2025 Global Wafer Fab Investment Overview: A Pivotal Year for Chip Supply Chain Restructuring

In 2025, the global semiconductor industry is witnessing an investment storm of unprecedented scale. According to the "Annual Global IC Fabs And Facilities Report" published by Semiconductor Engineering, more than 170 wafer fab and facility investment and update plans have been announced worldwide over the past 12 months, covering the entire industrial chain including manufacturing, materials, packaging, design, and R&D. From Hsinchu, Taiwan, China to Hokkaido, Japan; from Arizona, USA to Dresden, Germany, capital is flowing into chip capacity construction at an unprecedented speed.

This is not an ordinary capacity expansion cycle. The explosive growth in AI computing demand, combined with supply chain security anxiety triggered by geopolitics, has forced governments and companies to reassess the strategic value of semiconductor manufacturing. Over the past year, we have seen TSMC add US$100 billion in investment in the United States, Micron build an AI memory fab in Japan, the EU launch five pilot lines under the Chips Act, and even India approve four new wafer fab projects at once. Yet beneath the boom there is also a chill: Intel postponed its Ohio fab and canceled its European fab plans, while the multi-billion-dollar projects of Wolfspeed and SanDisk have flashed yellow. The ebb and flow of investment reflects a deeper contest between technological transitions and policy swings.

This article will analyze the real picture behind the 2025 global wafer fab investment boom from four dimensions—the industrial chain, technology roadmaps, market competition, and regional landscape—and attempt to answer a key question: as manufacturing capacity begins to be redistributed, how will the semiconductor landscape be reshaped over the next decade?

1. AI and Geopolitics: The Investment Logic Driven by Dual Engines

Understanding this round of investment boom cannot be separated from two keywords: AI infrastructure and supply chain autonomy.

On the one hand, the training and inference of large AI models consume astonishing amounts of computing power, and the AI chip orders of companies such as NVIDIA, AMD, and Broadcom are already booked years in advance. This has directly driven rigid demand for advanced process nodes (3nm/2nm) and advanced packaging (CoWoS, SoIC). TSMC's expansion of six wafer fabs at once in Taiwan, along with shifting some capacity to advanced packaging, is precisely intended to ease the bottleneck in AI chip capacity. On the memory side, HBM, as a key component of AI accelerator cards, has prompted SK hynix, Micron, and Samsung to make heavy bets—SK hynix's Yongin cluster could see total investment as high as KRW 600 trillion (approximately US$407 billion), while Micron is building a dedicated AI memory fab in Japan.On the other hand, geopolitics is profoundly reshaping the global division of labor in the semiconductor industry. Through the CHIPS and Science Act and its subsequent policies, the United States is attempting to bring advanced manufacturing back to its own shores; the European Union has introduced the Chips Act, funding pilot lines from 2nm to photonics; Japan is heavily subsidizing TSMC, Micron, and Rapidus; and India is leveraging its "Semiconductor Mission" to attract foreign investment in fab construction. These policy interventions are not purely economic incentives, but also a form of strategic positioning—whoever controls process nodes and manufacturing capacity will hold the foundation of the future digital economy.

It is worth noting that the investment boom in 2025 is not evenly distributed. Advanced logic and memory account for most capital expenditures, while mature process nodes, analog, power, and other areas are diverging. For example, onsemi received €450 million in the Czech Republic to build a SiC power device plant, Infineon is expanding in Germany, while NXP chose to shut down its GaN fab in Arizona—reflecting the vastly different supply-demand prospects facing different device types.

II. Technology Roadmap: From 3nm to 2nm, Advanced Packaging Becomes the New Battlefield

The 2025 investment list clearly outlines the trajectory of semiconductor technology evolution.

In logic process technology, TSMC is steadily advancing on its 2nm GAA (Gate-All-Around) process, and the expansion of its new fab in Taiwan is precisely to prepare for 2nm mass production. Japan's Rapidus also achieved trial production of 2nm GAA transistors this year, marking an important milestone in Japan's return to advanced process technology after many years. Meanwhile, mainland China's SMIC and Huawei are targeting the 5nm node, attempting to break through yield bottlenecks under existing equipment constraints. These efforts show that the race below 2nm is expanding from a three-way contest among TSMC, Samsung, and Intel to include participants from more countries and regions.

Advanced packaging has shifted from a supporting role to a leading one. AI chips' demands for transmission bandwidth and power consumption have made 2.5D/3D packaging technologies such as CoWoS and InFO the capacity bottleneck. TSMC is significantly expanding its advanced packaging facilities, and ASE is investing $578.6 million in Kaohsiung to build a new packaging plant. In the EU's Chips Act pilot lines, advanced packaging has been designated as a key direction. It could even be said that, against the backdrop of Moore's Law slowing down, packaging technology is becoming the key variable that determines chip performance.

