ANALYSIS: The AI Boom Is Choking East Asia — Kaohsiung and Yongin Pay the Price

An article that disturbs

On June 19, 2026, journalist Yujie Xue of Nikkei Asia published an article with a sober title but devastating content: "The AI boom is quietly choking people in East Asia." Her subject: the explosive growth in demand for AI semiconductor chips is driving a massive increase in electricity consumption in Taiwan and South Korea, consumption mainly covered by natural gas and coal — fossil fuels that emit air pollutants directly responsible for premature deaths, asthma, cardiovascular diseases, and significant economic losses.

The Nikkei Asia article relayed by The Asia Cable on June 22, 2026 points precisely to cities: Kaohsiung in southern Taiwan and Yongin in South Korea. These two cities share a profile: they host massive semiconductor manufacturing plants — for TSMC and Samsung/SK Hynix respectively — and their air quality is deteriorating as a direct consequence of those plants' expansion under AI boom pressure. The link is direct, documented, and morally uncomfortable for an industry that presents itself as the key to human progress.

The fundamental contradiction of green AI

The global tech industry prides itself on being at the forefront of the fight against climate change. Major tech companies — Microsoft, Google, Apple, Meta, Nvidia — all have carbon neutrality goals. But their semiconductor supply chains are concentrated in East Asian regions that depend heavily on fossil fuels to power their chip manufacturing plants. This dependence is not abstract. It translates into measurable, documentable air pollutant emissions directly attributable to AI chip demand. Carbon neutrality in San Francisco does not erase pollution in Kaohsiung.

Nvidia earns billions. TSMC runs its fabs. Chips power the LLMs that power AI usage that powers stock valuations. And in Kaohsiung, someone coughs. This causal chain is real, documented, and systematically absent from tech CEO presentations to their investors. This is not bad faith. It is incomplete accounting. And incomplete accounting produces incomplete decisions.

Greenpeace East Asia and the quadrupling of emissions

A report by Greenpeace East Asia published in April 2025 — cited by the South China Morning Post — had already documented the scale of the problem. Electricity consumption linked to AI chip manufacturing had "more than tripled" to approximately 984 gigawatt-hours, driving a fourfold increase in CO2 emissions to 453,600 metric tons. These figures, already concerning in 2025, have continued growing with the expansion of manufacturing capacity in 2026.

These emissions are not all CO2. Burning coal and gas also produces fine particulates (PM2.5), nitrogen dioxide (NO2), and sulfur dioxide (SO2) — pollutants whose effects on respiratory and cardiovascular health are well documented by medicine. Studies cited by Nikkei Asia in June 2026 projected concrete health consequences: premature deaths, asthma, respiratory diseases, and associated economic losses in areas adjacent to major semiconductor plants.

Taiwan: the TSMC paradox

Taiwan is at the heart of this contradiction. TSMC — which manufactures approximately 60% of the world's advanced chips according to industry estimates — is also the island's largest industrial electricity consumer. The expansion of its plants to meet AI demand (notably new plants for 2nm and 3nm nodes) requires additional gigawatts. And Taiwan, whose energy mix still includes a significant share of natural gas and which has closed part of its nuclear capacity, must compensate this growing demand with fossil fuels.

Air quality monitoring data in Kaohsiung — southern Taiwan city where TSMC has plants and other heavy industries are present — regularly shows PM2.5 levels exceeding the thresholds recommended by the World Health Organization. These levels have a documented correlation with industrial expansion and electricity consumption. Kaohsiung residents did not choose to live next to the world's most important semiconductor supply chain. But they pay its health cost.

I searched for premature mortality figures in Kaohsiung. They exist, they are published, but tech companies never cite them in their sustainability reports. It is variable-geometry accounting: we count tons of CO2 avoided by solar panels on the California headquarters roof, but we don't count years of life lost in Kaohsiung by workers and residents near the fabs manufacturing the GPUs. This accounting asymmetry is not bad faith. But it produces insufficient decisions.

