In September 2025, whilst Taiwan Semiconductor Manufacturing Company (TSMC) announced the construction of a new plant worth 500 billion Taiwanese dollars [1] for the production of 1.4-nanometre chips, the most advanced technology in the world, one detail passed almost unnoticed in the international press. A few weeks earlier, during the shareholders’ meeting, a woman had shown president Wei Zhejia the photograph of her brother, Huang Xinyi, who died in an accident on a TSMC building site. He was only the latest of at least ten subcontracted workers who had died since 2019. The contrast between the technological vanguard that promises to power the era of artificial intelligence and the broken bodies on building sites, between record profit margins and the growing energy pressures that threaten to paralyse the entire island, reveals an uncomfortable truth. The so-called Taiwanese ’silicon shield’ – the idea that Taiwan’s near-absolute monopoly in the production of advanced chips protects it from a Chinese invasion by making the island indispensable to both Beijing and Washington – is built on increasingly fragile foundations. Whilst the two superpowers compete for control of this strategic resource, TSMC finds itself trapped in a paradox. The more it expands globally to placate geopolitical pressures, the more it exposes the vulnerabilities of a production model that depends on social, energy and logistical conditions impossible to replicate elsewhere. Behind the rhetoric of technological progress lies a network of unresolved contradictions, from expendable workers to energy crises, from supply chains under siege to dependencies on China for renewable energies, which could transform Taiwan’s greatest industrial success into its most dangerous fragility.

When the shield becomes vulnerability

In 2020 the Taiwanese political scientist Tsai Ming-yen coined the expression ’silicon shield’ to describe Taiwan’s strategic position. The island’s near-monopolistic control over the production of advanced semiconductors made it too valuable to be attacked by China and too important to be abandoned by the United States. TSMC, which produces over 90% of the world’s most sophisticated chips, represented the material guarantee of this protection. The reasoning contained, however, a fundamental ambiguity: that position of strength depended entirely on the willingness of the great powers to consider Taiwan indispensable. And this perception can change rapidly, transforming the shield into a trap.

In the last five years, the evolution of American industrial policy has laid bare this fragility. The CHIPS and Science Act of 2022 [2] allocated billions to bring semiconductor production back onto American soil. For Taiwan this meant something disquieting, because if the United States rebuilt internal production capacity, the island would gradually lose its shield function. According to industrial forecasts, by 2032 the USA will hold 28% of global capacity for production of advanced chips, starting from zero in 2022. TSMC itself contributes massively to this growth, with an investment of 165 billion dollars (149 billion EUR) in Arizona for six production plants, two advanced packaging facilities and a research centre.

TSMC’s American expansion was not an autonomous strategic choice, but the result of explicit political pressures. When Trump announced in March 2025 the expansion of this commitment, the president of the Taiwanese company, Wei Zhejia, diplomatically emphasised that the decision responded to ’very large demand from American customers’. The formulation barely concealed the reality: TSMC was yielding to Washington’s demands, accepting significantly higher production costs and 440 million dollars (397 million EUR) of losses forecast for 2024 in Arizona, in order to maintain access to the American market.

The paradox of the silicon shield thus manifests itself in all its complexity. Taiwan still holds about 60% of global production capacity for advanced chips, but the trajectory is clear. The more TSMC builds plants abroad to respond to political pressures, the more it erodes the productive concentration that constituted the foundation of Taiwanese protection. Refusing these pressures would, however, mean losing access to Western markets and public funding, accelerating the company’s relative decline. There is no evident way out, because the shield works as long as Taiwan remains irreplaceable, but the demands of the countries that should give it support are gradually rendering it replaceable. It is a protection mechanism that contains within itself the conditions of its own obsolescence.

