AI's Impact on Semiconductor Geopolitics: Foundry Wars 2.0

In the high-stakes competition for global technology supremacy, semiconductor manufacturing has become the ultimate trump card. But unlike oil in the 20th century, the semiconductor race is being fundamentally reshaped by its own product, which is artificial intelligence.

The global semiconductor market for data centers alone is projected to grow from $209 billion in 2024 to $492 billion by 2030, with AI infrastructure spending expected to exceed $1 trillion within the next few years. This explosive demand isn't merely driving revenue, it's restructuring the entire geopolitical landscape of chip manufacturing, concentrating power in the hands of a small number of companies and nations.

Welcome to Foundry Wars 2.0, a competition more intense, more technologically demanding, and more geopolitically fraught than anything the semiconductor industry has witnessed before.

TSMC: The Indispensable Empire

At the epicenter sits Taiwan Semiconductor Manufacturing Company, which has achieved something unprecedented: near-monopolistic control over the most critical technology of our era. TSMC commands 64.9% of the global foundry market as of Q3 2024, projected to reach 66% by 2025. More importantly, the company produces over 90% of the world's most advanced chips, the processors powering everything from iPhones to ChatGPT to F-35 fighter jets.

TSMC's 2nm process technology entered volume production in the second half of 2025, with major customers including Apple, AMD, Qualcomm, and Intel lined up. These wafers command prices around $30,000 each, which is a staggering figure underscoring both the technological sophistication and the captive market TSMC has created.

In Q3 2025, TSMC reported record earnings with profits up 39%, driven overwhelmingly by AI chip demand. High-performance computing now accounts for 57% of quarterly sales. The company's advanced nodes—3nm, 5nm, and 7nm combined represent 74% of total wafer sales, highlighting how AI is reshaping not just demand, but the very architecture of the semiconductor industry.

The CHIPS Act: America's Industrial Policy Gambit

The Biden administration's recognition of this dangerous dependency culminated in the CHIPS and Science Act of 2022, a $52.7 billion intervention unprecedented in modern American history. TSMC received $6.6 billion in federal funding for its $65 billion investment in three Phoenix, Arizona fabrication facilities. Intel was awarded $7.86 billion, reduced from an initially expected $8.5 billion. Samsung received $4.75 billion, a 26% cut from the preliminary $6.4 billion award due to yield challenges and workforce reductions.

These aren't merely construction projects, they represent an attempt to rewrite the fundamental geography of semiconductor manufacturing. TSMC's Arizona facilities will produce 2nm chips, marking the first time the company will manufacture its most advanced nodes outside Taiwan.

Yet the scale is daunting. The semiconductor industry faces an estimated need for 300,000 additional skilled workers just to complete ongoing fab projects. Approximately 67,000 jobs for technicians, computer scientists, and engineers risk going unfilled by 2030, representing 22% of the total semiconductor workforce.

The Trump administration has further complicated matters, pursuing equity stakes in CHIPS Act recipients, with plans to acquire 10% of Intel in exchange for funding. Reports suggest both TSMC and Broadcom are weighing the possibility of acquiring part of Intel, with the administration interested in facilitating such deals.

The ASML Chokepoint

No company better illustrates the concentrated nature of semiconductor manufacturing than ASML Holding, the Dutch firm with a monopoly on extreme ultraviolet lithography machines, the $150 million tools absolutely essential for producing chips at 7nm and below.

ASML's sales to China accounted for 29% of its revenue in 2024, but that figure is expected to drop to around 20% in 2025 due to tightening export controls. In September 2024, the Netherlands announced expanded restrictions requiring ASML to apply for licenses to service immersion DUV systems previously sold to Chinese customers, effectively cutting off maintenance for equipment already installed in Chinese fabs.

This maintenance ban represents economic warfare by other means. Chinese chipmakers had stockpiled ASML's DUV machines in 2023-2024 ahead of restrictions, with the company sourcing as much as 49% of its sales from China in Q2 2024. ASML CEO Christophe Fouquet stated that the EUV ban effectively pushes China's chip manufacturing capabilities back by 10 to 15 years.

The Taiwan Question: Silicon Shield or Vulnerability?

No geopolitical issue looms larger than Taiwan's precarious status. Research using tabletop exercises suggests that if China attempted to take control of Taiwan, it would most likely employ a quarantine method, and Taiwan's semiconductor supply chain would be particularly vulnerable before 2027.

Any conflict over Taiwan could interrupt global semiconductor supply, creating an economic shock given TSMC's critical position. The smartphone in your pocket, the data center running cloud infrastructure, the AI models powering innovation, all depend on the continued stability of this island 100 miles from mainland China.

In October 2025, Taiwan firmly rejected a U.S. proposal to split semiconductor production evenly, with Vice Premier Cheng Li-chiun refusing demands that half of chips consumed by the U.S. be manufactured on American soil. This came after President Trump reportedly threatened tariffs as high as 100% to 300% on semiconductors if Taiwan didn't significantly relocate its manufacturing.

This standoff reveals fundamental tension: Taiwan views its chip industry as a strategic asset ensuring American protection, while Washington increasingly sees Taiwan's production concentration as an unacceptable vulnerability.