Samsung, SK Hynix, and the Yongin complex

In South Korea, the city of Yongin — in Gyeonggi-do province, south of Seoul — is becoming the center of a gigantic semiconductor complex. SK Hynix is building a memory chip manufacturing cluster there covering approximately 4.15 km², with a planned investment of 120 trillion won (approximately $90 billion) over several decades. Samsung has facilities in the same region. These investments are directly linked to demand for HBM (High Bandwidth Memory) chips for AI GPUs.

The electricity consumption of these facilities is colossal. South Korea still depends heavily on coal and natural gas for its electricity — approximately 60% of its electricity mix in 2024. The expansion of semiconductor plants increases this demand at a pace that renewable energy cannot yet compensate. Studies cited by Nikkei Asia document health burdens linked to air pollution in peripheral areas of major Korean semiconductor plants.

The paradox of the accelerated energy transition

There is a deep paradox in this situation. The semiconductor chips manufactured in these plants are indispensable to the global energy transition: electric vehicles, smart grids, renewable energy management, industrial efficiency. In other words, the tools of the solution to climate change are manufactured in industrial processes that contribute to climate change. This is not an absurd contradiction — every energy transition involves transition costs. But transparency about these costs is essential for political and industrial decisions to be properly informed.

Researchers from the chipping point, an analysis article from intellinews.com of April 2025, calculated that to achieve tech industry carbon neutrality goals while meeting growing AI chip demand, Taiwan, South Korea, and Japan would need to multiply their renewable energy capacity several times over and conclude long-term Power Purchase Agreements (PPAs) with clean energy producers. This transformation takes decades. AI chip demand, meanwhile, is growing exponentially now.

We use AI chips to better manage renewable energy grids. And we manufacture those chips by burning coal. There is a painful irony in that which no one in the industry really wants to highlight. This is not deliberate hypocrisy. It is the result of a transition whose different parts are unfolding at different paces. But naming this irony is the prerequisite to resolving it. And Yujie Xue's article names it with welcome clarity.

Governments under contradictory pressure

The governments of Taiwan and South Korea face contradictory pressures. On one side, they are encouraged — by Washington, by their domestic industries, by their commercial partners — to maximize their semiconductor manufacturing capacity to meet global demand and maintain their strategically crucial position in supply chains. On the other, they face growing pressures on greenhouse gas emissions and on their populations' air quality.

These two pressures cannot be simultaneously satisfied in the short term without massive and immediate investment in renewable energy. Taiwan has an energy transition plan targeting a higher renewable mix by 2050. South Korea likewise. But the construction timelines for offshore wind farms, solar installations, and — for some — next-generation nuclear power are offset from the explosive growth in electricity demand from semiconductor fabs. The gap between demand and renewable capacity is filled by coal and gas. And this gap has a direct health cost.

The nuclear decision: the return in question

In this context, the question of nuclear power becomes central again. Taiwan had progressively closed its nuclear plants under public pressure after the Fukushima accident in 2011. This decision, politically understandable, created a hole in its energy mix filled by gas plants. Facing growing electricity demand from semiconductor fabs, some Taiwanese experts and decision-makers are advocating for a return to nuclear, at least in a version using new-generation small modular reactors (SMRs).

South Korea has itself changed energy policy, with the Yoon government reversing its predecessor's nuclear phase-out plan and committing to a new reactor construction program. This about-face is directly linked to the energy needs of the semiconductor industry. The geopolitics of chips is now shaping national energy policy. This is the measure of semiconductors' strategic centrality in the contemporary world.

Nuclear power has returned to the Taiwanese and Korean energy debate largely because semiconductor fabs consume too much energy to be powered solely by renewables in the short term. I'm not saying it's the right decision. I'm saying it's a real decision, made by rational decision-makers facing real constraints. And this reality should change how we calculate the cost of AI policies in chip-producing countries.

Companies that profit without accounting for the pollution

The Nikkei Asia article of June 19, 2026 explicitly points to the responsibility of major tech companies that profit from the AI boom but externalize the environmental and health costs of their supply chains. The Greenpeace report of April 2025 cited by the SCMP named specific companies: "While fabless companies like Nvidia and AMD are raking in billions from the AI boom, they are neglecting the climate impact of their supply chains in East Asia."