Energy, water and the hidden cost of artificial intelligence

Whilst the geopolitical debate concentrates on the strategic control of TSMC, the company faces a more prosaic but equally determining constraint. The artificial intelligence revolution has transformed advanced semiconductors from a niche product into a fundamental resource for the global economy, but this same revolution has generated an energy demand that Taiwan increasingly struggles to sustain. The enthusiasm that greeted the launch of ChatGPT in November 2022 radically transformed the prospects of the semiconductor industry. Only a few years earlier, analysts at the Center for Security and Emerging Technology [3] in Washington had predicted that technological progress in chip miniaturisation would soon reach an economic limit. Their 2020 report calculated that upgrading production from the 10-nanometre process to the 7-nanometre one required 3.7 years to pay back, whilst the passage from 7 to 5 nanometres required as much as 8.8 years. The so-called Moore’s law [4], which described the doubling of transistor density every two years, seemed destined to be exhausted for lack of commercial incentives. Artificial intelligence completely overturned this prospect. Large language models require such massive computing capacity that demand for advanced chips exploded. TSMC immediately accelerated the development of the most advanced technological nodes. The 3-nanometre process is currently in mass production, the 2-nanometre one will enter series production at the end of 2025, whilst the 1.4-nanometre process is planned for the start of mass production in 2028.

This technological renaissance has, however, a material price that rarely appears in triumphalist narratives about artificial intelligence. Producing a single silicon wafer at 3 nanometres generates emissions equivalent to 4,460 kilograms of carbon dioxide and consumes 37.2 cubic metres of water, almost double compared to the 5-nanometre process and over ten times more than the 90-nanometre chips used for less advanced applications. The average energy consumption to complete a circuit layer on a twelve-inch wafer went from 21.4 kilowatt-hours in 2017 to 40.5 kilowatt-hours in 2023. Overall, TSMC consumed 25.3 billion kilowatt-hours in 2024 alone, equal to 8.9% of all Taiwanese electricity. The semiconductor industry as a whole absorbed 14.2% of national consumption. These figures are destined to grow vertiginously. Taiwan Power Company, the state electricity company, predicts that by 2030 the semiconductor industry will require another 28.4 billion kilowatt-hours, a 75% increase compared to 2023.

Taiwan finds itself in a paradoxical situation. The government has set the target of 30% renewable energy by 2030, but in 2025 the share is stuck at 13%. TSMC alone has promised to use 60% renewable energy by 2030, a figure that represents over two-thirds of all current Taiwanese renewable production. The pressure also comes from clients. Apple has imposed carbon neutrality by 2030 on its key suppliers, and for TSMC this means that the growth rate of the green share should double in the next five years. The problem is that Taiwan does not produce enough renewable energy, and the expansion of solar and wind is proceeding more slowly than forecast.

The energy question has rapidly transformed into an internal political clash. The Taiwan People’s Party [5] launched in May 2025 a referendum proposal to keep the Lungmen nuclear power plant [6] in operation, maintaining that the government’s denuclearisation policy has caused massive losses for Taiwan Power Company. This ideological clash blocks any pragmatic discussion about solutions. Taiwan is sacrificing its own energy sustainability to maintain TSMC’s position in the global artificial intelligence industry, with the government treating the expansion of the semiconductor industry as an incontestable premise. The current growth trajectory is materially incompatible with declared climate objectives, and no innovation in production processes manages to compensate for the exponential increase in demand generated by artificial intelligence.

Supply chains as vulnerable nervous system

The rhetoric surrounding TSMC tends to concentrate on technological sophistication, systematically obscuring the fact that behind every silicon wafer exists an extensive network of suppliers, subcontractors and workers whose condition concretely determines the company’s operational capacity. When on 5 September 2024 Huang Xinyi died crushed by a pipe on the building site of a TSMC plant in Tainan, the accident received no media coverage. Huang worked for a small subcontractor that answered to Yankee Engineering, the company that had received the overall contract from TSMC. According to the inspection report, TSMC had formally drawn up a safety management plan, but had not effectively implemented the on-site checks provided for in the contract.

Huang Xinyi is not an isolated case. The Taiwan Association for Victims of Occupational Injuries has documented at least ten deaths among subcontracted workers at TSMC building sites since 2019. On 26 May 2025 another worker died in a plant under construction in Chiayi, again for a contract managed by Yankee Engineering. According to the Ministry of Labour’s consultation system, Yankee Engineering has accumulated at least sixteen violations of the occupational safety law between 2017 and May 2025, including failures to prevent falls and collapses, as well as inadequate contractual management. The company, listed on the stock exchange and with contracts for the construction of TSMC plants throughout Taiwan, continues to receive contracts despite this record.