In addition, niche tracks such as photonics, SiC, and power ICs are also beginning to attract substantial investment. imec and TNO have opened a photonics laboratory in the Netherlands, and an EU-funded photonics center has landed in Brussels; Ranovus is investing $100 million in Canada to expand its optoelectronics plant. On the SiC front, in addition to onsemi and Infineon, ams OSRAM has also received Chips Act funding in Austria. These investments indicate that the semiconductor industry's technology landscape is diversifying, extending from purely silicon-based logic devices to optical interconnects, wide-bandgap semiconductors, and other extended fields.In terms of technological barriers, the most critical remains advanced-process equipment and processes. The lithography step is monopolized by ASML, and the introduction of High-NA EUV lithography machines will further raise the bar. The new materials and new processes brought by the GAA architecture also require foundries to have extremely high R&D investment and yield capabilities. For most countries, even with funding, it is difficult to establish a complete advanced-process supply chain in the short term.

III. Industry Chain Impact: Upstream Benefits, Midstream Diverges, Downstream Demand Continues to Surge

Looking at the broader industry chain, the impact of this round of investment is layered.

Upstream equipment and materials are undoubtedly the biggest winners. New factories around the world mean huge demand for etching machines, thin-film deposition, metrology and inspection equipment, and more. ASML's $164 million investment in an office building in South Korea is precisely to get closer to Samsung and SK Hynix and provide faster service response. Equipment makers such as Applied Materials, Lam Research, and KLA are expected to receive record orders in the coming years. On the materials side, demand for silicon wafers, photoresists, and specialty gases will also grow in tandem, but some material supply chains are highly concentrated in Japan and Germany, which to some extent intensifies regional risk.

The midstream manufacturing segment is showing clear divergence. Advanced logic foundry and memory manufacturing are enjoying the AI dividend, but capacity in mature processes (28nm and above) is expanding rapidly and may face oversupply in the future. For example, China's continued investment in mature processes and the entry of new players such as India may trigger price wars in the mature-process market in 2027-2028. In addition, the IDM model is also under pressure—Intel's cancellation of its plant plans in Germany and Poland serves as a footnote to the transformation difficulties of the traditional manufacturing model.

Downstream packaging and testing is being revalued due to advanced packaging. ASE, Amkor, and TSMC's packaging division have become key links in AI chip delivery. In the future, the boundaries between packaging and testing companies and foundries will become even blurrier. TSMC's huge investment in packaging may further squeeze the living space of independent packaging and testing companies.

IV. Competitive Landscape: TSMC Far Ahead, a Multipolar World Begins to Emerge

The investment dynamics of 2025 have essentially established the future competitive landscape.

TSMC remains the center of global semiconductor manufacturing. It is expanding six wafer fabs in Taiwan, adding $100 billion in the United States, continuing to advance its second fab in Japan, and aggressively expanding advanced packaging capacity. This strategy of "Taiwan core + global branches" allows TSMC to take the initiative amid geopolitical risks while also making it a target that every country competes to attract. Samsung Electronics is betting on both memory and foundry, but the gap between its foundry business and TSMC has not narrowed; Intel, meanwhile, has fallen into financial difficulties, and its negotiations with the U.S. government over a 10% stake reflect the trend of traditional IDMs becoming nationalized.The new variable worth watching is Japan. Although Rapidus's 2nm trial production is still far from mass production, it marks Japan's restart in advanced process technology. Micron's investment in building an AI memory fab in Japan has injected a shot in the arm for the country's memory industry. In addition, Taiwan's ASE and Europe's imec also play unique roles in the industry chain through the influence of research institutions.

In mainland China, companies such as SMIC and YMTC remain resilient under export controls. YMTC has announced the construction of a third memory fab, and Huawei and SMIC are jointly advancing 5nm. Despite equipment restrictions, the process of domestic substitution has not stopped. It is foreseeable that two major technology ecosystems will take shape: one led by the United States and its allies in advanced processes, and the other led by China in a self-controllable ecosystem (focusing on mature process nodes and specialty processes).

V. Regional Perspective: Accelerating Supply Chain Decentralization

From a regional distribution perspective, this wave of global investment is shifting the semiconductor supply chain from a high concentration in East Asia (especially Taiwan and South Korea) toward multi-regional dispersion.

The United States remains the focal point of global semiconductor policy. The Trump administration's implementation of the CHIPS Act has been full of uncertainties: establishing an investment accelerator, canceling the contract with Natcast, and having NIST take over—the policy swings have left companies at a loss. However, TSMC's additional $100 billion investment, Apple's $500 billion plan, and Micron's $30 billion increase all indicate that the prospects for U.S. domestic manufacturing are still viewed favorably. Still, the gaps in mature process nodes and packaging and testing remain enormous.