This accusation is legally and ethically complex. These companies do not own TSMC's or Samsung's fabs. They do not choose Taiwan's or South Korea's energy mix. But they create the demand that makes investments in new manufacturing capacities necessary, and they could use their considerable purchasing power to require their suppliers to source electricity from renewables, through PPAs or other mechanisms. Buyer power is a form of responsibility.

Carbon neutrality commitments: what about the supply chain?

All major US tech companies have carbon neutrality commitments of the "Scope 1 and 2" type (direct emissions and purchased electricity). Some, like Apple and Microsoft, have commitments extended to "Scope 3" emissions (supply chain). But even Scope 3 commitments generally allow implementation timelines and accounting methodologies that can obscure part of the problem.

CO2 and pollutant emissions from chip manufacturing are often accounted for in TSMC's or Samsung's balance sheet, not in Nvidia's or Apple's. And these suppliers have their own decarbonization targets and timelines — which do not necessarily coincide with the immediate needs of populations in Kaohsiung or Yongin. There is a responsibility gap between those who profit and those who suffer. This gap must be closed by regulation and by market and investor pressure.

ESG reports. Carbon neutrality commitments. Pretty infographics on avoided emissions. I've read hundreds of them. And in none of them did I find a line saying: "Our chips contributed X tons of PM2.5 to Kaohsiung this year, resulting in Y emergency room visits." That line does not exist in sustainability reports. It should. And its absence is not accidental — it reflects accounting that maximizes positive communication and minimizes concrete responsibility.

Taiwan as strategic sacrifice zone?

There is a geopolitical dimension to this industrial pollution question that deserves to be named. Taiwan is both the indispensable producer of chips that global AI depends on and a territory under existential military threat from China. The TSMC fabs that pollute Kaohsiung's air are the same ones that allow the US and its allies to maintain their technological lead over China. The question arises: to what extent does Taiwan bear a double burden — that of security (military threat) and that of sustainability (industrial pollution) — for the benefit of the rest of the world?

This question does not call for relocating TSMC's fabs — that would be catastrophic for global supply security and ruinous for Taiwan. But it calls for a more equitable contribution from final beneficiaries to Taiwan's energy transition. Companies like Apple, Nvidia, and AMD, which earn massive margins on the products TSMC manufactures for them, could invest directly in deploying renewable energy in Taiwan through long-term PPAs. This is not altruism. It is supply chain risk management.

South Korea in the geopolitical equation

South Korea is in an analogous position. Its memory chip plants power global AI and rely on still heavily carbonized energy. It is also a strategic US ally in a region under pressure from China and North Korea. Modernizing its energy mix toward clean sources is in its own interest and that of its allies — a South Korea whose semiconductor plants depend on energy infrastructure vulnerable to disruptions (gas supply cuts, coal price fluctuations) is a less reliable South Korea in its ability to maintain production.

The alliance between energy security, environmental sustainability, and technological competitiveness is real. And governments that have the lucidity to address them together — rather than separately in the usual ministerial silos — will be those that build the most resilient technological supply chains of the 21st century.

Taiwan bears the military risk of manufacturing chips for the world. And it also bears the health risk. This double burden is real and rarely named in discussions about technological supply chains. Perhaps because it is uncomfortable for multinationals that prefer to talk about their carbon neutrality commitments in well-crafted press releases rather than looking squarely at what their supplier chains do to Kaohsiung's air. Discomfort is not a reason to avoid naming reality.

100% renewables in supply chains: the 2030 goal

The intellinews.com analysis article of April 2025 formulated precise recommendations, reprised in the June 2026 debate. Major tech companies — Nvidia, Microsoft, Meta, Google — should commit to sourcing 100% renewable energy in their supply chains by 2030. Chip manufacturers — primarily in East Asia — should set ambitious renewable energy targets and adopt direct high-impact procurement strategies.

These recommendations are technically feasible but financially demanding. Building the renewable capacity needed to power Taiwan's and South Korea's fabs on 100% clean energy would require investments on the order of several tens of billions of dollars. Part of these investments can come from manufacturers' own funds. But part will have to come from their customers — major tech companies — in the form of long-term PPAs guaranteeing stable revenues for renewable energy producers.

