From TSMC’s Earnings to the U.S.–Taiwan Tariff Deal: AI, Capital Flows, and the Geopolitical Reordering of Semiconductors
Original Article by SemiVision
Record-High Q4 and Full-Year 2025 Results; Optimistic Outlook for Q1 2026
TSMC delivered its strongest results in history in Q4 2025. Consolidated quarterly revenue surpassed the NT$1 trillion milestone for the first time, reaching approximately NT$1.046 trillion (about USD 33.7 billion), while net income also exceeded NT$500 billion for the first time, at roughly NT$505.7 billion (about USD 16.3 billion). Earnings per share for the quarter came in at NT$19.50.
Compared with the same period in 2024, Q4 revenue grew 20.5% year over year, while net profit surged 35.0%. Driven by robust demand for advanced nodes and continued optimization of product mix, gross margin climbed to 62.3%, setting a new historical high and significantly exceeding the company’s prior guidance range of 59%–61%. This margin level even edged above the previous record of 62.22% in Q4 2022, underscoring TSMC’s strong pricing power in advanced manufacturing, despite rising depreciation from massive capital investments.
On a full-year basis, 2025 consolidated revenue reached NT$3.809 trillion (approximately USD 122.9 billion), representing 31.6% year-over-year growth.
Full-year gross margin averaged 59.9%, up 3.8 percentage points from the prior year.Net income hit a record NT$1.7178 trillion (around USD 55.4 billion), with EPS of NT$66.25, marking the strongest annual performance in TSMC’s history.
Supported by these exceptional results, TSMC also delivered a confident outlook for early 2026.
The company expects Q1 2026 revenue to reach another all-time high, in the range of USD 34.6 billion to USD 35.8 billion, representing more than 4% sequential growth. Gross margin is projected to remain elevated at 63%–65%, with operating margin of 54%–56%, indicating that near-term profitability will stay resilient even as depreciation expenses rise with aggressive capacity expansion.
In addition, TSMC announced a significant upward revision to its capital expenditure plan. For 2026, capex is now expected to reach USD 52–56 billion, up roughly 30% from the USD 40.9 billion invested in 2025, setting a new historical record. This level of investment far exceeds prior market expectations and reflects management’s strong conviction in sustained multi-year demand growth.
TSMC emphasized that even with record-high capital spending in 2026 to position the company for future growth, it remains committed to delivering long-term earnings expansion and shareholder value. Notably, approximately 70%–80% of capex will be allocated to advanced process technologies, around 10% to specialty technologies, and 10%–20% to advanced packaging, testing, mask production, and other supporting infrastructure.
This allocation highlights a clear strategic focus: concentrating capital on the most advanced technologies to meet surging market demand—particularly from AI and HPC—while maintaining TSMC’s long-term technology leadership.
AI Demand Surge: C.C. Wei on Its Authenticity and Investment in Advanced Technologies
In response to market skepticism that the recent AI boom could be a “bubble,” TSMC Chairman and CEO C.C. Wei stated unequivocally at the earnings call that AI demand is real and sustainable, and that the company has strong confidence in its long-term trajectory. Wei revealed that over the past three to four months, he personally engaged in intensive discussions with key customers to verify whether demand for AI-related chips is supported by a solid long-term foundation. The outcome, he said, was reassuring: customers did not merely offer verbal assurances, but also presented concrete data showing that AI applications are actively driving healthy business growth and generating substantial returns. He humorously added that after reviewing these customers’ financial statements, he found that “they are making even more money than TSMC,” drawing lighthearted laughter from the audience. This further underscored the significant profitability downstream service providers are enjoying from the AI wave. Taken together, TSMC is clearly not concerned that AI demand is illusory—the strong order momentum from end customers such as cloud service providers (CSPs) provides compelling evidence that the AI boom rests on a solid foundation.
