Somewhere in the Arizona desert, concrete is being poured for what will become one of the most advanced semiconductor fabrication plants ever built on American soil. Thousands of miles away, in southern Taiwan, construction crews are working on yet another cleanroom expansion. In Kumamoto, Japan, a second fab is rising next to one that only recently started shipping chips. And in Dresden, Germany, ground has been broken on TSMC’s first European plant.
Taiwan Semiconductor Manufacturing Co. is not building one factory. According to a tally of the company’s public announcements, earnings disclosures, and government filings, TSMC has roughly 18 fabrication and advanced packaging facilities in various stages of construction simultaneously. No chipmaker has ever attempted anything at this scale, and the reason is straightforward: artificial intelligence has created a demand spike so severe that the world’s most important chip manufacturer is spending more than $38 billion a year just to keep up.
The scale of what TSMC is building
TSMC’s latest annual filing with the U.S. Securities and Exchange Commission, covering fiscal year 2025, lays out the financial backbone of this expansion. The company guided capital expenditure in the range of $38 billion to $42 billion for 2025 alone, a figure that has roughly doubled since 2021. The filing identifies surging demand from AI and high-performance computing applications as the primary driver and flags the operational complexity of running parallel construction programs across multiple countries.
The buildout spans at least four countries. In Arizona, TSMC is constructing three fabs near Phoenix, with the first already producing chips on its N4 process node. In Japan, the company’s JASM joint venture has one operational fab in Kumamoto and a second under construction, with a third facility planned. In Germany, TSMC’s ESMC joint venture broke ground on a fab in Dresden. And in Taiwan itself, multiple new fabs and advanced packaging plants are being built or expanded in Kaohsiung, Taichung, and other sites to support the company’s most cutting-edge process nodes, including N2 and its successors.
The precise count of 18 simultaneous projects is assembled from these individual announcements rather than from a single company document. TSMC has not published a consolidated list. But the sheer number of disclosed sites, combined with the capital expenditure figures in the SEC filing, makes clear that the company is running one of the largest parallel construction programs in industrial history.
Why advanced packaging is the real bottleneck
Building transistors is only half the problem. The AI chips designed by Nvidia, AMD, Apple, and Broadcom increasingly rely on advanced packaging, a process that stacks and connects multiple chiplets into a single high-performance module. TSMC’s CoWoS (Chip-on-Wafer-on-Substrate) packaging technology has become the industry standard for the most powerful AI accelerators, and demand for CoWoS capacity has consistently outstripped supply since 2023.
The U.S. Department of Commerce’s CHIPS Program award page for TSMC’s Arizona project confirms that federal incentives, totaling up to $6.6 billion in direct funding plus approximately $5 billion in proposed loans, are tied to specific commitments that go beyond wafer fabrication. The award terms require TSMC to build advanced packaging and testing capabilities at the Phoenix site, a move designed to localize more of the AI chip value chain on American soil.
This is a significant shift. Historically, nearly all advanced packaging for leading-edge chips happened in Taiwan. The CHIPS Act award effectively requires TSMC to replicate part of that capability in Arizona, creating redundancy that serves both commercial and national security interests. The binding nature of these commitments, enforceable by a federal agency with the power to withhold subsidies, gives them more weight than a typical corporate press release.
The risks TSMC itself is flagging
TSMC’s SEC filing is not just a growth story. It is also a catalog of risks, written under the legal constraints of securities law, where officers and directors face personal liability for material misstatements. Several of those risks bear directly on whether this expansion can be executed on schedule.
Equipment supply: Building multiple advanced fabs simultaneously requires extreme ultraviolet (EUV) lithography machines, each costing upward of $350 million, manufactured exclusively by the Dutch firm ASML. The filing acknowledges supply-chain risk in general terms but does not isolate EUV availability as a discrete bottleneck. Industry analysts have flagged ASML’s production capacity as a potential constraint, given that Intel, Samsung, and other chipmakers are also placing large orders.
Labor: Operating greenfield fabs and packaging plants across four countries requires thousands of engineers, technicians, and construction workers with highly specialized skills. TSMC has faced well-documented challenges recruiting and retaining workers at its Arizona site, where cultural friction and compensation disputes slowed early progress. The filing acknowledges workforce challenges broadly but does not detail recruitment plans for each location.
Execution complexity: Running 18 construction projects in parallel is not 18 times harder than running one. It is qualitatively different. Each site involves local permitting, environmental review, utility infrastructure, and coordination with equipment vendors whose delivery schedules are already stretched. A delay at one site can cascade if it ties up shared engineering resources or diverts management attention.
Demand uncertainty: The filing references strong demand from AI applications but does not break out order data by customer or product line. That opacity matters. If AI training workloads plateau, shift to new architectures, or consolidate around fewer chip designs, some of the capacity TSMC is building could arrive into a softer market than current projections suggest.
How TSMC’s bet compares
TSMC is not the only chipmaker expanding, but it is operating at a different scale. Intel has committed roughly $100 billion over five years to rebuild its manufacturing competitiveness, including new fabs in Ohio, Arizona, and Germany. Samsung is expanding its foundry operations in Taylor, Texas, and Pyeongtaek, South Korea. But neither company matches TSMC’s combination of leading-edge process technology and volume. TSMC manufactures more than 90% of the world’s most advanced logic chips, a dominance that makes its expansion plans uniquely consequential for the AI industry.
The competitive dynamic also explains the urgency. Every major AI chip designer, from Nvidia to Apple to Amazon’s Annapurna Labs, depends on TSMC’s most advanced nodes. If TSMC cannot deliver enough capacity, these companies cannot ship products. That dependency gives TSMC enormous pricing power but also enormous responsibility. A construction delay in Kaohsiung or a staffing shortfall in Phoenix does not just affect TSMC’s quarterly earnings. It ripples through the entire AI supply chain.
What to watch over the next 12 months
The gap between TSMC’s ambition and its execution is where the real story will unfold through the rest of 2026 and into 2027. Several markers will signal whether this expansion is on track or running into trouble.
First, capital expenditure updates in TSMC’s quarterly earnings calls will show whether spending is accelerating, holding steady, or being trimmed. Any reduction in capex guidance would be an early warning that demand assumptions are shifting. Second, CHIPS Act milestone reports, which the Department of Commerce is required to track, will reveal whether the Arizona advanced packaging commitments are being met on schedule. Third, ASML’s order book and delivery timelines will serve as a proxy for whether EUV equipment is flowing fast enough to keep multiple fabs on track.
For now, the verified elements are substantial. TSMC has committed tens of billions of dollars in capital, made binding promises to the U.S. government, and disclosed to its shareholders that AI-driven demand is reshaping its entire manufacturing strategy. The company is building across four countries at a pace that has no precedent in the semiconductor industry. Whether all the moving parts, from lithography machines to construction crews to customer orders, align in time is the question that will determine whether this expansion becomes a landmark in industrial history or a cautionary tale about building for a boom that shifted before the concrete dried.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