Regulation as accelerator: Europe ahead

The European Union is ahead on this subject. The Corporate Sustainability Reporting Directive (CSRD) now requires large companies to report their Scope 3 emissions — including their supply chain. The Corporate Sustainability Due Diligence Directive (CS3D) introduces legal liability for environmental damage in supply chains. These regulations, applying to companies operating in the EU or selling products there, have a pull effect on East Asian manufacturers wanting to maintain access to the European market.

The US does not yet have equivalent regulation, although the SEC proposed climate reporting rules still under debate. In this context, European regulatory pressure is one of the most effective drivers available to accelerate the energy transition in semiconductor supply chains. And this is not hostile pressure — it is pressure that will first benefit the populations of Kaohsiung and Yongin.

Europe regulates. The US debates. East Asia produces by burning coal. This division of labor is consistent with recent history, but it cannot last. Because the populations of Kaohsiung and Yongin are not going to wait indefinitely for US multinationals and their shareholders to decide that sustainability is profitable enough to invest in. Change will come either from regulation or from popular protest. And in both cases, it is better to anticipate than to suffer it.

Beyond fabs: the data centers themselves

The Nikkei Asia article of June 19, 2026 focuses on chip manufacturing. But there is a second wave of AI-related electricity consumption adding to the first: the data centers themselves, where chips are deployed to train and run AI models. These data centers are exploding in capacity across all of East Asia.

An article from techpulse of June 8, 2026 documented that AI data center construction "is running into the limits of East Asian electricity grids" in Japan, South Korea, and Taiwan. Data center power demand adds to that of semiconductor fabs, creating cumulative pressure on electrical grids not designed to absorb this growth. Blackouts, outages, and grid access constraints are becoming real operational risks for the industry.

Kaohsiung and Yongin as symbols of a broader reality

The cities of Kaohsiung and Yongin cited by Nikkei Asia of June 19, 2026 are not isolated cases. They are local expressions of a global trend: the concentration of technology production in specific geographic regions creates concentrations of energy consumption and air pollution that residents of those regions suffer directly, without having chosen to be at the epicenter of the AI revolution.

This reality has a social justice dimension that exceeds the purely technological or economic framework. Workers and residents near the semiconductor plants of Kaohsiung and Yongin are the first victims of a technological transition whose main financial beneficiaries are located thousands of kilometers away. This is a question of environmental justice on a global industrial scale.

Environmental justice. This concept, born in Black and poor US communities confronted with polluting factories, now applies on a global and industrial scale that its founders could not have imagined. Workers in Kaohsiung, residents of Yongin — they are not anonymous victims of abstract progress. They are people, with children who have asthma, parents who have cardiovascular disease. And this reality must figure in the industrial and political decisions that cause it.

The impossible short-term decoupling between AI and fossil fuels

The factual reality is that the AI boom is, in the short term, incompatible with the rapid decarbonization of economies hosting its production. This reality does not condemn AI. But it implies difficult decisions: accept slower AI growth if renewables cannot keep up; accept increased short-term pollution in the hope that the energy transition catches up with growth in the medium term; or invest massively and immediately in renewables to hold both objectives simultaneously.

The third option is the only one ethically and strategically acceptable. It is also the most costly in the short term. And in an environment where markets favor quarterly returns and governments favor immediate electoral benefits, investing massively now for medium-term benefits is structurally difficult. This is where regulation plays its irreplaceable role: creating incentives that make sustainability the economically rational short-term decision.

International cooperation as the only systemic solution

No country can alone solve the pollution problem generated by the global semiconductor supply chain. The solution requires cooperation between consumer-country governments (US, Europe), producer-country governments (Taiwan, South Korea, Japan), and companies at every stage of the chain. Mechanisms like the Critical Minerals Partnership among Quad democracies, supply chain sustainability standards, and transboundary renewable energy PPAs are instruments beginning to be explored.

These mechanisms must be accelerated. The window of time to avoid serious long-term health consequences in production zones is narrowing as investments in new manufacturing capacities materialize. What is decided today in tech company boardrooms and environment ministries will show up in Kaohsiung's and Yongin's public health statistics in ten years.