Huang Xinyi’s sister described the mediation process to the Injury Association as ’being pushed into a ring to fight against a monster’. Since Huang was unmarried and his parents had died, the siblings could claim only a limited indemnity under the labour standards law, with no right to compensatory damages. The first proposed settlement provided for a disproportionately low sum. Only the Association’s intervention permitted more dignified conditions. After other fatal accidents in the same year, the sister presented herself at the TSMC shareholders’ meeting on 3 June 2025, declaring to president Wei Zhejia that ’TSMC is a great whale that the country has raised until making it so big’ and asking whether the company still had a sense of social responsibility. Wei apologised stating that workplace safety is an absolute priority.

The public apologies have not modified the logic of the production system. TSMC operates through a cascade subcontracting structure that progressively offloads risks and costs downwards. The company maintains strategic and technological control, defines delivery times and quality standards, but delegates material execution to contractors who in turn subcontract to smaller firms. This model permits TSMC to maintain high margins and operational flexibility, but creates a grey zone of responsibility where accidents can be formally attributed to subcontractors. Although the occupational safety law establishes that TSMC has the responsibility to inspect and train contractors, enforcement of this norm remains weak and the company can continue to maintain that safety problems primarily concern subcontractors.

This network structure is precisely what makes TSMC so efficient in Taiwan and so difficult to replicate elsewhere. The Taiwanese ecosystem has developed over decades around the specific needs of the semiconductor industry. Material suppliers, specialised equipment, maintenance services, even the laundries that clean the sterile garments for cleanrooms, all operate in synchrony with TSMC’s production cycles. The Taiwanese government has provided systematic support through subsidies, tax breaks and dedicated infrastructure. But this same ecosystem is based on asymmetric power relations.

To this is added that from 2026 onwards, suppliers with significant commercial activities linked to China will probably be excluded, and several companies have already suffered order cancellations. The US pressure for decoupling from China thus translates into a forced restructuring of the entire Taiwanese chain, with TSMC acting as intermediary for Washington’s demands. Equipment suppliers worry that the stricter supply chain management is now hitting Taiwanese firms beyond simple decoupling from China, even though this approach helps TSMC to understand the operational state of partners and avoid regulatory risks or American sanctions.

Global expansion and the failure of replicability

TSMC’s difficulty in reproducing its own operational model outside Taiwan emerges with particular clarity in the American experience. When the company announced in 2020 the construction of the first plant in Arizona, the prevalent narrative described the initiative as a triumph of transatlantic technological cooperation. The reality on the ground has proved much more complex. The Phoenix plants began mass production in 2025, achieving operational profitability, but the path was marked by delays, unforeseen costs and profound cultural tensions. A Taiwanese technician interviewed by Initium Media [7] on the Phoenix building site synthesised the problem in direct terms. ’It’s a question of human nature. American speed is what it is, and Taiwanese think it’s enough to be a bit faster than them. You receive the same salary, but others’ efficiency is only 50% of yours. Would you continue to commit yourself 100%?’ The frustration does not concern simply superficial differences of work rhythm. It instead reflects a clash between production systems built on incompatible assumptions regarding the relationship between work and personal life, between flexibility and regulation, between improvisation and proceduralisation.

TSMC executives have recognised that the principal challenge of overseas expansion consists in adaptation to local regulations and cultures. Supply chains must learn to conform completely to local laws, where regulations have absolute binding force and cannot be circumvented through personal relationships or interpretative flexibility. In Taiwan suppliers have become accustomed to collaborating closely with TSMC, satisfying functional needs but without familiarity with foreign procedures. The Taiwanese approach permits construction to begin even before projects are finalised, adapting flexibly to advance rapidly. This cannot in any way happen in the United States, where everything requires explicit documentation and no work can begin before completion of design.

The expansion in Japan and Germany is also encountering significant obstacles. President Wei Zhejia has recognised delays in the second plant in Kumamoto. Similarly, the German plant designed to serve the European automotive market faces weak macroeconomic conditions and a sector in declining phase, with the principal chip producers having announced massive redundancies. These developments make evident that TSMC’s expansion model might not support large-scale investments in Europe at this moment. The company is instead accelerating investments in the United States, committing to invest an additional one hundred billion for a total of one hundred and sixty-five billion (149 billion EUR), creating a compression of resources for other international projects.

