluster, with ITRI spinning off not just chipmakers but also suppliers such as Taiwan Mask Corporation in 1989. The Hsinchu Science Park became home to UMC, and Hsinchu is also where TSMC is based. TSMC now operates six 12-inch GIGAFAB facilities offering processes from 0.13 micrometres down to 2 nm. The company says it is expanding advanced packaging capacity in Chiayi and Tainan and building a facility in Taichung for a future 1.4 nm process. [1][2][8][13]

Proximity between foundries, packaging firms, mask makers, equipment service engineers and chip designers shortens the time needed to solve production problems. SIA and BCG found in 2021 that mainland China and Taiwan together held more than 60% of global assembly, packaging and test capacity, meaning much of the back end of the supply chain sits close to Taiwan's fabs. [3]

The 2021 SIA and BCG study estimated that all of the world's logic manufacturing capacity below 10 nm was in Taiwan (92%) and South Korea (8%). Their 2024 update says that in 2022 advanced logic capacity was still almost entirely split between those two places, and projects that by 2032 more than 40% will be located elsewhere as new fabs open, including in the United States. [3][4]

TSMC's position in contract manufacturing has grown alongside AI demand. TrendForce estimates that TSMC earned 40.2 billion dollars in the second quarter of 2026, or 72.5% of top-10 foundry revenue, citing strong demand for AI server GPUs and custom accelerators and noting that 2 nm began contributing revenue. TSMC reported 2025 revenue of about 122 billion dollars and 15 million 12-inch equivalent wafers shipped. [5][6]

Geopolitical risk receives the most attention. A 2023 RAND tabletop exercise concluded that production of the highest-end chips exists almost entirely in Taiwan and that there were generally no good short-term options if that supply were disrupted. Rhodium Group estimated in 2022 that well over 2 trillion dollars of economic activity would be at risk from a blockade, which it described as a floor. A 2025 CSIS series of 26 wargames on a possible blockade highlighted Taiwan's energy system as a key vulnerability, while also finding that a blockade would create escalation pressures that are difficult to contain. [9][10][11]

Natural hazards and resources are also discussed. After a magnitude 7.4 earthquake in Hualien in April 2024, TSMC said about 70% of tools had recovered within 10 hours and that critical equipment such as EUV machines was undamaged; analysts nonetheless used the event to highlight concentration risk. In early 2021 a severe drought led TSMC to order water by truck, and a reservoir serving Hsinchu fell to under 15% of capacity. [12][13]

Analysts differ on how to respond. RAND's report recommends incentivising some relocation of capacity and coordinating with allies, while also noting that building alternative production is expensive and slow, a point its exercise participants had often underestimated. Moody's analysts quoted after the 2024 earthquake argued for greater geographic diversification of suppliers. These are live policy debates in Taiwan, the United States, Japan and Europe, and assessments of the likelihood of any disruption vary widely. [9][12]

TSMC is expanding abroad while keeping its most advanced development in Taiwan. According to its 2025 annual report, its first Arizona fab entered high-volume production in late 2024 and a second is expected in the second half of 2027; its first fab in Kumamoto, Japan began volume production at the end of 2024; and a specialty fab in Dresden, Germany is progressing. In March 2025 TSMC announced plans to raise its planned US investment to 165 billion dollars, including more fabs and two advanced packaging plants. [6][7]

SIA and BCG project that the US share of global fab capacity will rise from 10% in 2022 to 14% in 2032 and that the US will host nearly 30% of sub-10 nm logic capacity by then, while Taiwan remains one of the largest centres. [4]

Nearly every high-end data center GPU on the market today depends on Taiwanese foundry and packaging capacity at some stage. That makes TSMC's capacity plans and any disruption in Taiwan one of the most important upstream factors for GPU supply, even for buyers who only ever rent compute in the cloud. Overseas fabs add resilience over time, but for the next few years the leading edge will remain heavily concentrated.

Kovara shows how these upstream conditions feed through to the market you actually buy in. Compare current GPU prices across providers, check which regions list specific accelerators, or ask Kova how supply for a given GPU has moved recently.