I want to believe that the leaders of TSMC, Samsung, Nvidia, and Microsoft read these articles. I want to believe that Greenpeace report co-author Katrin Wu and journalist Yujie Xue reach decision-makers who have the power to change something. But I also know that the path between a Nikkei Asia article and a $10 billion renewable energy investment decision in Kaohsiung is long and tortuous. What shortens this path is pressure — regulatory, shareholder, popular. And this pressure begins with the ability to name what is happening.

TSMC and its growing share of Taiwan's electricity consumption

The analysis published by Nikkei Asia on June 19, 2026 placed TSMC at the center of the debate on AI energy consumption. In 2024, TSMC consumed approximately 8% of Taiwan's total electricity — a proportion reflecting the concentration of global advanced chip manufacturing on an island of 23 million inhabitants. The company's internal projections, partially made public in its annual sustainability reports, anticipate electricity consumption growth of 12 to 15% per year through 2030, driven primarily by fabs dedicated to 3nm and 2nm chips needed for AI processors. If these projections materialize, TSMC could consume up to 20% of Taiwan's total electricity by the end of the decade.

This concentration of energy consumption creates a structural tension for Taiwan. The Taiwanese government has committed to achieving carbon neutrality by 2050 and raising the renewable share to 30% of the electricity mix by 2030. At the same time, it actively supports TSMC's capacity expansion as a pillar of national security strategy — Taiwanese fabs are both an economic asset and a geopolitical shield. These two objectives enter direct contradiction: each new leading-edge fab is an energy demand that slows the decarbonization trajectory. Taiwan cannot simultaneously be the world's capital of advanced chip manufacturing and meet its climate goals without a revolution in its electricity supply.

The return to nuclear: the question Taipei hesitates to raise

Facing this tension, a sensitive question is returning to Taiwan's energy debate: returning to nuclear power. Taiwan had gradually closed its nuclear plants under public pressure after Fukushima in 2011, with the last Taiwanese nuclear unit shut down in July 2023. But the explosive growth in electricity demand linked to AI and fabs has revived the debate. The Taiwanese Parliament, dominated by the Kuomintang since 2024, adopted a non-binding resolution in 2025 recommending the government study extending the life of existing reactors or building new ones.

President Lai Ching-te's government remains officially opposed to nuclear but faces a practical reality: renewables — solar and wind — cannot alone absorb short-term electricity demand growth. Investment in liquefied natural gas plants fills part of the gap but increases short-term CO₂ emissions. Technologies like small modular reactors (SMRs) — politically less controversial than large traditional nuclear plants — are emerging as a potential compromise in Taiwan's energy debate. The Taiwanese nuclear question is no longer just an environmental question — it is a question of energy sovereignty in the context of the AI race.

TSMC consumes 8% of Taiwan's electricity. A single industrial group. This proportion doesn't shock economists measuring the value generated. It should shock urban planners, electrical engineers, resilience planners. Because an island where 8% — soon 20% — of electricity consumption depends on a single actor has a systemic vulnerability that GDP figures don't capture. If TSMC loses electricity, Taiwan loses global AI. The world depends on it.

The SK Hynix Yongin complex: $90 billion over ten years

While the debate on TSMC's electricity consumption occupies Taiwan, South Korea is building the largest industrial semiconductor concentration ever undertaken. The SK Hynix complex at Yongin — announced in 2023 with total investment of 120 trillion won (approximately $90 billion over ten years) — aims to bring together on one geographic site four mega-fabs for advanced memory (HBM, DRAM, NAND) and the entire associated supplier ecosystem. This complex is not just a factory. It is an industrial city dedicated to producing memory for AI. The Nikkei Asia analysis of June 19, 2026 specifically cited Yongin as an example of industrial density amplifying AI's environmental impacts.

Yongin complex electricity consumption projections, at full capacity, are staggering. Each advanced memory mega-fab consumes between 500 MW and 1 GW of electricity — equivalent to the consumption of a city of 250,000 to 500,000 inhabitants. Four mega-fabs on one site, plus support infrastructure, represent an additional electrical load for the Seoul region exceeding the production capacity of several power plants. The Korea Electric Power Corporation (KEPCO) has had to revise its grid projections for the Gyeonggi-do region — the province where Yongin is located — accounting for the SK Hynix complex development. The geography of AI is imprinting itself on the geography of electrical grids.