The precarious future between crossed dependencies and structural vulnerabilities

TSMC’s global expansion generates a paradoxical consequence. Whilst the company builds production capacity in the United States to respond to Washington’s pressures, it simultaneously risks increasing its own dependence on China in the renewable energy sector. In 2024, TSMC used 3.61 billion kilowatt-hours of certified renewable energy, of which about 60% came from certificates purchased abroad. The company has never made public the precise origin of these certificates, but the context suggests an embarrassing reality: TSMC, as Apple’s principal manufacturing partner, might purchase Chinese green certificates to satisfy the sustainability objectives required by the American client. China has installed a total solar and wind energy capacity five times greater than the United States and dominates the solar component supply chain. The One Big Beautiful Bill Act of July 2025 seeks to limit this dependence by prohibiting tax incentives for renewable energies connected to foreign entities of interest, but risks further hindering the development of American renewables.

TSMC’s position regarding China is further complicated on the commercial and technological front. The company announced in November 2024 that it would completely interrupt the use of Chinese equipment in the 2-nanometre production lines, in response to growing American restrictions on exports of advanced technology. These restrictions have already forced TSMC to cease shipments to Huawei in September 2020, and subsequent rules issued in October 2022 and 2023 have further restricted the possibilities of supplying artificial intelligence chips and production equipment to Chinese clients. In October 2024, Canadian analysts at TechInsights [8] discovered that a chip produced by TSMC had ended up inside Huawei’s Ascend 910B processor for artificial intelligence, despite the sanctions. The incident unleashed bipartisan concerns in Washington, with senator Mark Warner criticising the ’weak enforcement’ of existing controls by the American government. TSMC could face fines exceeding one billion dollars (903 million EUR), although so far no sanction has been imposed. The episode underlines the growing difficulty of controlling chips’ final destinations in global supply chains where design companies can obscure downstream relationships.

American tariff pressures add a further layer of uncertainty. In April 2025, President Trump proposed a commercial regime that includes a baseline tariff of 10% on all imports, a reciprocal tariff of 32% on Taiwanese products and potentially a 25% levy on semiconductors. TSMC’s annual report warns that such tariffs could increase costs for clients and brake American growth. All this creates a paradox: TSMC invests to satisfy American national objectives whilst finding itself facing policies that could penalise it. The company emphasises that semiconductors represent only 2.5% of American imports, therefore its investment will not perceptibly reduce the United States’ overall trade deficit.

The alliance announced in September 2025 between Nvidia and Intel further complicates the competitive landscape. Nvidia invested five billion dollars (4.5 billion EUR) in Intel, an operation interpreted by the sector as a concession to the Trump administration rather than a financial investment with return expectations. Intel has received overall 12.7 billion dollars (11.5 billion EUR) of fresh capital through investments from Nvidia, SoftBank and CHIPS Act subsidies. The alliance provides for Intel to build customised processors based on x86 architecture [9] for Nvidia’s artificial intelligence infrastructure. For TSMC, the alliance creates uncertainty about future order allocation: if Intel makes its own processes competitive compared to TSMC’s technology below three nanometres, even American partners like Nvidia might hesitate to shift orders.

TSMC’s technological supremacy remains for now unchallenged. The company has consolidated a global market share of 38% in the second quarter of 2025, with revenues increased by 44% on an annual basis, fuelled by demand for the 3-nanometre process and advanced packaging. Samsung has seen its own share drop to 4%, widening the gap with TSMC to 33 percentage points. However, TSMC’s dominant position does not eliminate structural vulnerabilities. The company depends on specific Taiwanese conditions that it cannot replicate elsewhere, faces domestic energy pressures incompatible with its own growth, suffers contradictory demands from Washington and Beijing that force it simultaneously to decouple from China and to depend on it for renewable energies, and maintains a local supply chain under growing stress. The model that has made TSMC indispensable contains within itself the conditions of its own fragility. Advanced technology does not float in an abstract space, and is instead anchored to limited material resources, asymmetric power relations and political choices that determine who pays the costs of progress and who reaps the benefits.

Andrea Ferrario

Sources: Initium Media, Digitimes, Taipei Times, Asia Society, Radio Free Asia, Tai Sounds, Taiwan Insight, Nikkei Asia, Bureau of Economic Analysis, TechInsights