Intel's Struggle and Samsung's Stumble

While TSMC extends its lead, competitors face mounting challenges. Intel showcased its first 18A process chip (roughly equivalent to 2nm) at CES 2025, announcing mass production in the second half of the year. But Intel's foundry division reported a $2.3 billion loss in Q4 2024 and didn't make the top ten foundry rankings.

Samsung's challenges are equally severe. The company's 3nm GAA FET node has failed to attract new customers, with trial production of its Exynos 2600 processor achieving only a 30% yield rate, which is far below the 60-70% yields TSMC achieves with 3nm. Samsung's foundry revenue declined 12.4% in recent quarters, and the company has publicly apologized for setbacks.

The 2nm Race: Commanding Heights of the AI Economy

The competition to mass-produce 2nm chips represents more than technological one-upmanship—it's a contest for control of the AI economy. Global advanced process capacity at 7nm and below is projected to increase by 69% from 2024 to 2028, reaching 1.4 million wafers per month. Capacity at 2nm and below will expand from under 200,000 wafers per month in 2025 to over 500,000 by 2028.

Capital expenditure on advanced process equipment is forecast to surge to over $50 billion by 2028, representing a 94% increase from the $26 billion invested in 2024. This investment tsunami reflects a simple reality: whoever controls 2nm manufacturing controls the AI supply chain.

TSMC began risk production of 2nm in July 2024, with mass production in the second half of 2025. Samsung plans to start production in 2025, while Intel scrapped its 2nm node in favor of 18A. TSMC's N2 process is expected to offer 10-15% higher performance at the same power or 20-30% lower power at the same performance, with over 20% higher transistor density compared to 3nm.

The China Dilemma

The U.S. export control regime represents the most aggressive technology denial campaign since the Cold War. Export controls since 2018 have restricted China's access to chipmaking tools, severely hindering the industry and making China a marginal producer of AI chips. The EUV ban has been particularly effective—no Chinese company can currently manufacture chips below 7nm at scale.

Yet technical benchmarks suggest that China's AI models don't have a significant capability gap compared with U.S.-produced models, with Chinese AI labs remaining fast followers. This paradox suggests that software innovation can partially compensate for hardware disadvantages, at least for now.

China's response has been multifaceted. Beijing imposed export controls on rare earth materials including holmium, thulium, and ytterbium, with licensing requirements extending to products containing these materials if their value content exceeds 0.1%. A European think tank has suggested responding by further tightening restrictions on legacy chip tools, specifically targeting DUV systems.

AI: The Demand Driver Reshaping Everything

The AI boom is restructuring what kinds of chips get made, where they're manufactured, and who has access to cutting-edge capacity. The global semiconductor industry is expected to grow by 15% in 2025, with AI and high-performance computing driving demand for advanced chips. AI server volumes are projected to grow at a 40-50% compound annual growth rate, testing production capacity to its limits.

This demand is highly concentrated. Graphics processing units take up about a third of the semiconductor die area of an AI server, with Apple and NVIDIA accounting for 25% and 10% of TSMC's revenue respectively in 2023. High-bandwidth memory revenue is projected to increase by up to 70% in 2025, with severe DRAM shortages predicted for 2026.

NVIDIA maintains an 85-94% market share in AI GPUs as of 2025, with its H100, Blackwell, and upcoming Rubin architectures serving as the backbone of the AI revolution. But NVIDIA's dominance depends entirely on TSMC's manufacturing capability.

The 2027 Inflection Point

Multiple timelines converge around 2027, creating what industry observers describe as a critical inflection point. Analysis suggests Taiwan's semiconductor supply chain would be particularly vulnerable to Chinese quarantine action before 2027, after which diversification and stockpiling strategies may provide greater resilience. By then, TSMC's Arizona fabs should be producing leading-edge chips, Intel's 18A process should be mature, and Samsung may have resolved yield issues.

Advanced node capacity is projected to maintain a robust 14% compound annual growth rate from 2025 through 2028, with production capacity surpassing one million wafers per month for the first time in 2026. But 2027 also represents a danger zone. If Beijing calculates that Taiwan's strategic value is declining as chip manufacturing diversifies, the incentive to act before the window closes could increase.

Conclusion: The Permanent Crisis

The semiconductor industry has entered a permanent state of crisis, not imminent collapse, but rather a critical turning point. AI has transformed semiconductors from an important technology sector into the strategic chokepoint of the global economy. The companies and countries that control advanced chip manufacturing will shape the 21st century in ways comparable to how oil producers shaped the 20th.

TSMC's massive U.S. investments demonstrate Taiwan's effort to show its indispensable role in the AI revolution, potentially influencing how the United States views Taiwan's value. But this creates its own risks as production diversifies. Does Taiwan's "silicon shield" weaken?

The next decade will determine whether the semiconductor industry can maintain the global cooperation necessary for continued innovation, or whether it fractures into competing regional ecosystems, each less capable than the integrated whole, but each offering the political security of reduced dependence.

In Foundry Wars 2.0, the battles are fought in cleanrooms rather than battlefields, but the stakes are just as high. Control the chips, control the AI. Control the AI, control the future. And right now, that control rests in the hands of a very small number of companies on a very small island in an increasingly contested sea.

The question isn't whether this situation is sustainable, it clearly isn't. The question is how it will change, and whether that change will come through strategic repositioning or strategic catastrophe.