Samsung and its decarbonization strategy under pressure

Samsung Electronics, the other Korean semiconductor giant, faces the same challenge. The company committed in 2022 to achieving carbon neutrality for its global operations by 2050 and to using 100% renewable energy in its global operations by 2050. But the explosive growth of its semiconductor division — driven by demand for HBM (High Bandwidth Memory) chips for AI processors from Nvidia, AMD, and Google — creates an identical tension to TSMC's. Each new fab built to meet AI demand is a new focus of electricity consumption and greenhouse gas emissions.

The Greenpeace analysis cited by Nikkei Asia projected a fourfold multiplication of the carbon footprint of East Asian data centers and semiconductor manufacturing facilities by 2030 if current trajectories are maintained. For Samsung and SK Hynix, this projection raises a corporate governance question: can they maintain their ESG (Environmental, Social, Governance) commitments while meeting growing AI customer demand? The honest answer is: not with their current energy mix. The semiconductor industry's carbon neutrality commitments are unachievable without a radical transformation of electricity sources in South Korea and Taiwan.

Four mega-fabs. An industrial city of $90 billion. Hundreds of thousands of workers, technicians, engineers. And somewhere in the environmental impact projections, the electricity consumption figure that will push CO₂ emissions of an entire province upward. I'm not saying we should not build Yongin. I'm saying we need to look at this figure squarely — and decide what we are prepared to accept as the cost of keeping AI production outside China.

The AI value chain and its environmental geography

The Nikkei Asia analysis of June 19, 2026 raised a question that Western tech companies generally prefer not to address frontally: to what extent do the carbon neutrality commitments of American and European data centers simply relocate their carbon footprint toward East Asian semiconductor manufacturers? A Microsoft Azure or Google Cloud data center powered 100% by renewables in Iowa or Ireland uses Nvidia H100 GPUs manufactured by TSMC in Taiwan with electricity partially produced by gas or coal plants. The carbon footprint of manufacturing those chips is not in the data center's carbon ledger — it is in TSMC's.

This accounting segmentation of the carbon footprint has concrete consequences on AI industry procurement policies. Companies that tout carbon neutrality in their operations are buying chips whose manufacturing emits considerable amounts of CO₂ and PFAS (per- and polyfluoroalkyl substances) — persistent chemical compounds used in semiconductor etching and cleaning processes. The real footprint of an AI deployment includes Scope 3 supplier emissions — a category that few tech companies honestly integrate into their sustainability reports.

Transparency initiatives: investor and regulator pressure

Facing this fragmented carbon accounting problem, growing pressures are being exerted on the tech industry to adopt a more holistic approach to its environmental footprint. The US Securities and Exchange Commission (SEC) adopted in 2024 climate disclosure rules requiring listed companies to report their Scope 3 emissions — including those from their supply chain. The European Union, through the Corporate Sustainability Reporting Directive (CSRD) and the green taxonomy regulation, imposes even more extensive requirements. These obligations create regulatory pressure on tech companies to account for their chip manufacturing emissions in their carbon ledgers.

Organizations like Greenpeace, the Electronic Industry Citizenship Coalition (EICC), and the Science Based Targets initiative (SBTi) publish assessments of the tech industry's complete value chain that highlight the gap between operational carbon neutrality commitments and total carbon footprint including equipment manufacturing and end-of-life. The Greenpeace analysis cited by Nikkei Asia is part of this growing transparency dynamic. The projected fourfold multiplication of East Asian data center and fab emissions by 2030 will only be avoided if Western customers integrate manufacturing carbon footprint into their procurement decisions — and if manufacturers are held accountable by binding rules.

Scope 3. These two technical words hide a reality that many sustainability reports prefer to ignore. My data center is green. My chips come from a gas-powered fab. My carbon ledger is incomplete. This is not bad faith — it is incomplete accounting. The good news: rules are changing. The SEC and the EU are pushing for more transparency. The bad news: they are arriving while the real footprint is growing faster than the rules.