Driven by this powerful demand, TSMC has revised upward its growth outlook for the coming years. Wei announced that the compound annual growth rate (CAGR) of revenue from AI accelerators between 2024 and 2029 is now expected to reach 54%–56%, significantly higher than the previous estimate of 45%. Rapidly growing AI chip orders have already lifted AI accelerator–related revenue to approximately 17%–19% of TSMC’s total wafer revenue in 2025, making it one of the company’s most important growth engines. Wei further projected that, fueled by AI demand, TSMC’s overall revenue CAGR over the five years starting in 2024 could reach around 25%, well above the earlier expectation of 20%. He emphasized that there are currently no signs of weakening AI demand; on the contrary, growth remains exceptionally strong with virtually no indication of deceleration. As a result, 2026 is expected to be another year of robust growth for TSMC, with the company confident that revenue growth in U.S. dollar terms could approach 30%. Overall, TSMC’s management holds a firm conviction in the long-term demand outlook for AI, viewing it not as a short-lived bubble but as a powerful new pillar capable of supporting revenue growth for many years to come.
To meet surging AI demand, TSMC is aggressively expanding its most advanced manufacturing and packaging capabilities. On the 2-nanometer (N2) process, Wei announced that development is progressing smoothly: N2 entered volume production as scheduled in Q4 2025 at both the Hsinchu and Kaohsiung sites, with initial yields exceeding expectations. Orders from smartphones, high-performance computing (HPC), and AI accelerators are ramping rapidly, and 2026 is expected to mark the first major wave of capacity expansion for 2nm. Continuing its comprehensive process-family strategy, TSMC plans to follow N2 with enhanced N2P variants that deliver improved performance and power efficiency, as well as the A16 advanced node, which introduces a new power delivery architecture tailored for high power-density AI and HPC applications. Industry observers interpret A16 and similar nodes as targeting data center energy-efficiency bottlenecks, signaling that AI computing is shifting from simply “stacking more compute” toward a stronger emphasis on efficiency. TSMC has also acknowledged that the capital required for fab construction and equipment at the 2nm generation is significantly higher than for prior nodes. However, strong market demand for high-performance, high-density chips positions the N2 family to become a cornerstone technology supporting AI and HPC applications over the next decade. In short, the AI megatrend is driving TSMC to double down on the most advanced process technologies, ensuring it remains ahead in the next generation of semiconductor competition.
Beyond wafer fabrication, TSMC is also intensifying its investment in advanced packaging. As AI high-performance chips increasingly rely on heterogeneous integration—such as high-bandwidth memory (HBM) stacking—TSMC’s advanced packaging capacity has at times struggled to keep pace with demand in recent years. Wei noted that TSMC maintains a leading position in advanced packaging technologies and that market demand remains robust. In the 2026 capital expenditure plan, more than 10% of total spending will be allocated to advanced packaging, testing, mask production, and related areas. This indicates that TSMC is accelerating the expansion of advanced packaging capacity, including high-end wafer-level stacking solutions such as CoWoS, InFO, and SoIC, to eliminate bottlenecks and meet customer demand. At the same time, this expansion is expected to benefit the broader ecosystem of equipment and materials suppliers.
Market analysts point out that TSMC’s massive capital spending effectively signals to the market that it is backed by long-term customer commitments. Historically, TSMC has been highly disciplined, refraining from aggressive capacity expansion without clear visibility into long-term orders and customer prepayments. The decision to raise 2026 capex to as much as USD 56 billion, with more than 70% allocated to advanced nodes, sends a strong message: the global cloud giants’ arms race in AI compute is not cooling off—instead, it is pulling forward capacity investments to secure supply for 2027–2028 and beyond. Looking ahead, by scaling advanced process and packaging capacity, TSMC is set to further solidify its central role in the AI chip supply chain and ensure that it fully captures the long-term value of this transformative technology wave.