Europe is accelerating its catch-up, driven by the Chips Act. Dresden, Germany, has become Europe's "silicon heart": Infineon and GlobalFoundries have received a total of approximately €1.5 billion in government subsidies; the Czech Republic has attracted onsemi's SiC fab; and Austria's ams OSRAM has obtained funding. But there are also shadows: Wolfspeed canceled its fab construction, and the GF-ST project in France has stalled, showing that Europe's appeal in advanced processes is limited, with its comparative advantages lying more in specialty processes such as automotive chips, power devices, and photonics.

Japan, meanwhile, has regained vitality through cooperation with TSMC, Micron, and Rapidus. The Japanese government's subsidies for the semiconductor industry are even more generous than those of the United States, with the goal not only of restoring manufacturing capacity but also of ensuring supply chain security. South Korea still dominates the memory sector, with SK Hynix's Yongin cluster set to become the world's largest semiconductor industrial park, and Samsung is also accelerating foundry investment.

As for China, YMTC's third fab is planned to begin production in 2027, and SMIC and Huawei are targeting 5nm. However, constrained by the embargo on EUV lithography machines, the development path for advanced processes remains unclear. Notably, India approved a batch of four wafer fab projects in 2025; although small in scale, they mark South Asia's initial attempt to enter semiconductor manufacturing.Southeast Asia, although it has seen limited investment activity this year, is likely to become a destination for the relocation of packaging and testing as well as mature process nodes in the future, given its neutral geopolitical position and labor cost advantages.

VI. Investment Perspective: The Dual Drivers of Capital Expenditure and Government Support

From a capital market perspective, the cyclicality of the semiconductor industry is being smoothed out by the long-term growth of AI demand. The significant upward revisions in capital expenditure by TSMC, SK hynix, Micron, and others imply that production capacity supply will increase substantially in the coming years. However, fab construction cycles are long (typically 3-5 years), and if AI demand growth slows, structural oversupply could emerge in 2027-2028.

Government subsidies, on the one hand, reduce the risk of corporate investment, but on the other hand, they distort market signals — some "politically driven" projects (such as Intel Ohio and Wolfspeed Germany) ultimately stalled, demonstrating that policy support does not guarantee project success. Long-term investors need to focus on a company's technological competitiveness, yield ramp-up capabilities, and visibility into customer orders, rather than simply making judgments based on fab construction news.

VII. Long-term Outlook: Industry Transformation Over the Next Decade

Looking ahead three years, 2nm process technology will enter mass production with gradually improving yields, and advanced packaging will continue to expand. TSMC's global multi-site capacity will begin to be released, with fabs in Arizona, USA, and Kumamoto, Japan, producing actual chips. China's capacity expansion in mature process nodes will impact the global mid-range market, making price competition likely unavoidable.

Within five years, technologies such as GAA and backside power delivery will enter the mainstream. If Rapidus can achieve mass production of 2nm, Japan will re-emerge as one of the three poles of foundry supply. Europe's chiplet protocol stack could transform automotive chip design models. Meanwhile, the effects of reshoring advanced manufacturing in the United States will gradually become visible, but whether costs can be continuously reduced remains a challenge.

After ten years, quantum computing and photonic computing may bring disruptive changes to the semiconductor industry, but traditional silicon-based processes will still remain the core of information processing. The truly important change will be the "regional multipolarization" of supply chains — with North America, East Asia, and Europe each forming relatively complete semiconductor ecosystems. For semiconductor industry practitioners, understanding this divergence is more important than chasing every investment boom.

Conclusion: Sober Reflections Amid the Investment Boom

The unprecedented surge in global fab investment in 2025 is essentially a forward-looking deployment for AI-era computing demand, and also an inevitable result of supply chain restructuring driven by geopolitics. A few leaders such as TSMC, Samsung, and Micron have received the most resources, while more countries and regions are trying to find their own positions in specific niche segments.The industry judgments most worth paying attention to are threefold. First, advanced packaging has become a competitive high ground as important as process technology. Second, policy subsidies can determine factory construction decisions, but they cannot determine commercial success or failure; technological capability and customer stickiness are the long-term moats. Third, the global semiconductor supply chain is shifting from "efficiency first" to "security first," which means rising costs and redundant construction will be the norm for the next decade.

For corporate decision-makers, staying sober amid the surging wave of investment is more critical than diving in themselves. After all, every cyclical peak is accompanied by massive capacity elimination and misallocation of resources. History may not repeat itself simply, but the laws of the industry still hold.