Kaohsiung: from polluted industrial city to new AI capital

Kaohsiung, Taiwan's second city with 2.7 million inhabitants, illustrates particularly acutely the tension between industrial development and public health. For decades, Kaohsiung was the capital of Taiwanese heavy industry — steelworks, petrochemicals, shipbuilding — before a partial reconversion toward less polluting industries in the 1990s and 2000s. This reconversion did not completely resolve air quality problems: the city regularly records among Taiwan's highest air pollution indices, driven by its Linyuan petrochemical complex and port traffic.

The expansion of semiconductor fabs in the Kaohsiung region — notably new TSMC plants and data center parks — adds a new layer to this public health dossier. Semiconductor manufacturing processes emit volatile organic compounds (VOCs), fluorinated gases with very high greenhouse effect, and contribute to tropospheric ozone formation. The Nikkei Asia analysis of June 19, 2026 documented air quality measurements in residential areas near the new fabs exceeding World Health Organization PM2.5 recommendations in certain seasons. The industrial expansion of Kaohsiung in the name of AI has a public health cost that deserves to be named.

Riverside communities: between jobs and air quality

Kaohsiung residents living near industrial zones face what economists call a "jobs-health tradeoff" — a tension between the well-paying jobs that fabs create and the impacts on air quality and public health in their neighborhoods. Jobs in Kaohsiung's semiconductor industry represent salaries well above local averages. But epidemiological studies conducted in the 2010s around existing Kaohsiung industrial zones documented higher rates of respiratory and cardiovascular disease in neighborhoods most exposed to industrial emissions.

Environmental advocacy organizations in Taiwan — notably the Taiwan Environmental Information Center and local Kaohsiung groups — have requested in-depth environmental impact studies for new fabs and data centers, real-time open-data air quality measurements, and compensation mechanisms for communities near the plants. These requests have found partial echo in Kaohsiung's municipal government policies. But the economic and strategic pressure to keep fabs on Taiwanese soil — including US incentives to relocate part of production to the US — gives industrial companies considerable leverage in regulatory negotiations. The negotiating power of riverside communities facing semiconductor giants remains structurally asymmetric.

Kaohsiung's residents. They have no name in TSMC's sustainability reports. They do not appear in Indo-Pacific geopolitical analyses. They exist between the lines of air quality statistics and epidemiological studies. I don't know what they think about AI's expansion in their city. I know they deserve to be in the conversation — not just as a footnote on environmental impact, but as citizens whose right to breathable air is as legitimate as tech companies' right to affordable chips.

Innovations in data center energy efficiency

Facing the explosion in AI-related electricity consumption, the tech industry is investing massively in technologies that reduce energy per unit of computation. The industry's reference metric — Power Usage Effectiveness (PUE) — measures the ratio of total data center energy consumption to energy actually used for computing. The best modern data centers from Google, Microsoft, and Meta achieve PUE of 1.1 to 1.2 — versus averages of 1.5 to 2.0 for older data centers. These improvements have been made possible by innovations in cooling (immersion cooling, direct liquid cooling on chips), rack architectures, and workload management software optimization.

Processors themselves are becoming more energy-efficient per generation. The move from 7nm to 3nm to 2nm manufacturing nodes reduces power consumption per transistor — which is precisely one of the commercial motivations for the leading-edge node race. Nvidia's H100 GPU consumes approximately 700W. Its successor, the B200, consumes more in absolute terms but offers superior energy efficiency per computation — meaning the energy cost of a given AI computation decreases per chip generation. Moore's Law is not dead — it has partly transformed into an energy efficiency law.

Innovations in semiconductor manufacturing: reducing the footprint at source

On the fab side, important initiatives aim to reduce the environmental footprint at source. TSMC announced in 2023 its "Green Fab" program targeting a 22% reduction in fluorinated gas emissions per unit produced by 2030 through abatement gas treatment technologies and less emissive process chemistry substitutions. Samsung is investing in recovery and destruction systems for perfluorocarbons emitted in its etching and deposition processes. These initiatives do not solve the problem of absolute emissions growth — intensity reduction per unit is canceled out by the multiplication of units produced — but they reflect growing industry awareness of the problem.