Accelerating U.S. Fab Expansion: Scaling Capacity to Meet AI Customer Demand
Notably, as TSMC undertakes a major capacity expansion, its overseas fab construction plans are also being accelerated in parallel—most prominently in Arizona, where progress has drawn significant market attention. C.C. Wei revealed that recent media reports suggesting TSMC may add four to five additional fabs in the United States are not without basis. He confirmed that TSMC has recently acquired a second large parcel of land in the U.S., stating candidly, “That is a hint—we need this land.” This move clearly signals that the company is planning further expansion to meet the continuous and growing order demand from U.S. customers. Wei explained that the primary driver is the strong pull from major U.S. AI customers, who are requesting substantial capacity support, leaving TSMC little choice but to accelerate fab expansion in order to keep pace with customer needs. Fueled by this demand, TSMC’s U.S. manufacturing footprint is rapidly expanding, taking shape as an independent GigaFab cluster.
According to Wei, the first Arizona fab (Fab 21 Phase 1) successfully entered volume production in Q4 2024. Although the project previously faced concerns over delays due to labor availability and construction timelines, it is now producing as scheduled, demonstrating TSMC’s ability to overcome execution challenges and reach key milestones. The second fab (Phase 2) at the same site has completed its main construction and is expected to begin tool installation in 2026, with volume production pulled in to the second half of 2027. Originally, the Phase 2 ramp was expected to occur later, but in response to strong and urgent demand from U.S. customers, TSMC decided to accelerate the production timeline. In addition, construction of a third fab is already underway, while applications have been submitted for a fourth fab and the first advanced packaging facility at the Arizona site. These developments indicate that multiple new production facilities will be added to the Arizona campus over the coming years. To support this large-scale expansion and maintain long-term flexibility, the recently acquired second land parcel is intended to accommodate future phases of growth. Ultimately, TSMC envisions transforming the Arizona site into a fully integrated fab cluster, spanning from wafer manufacturing to advanced packaging, and designed to serve rising demand for advanced-node technologies from smartphone, AI, and HPC customers.
Wei also noted that yields and defect density levels at the Arizona fabs are now approaching those of TSMC’s Taiwan fabs, a highly encouraging signal for overseas expansion. Previously, the market had expressed concerns that U.S.-based manufacturing might suffer from lower yields and higher costs due to a lack of local experience. Current results, however, indicate that TSMC has successfully transferred a significant portion of its manufacturing expertise and operational discipline to the U.S., achieving product quality close to Taiwan benchmarks. This suggests that while overseas fabs inherently carry higher costs, there has been no material compromise in quality or efficiency, and further optimization remains possible as operational experience accumulates.
Beyond the U.S., TSMC’s specialty-technology fab in Kumamoto, Japan, entered volume production in late 2024, with a second Japan fab currently under construction, while preparations are also underway for its European fab in Dresden, Germany. Together with the continued build-out of multiple 2nm fabs in Taiwan, these overseas sites form a new global manufacturing footprint for TSMC. Wei emphasized that all overseas expansion decisions are fundamentally customer-driven, reflecting the desire of global customers for greater regional supply-chain flexibility. At the same time, TSMC factors in the importance of adequate government support to ensure long-term shareholder value. With strong collaboration from advanced U.S. customers as well as federal, state, and local governments, TSMC is accelerating capacity expansion in Arizona, with all projects progressing smoothly and in line with plan. Taken together, U.S. government policy and customer demand are reinforcing one another, driving TSMC’s U.S. fab strategy to expand well beyond the originally planned two fabs and positioning the company to meet the massive chip demand of the AI era ahead.
U.S.–Taiwan Tariff Agreement Reached: Tariffs Cut to 15% and a USD 500 Billion Investment Commitment
Coinciding with the day TSMC released its strong earnings results (January 15, 2026), a major breakthrough emerged in U.S.–Taiwan economic and trade relations. The U.S. Department of Commerce announced that the United States and Taiwan had reached a new trade agreement under which the U.S. will reduce the “reciprocal tariff” rate applied to Taiwanese goods from 20% to 15%. In exchange, Taiwan’s technology sector will significantly expand its investment footprint in the United States. The news, first reported by outlets such as The New York Times and CNBC, was quickly confirmed by both governments and is widely viewed as a milestone in U.S.–Taiwan economic cooperation.