More radical technologies are in development for the coming decades: high-power EUV lithography that reduces the number of processing steps, 3D stacking chip architectures that increase functional density without increasing silicon area, and alternative materials to silicon like silicon carbide (SiC) and gallium nitride (GaN) offering better electrical efficiency for certain applications. The race for AI computing power and the transition to a greener semiconductor industry can be compatible — but only if investment in energy efficiency is as prioritized as investment in raw performance.

Immersion cooling. High-power EUV. 3D stacking. These technologies exist. Some are already deployed at scale. What is missing is not technical innovation — the semiconductor industry is perhaps the world's most R&D-intensive. What is missing is incentive alignment: as long as the manufacturing carbon footprint remains off chip buyers' balance sheets, investment in green technologies remains less of a priority than investment in performance. Changing that — through regulation, transparency, or buyer demand — is the key.

What the Nikkei Asia article really says

The article by Yujie Xue in Nikkei Asia of June 19, 2026 says something simple and profound: the digital world has a physical and biological cost, and this cost is borne disproportionately by populations that did not choose their industrial geography and are not the primary beneficiaries of the technological revolution they fuel. This truth deserves to be at the center of industrial, political, and regulatory decisions — not in their footnotes.

The growth of AI is inevitable and, properly governed, beneficial for humanity. But this governance must include governance of its supply chain, its energy footprint, and its health externalities. Ignoring these dimensions does not make them disappear. They accumulate, silently, in the public health statistics of cities that most AI users have never heard of.

A call for complete accounting

The real challenge is not technological. It is accounting and political. As long as the cost of Kaohsiung's pollution does not enter the profitability calculation of the chips TSMC manufactures for Nvidia, and as long as that calculation does not affect Nvidia's investment decisions, the incentive to fund TSMC's energy transition will remain insufficient. Complete accounting — including health and environmental externalities — is the prerequisite to good 21st-century industrial governance. The Nikkei Asia article of June 19, 2026 is a contribution to this accounting. It is welcome and necessary.

453,600 metric tons of CO2. 984 gigawatt-hours. Premature deaths in Kaohsiung and Yongin. These figures, published in a Greenpeace report and picked up by Nikkei Asia, describe the biological price of the digital world. I'm not saying we should stop AI. I'm saying we should stop pretending this price doesn't exist. Because as long as we hide it, the right people don't make the right decisions. And the wrong people, they, quietly continue not making any at all.

By Maxime Marquette, columnist

Editorial stance

I am not a journalist but a columnist and analyst. My expertise lies in observing and analysing the geopolitical, economic and strategic dynamics that shape our world. My work is to take political strategies apart, to understand global economic movements, to put the decisions of international actors in context and to offer analytical perspectives on the changes that are redefining our societies.

I do not claim the cold objectivity of traditional journalism, which stops at reporting facts. I claim analytical clarity, rigorous interpretation and a deep understanding of the complex issues that concern us all. My role is to give meaning to the facts, to place them in their historical and strategic context, and to offer a critical reading of events.

Method and sources

This text keeps the basic distinction between verified facts and interpretive analysis. The rule never changes: a factual claim is published only when a verifiable source supports it, and the sources this article actually used are listed under Sources, never here.

Primary source families used by the publication, when the story has them: official statements from governments and international institutions, public statements by political leaders, reports from intergovernmental organisations, dispatches from recognised international news agencies.

Secondary source families: specialised publications, internationally recognised news media, analyses from established research institutions, reports from industry organisations.

When an article cites statistical, economic or geopolitical data, it comes from the institutions that produce it (intergovernmental organisations, central banks, national statistics offices), and the exact institution is named under Sources.

Nature of the analysis

The analyses, interpretations and perspectives in the analytical sections of this article are a critical, contextual synthesis based on the information available, the trends observed and the expert comments cited in the sources consulted.

My role is to interpret these facts, to set them in the context of today's geopolitical and economic dynamics, and to give them a coherent meaning within the larger story of the changes shaping our time. These analyses reflect expertise built through continuous observation of international affairs and an understanding of the strategic mechanisms that drive global actors.

This article describes a documented state of affairs on its publication date, not a prediction: later developments may shift its outlook. No update is promised in advance; when an article is corrected or completed, the change is dated in the text.