TSMC’s AI chip boom sparks margin jitters, investor doubts

TSMC's financial performance reached unprecedented heights in mid-2026, with quarterly revenue soaring to approximately $39.6 billion—a 36% year-over-year increase—and an extraordinary 68% jump in June alone, fueled primarily by surging demand for advanced AI chips. This record-breaking growth underscores the company's pivotal role in powering AI infrastructure, as it manufactures cutting-edge processors for industry giants like Nvidia, Apple, and AMD, who have aggressively secured capacity for TSMC's next-generation 2nm technology slated for 2026.

The structural shift in TSMC's revenue mix toward AI and high-performance computing chips is profound, with these segments accounting for over 60% of revenue and advanced nodes (7nm and smaller) representing nearly three-quarters of wafer sales. CEO C.C. Wei highlighted the relentless strength of AI-related demand, prompting the company to raise its full-year 2026 revenue growth forecast to more than 30% in U.S. dollars, even as capacity constraints intensify amid fierce competition from hyperscalers like Nvidia, Apple, and AMD.

Despite TSMC's record-breaking revenue surge driven by the global AI infrastructure boom—expected to exceed $725 billion in spending this year—the company faces a significant capacity crunch, particularly in meeting American customer demand. CEO C.C. Wei candidly acknowledged that even with new U.S. manufacturing facilities coming online, TSMC will be unable to fully satisfy this demand for years, a bottleneck that simultaneously bolsters the company's pricing power and underscores the intense competitive dynamics shaping the semiconductor industry.

TSMC's cutting-edge 2nm process is already fully booked, underscoring the intense demand for advanced nodes driven by AI chip requirements. To address critical bottlenecks in advanced packaging, especially for AI accelerators, TSMC is rapidly expanding its CoWoS capacity by adding three new factories in Taiwan’s Chiayi Science Park, aiming to reach 170,000 wafers per month by 2027. Despite these efforts, current packaging capacity remains strained, forcing TSMC to outsource overflow orders to competitors like Intel’s EMIB technology and Taiwanese subcontractors such as ASE and SPIL, highlighting persistent supply chain challenges.

Leveraging its dominant position—controlling approximately 73% of the global pure-play foundry market and an even larger share of leading-edge capacity—TSMC is exercising significant pricing power amid these capacity constraints. Starting January 2027, the company plans to implement its first mature-node price increases in over three years, with advanced process nodes seeing hikes of 5% to 10%, and surcharges up to 15% for orders exceeding pre-negotiated volumes, potentially pushing total cost increases to 25%. This pricing strategy reflects both robust demand and the substantial cost pressures from expanding capacity, especially given that U.S. fab construction costs run four to five times higher than in Taiwan due to labor, infrastructure, and regulatory challenges.

TSMC’s aggressive capital expenditure plans, including a historic $265 billion U.S. investment and a raised 2026 capex guidance of $60–64 billion, are aimed at alleviating the advanced node and packaging bottlenecks by building multiple new fabs for 2nm-class and below technologies in the U.S., Japan, and Germany. The Chiayi Science Park’s development as a dedicated advanced packaging cluster is expected to generate significant economic impact, with an annual output valued at around US$9.35 billion and creating over 9,000 jobs, signaling TSMC’s strategic commitment to scaling capacity while navigating the complex cost and supply challenges inherent in next-generation semiconductor manufacturing.

TSMC's historic $265 billion U.S. investment plan, with a recent $100 billion increase focused on expanding its Arizona manufacturing campus to 12 fabs, exemplifies a strategic shift toward regionalizing semiconductor production. This move aims to mitigate geopolitical risks tied to Taiwan's vulnerabilities and align with U.S. government policies like the CHIPS Act, which provides $6.6 billion in direct funding and $5 billion in loans to support these projects. Despite the aggressive capital expenditure, TSMC has not committed to fixed timelines for all expansions, reflecting uncertainties amid construction delays and labor shortages that have pushed the second Arizona fab’s production start to late 2027.

The Arizona expansion specifically targets alleviating critical AI packaging bottlenecks through CoWoS advanced packaging technology, which currently limits AI accelerator shipments from major customers like NVIDIA and AMD. By regionalizing this capacity within the U.S., TSMC not only addresses supply chain resilience and geopolitical concerns but also meets domestic sourcing requirements of hyperscalers, creating a strategically distributed manufacturing footprint that balances efficiency with political acceptability. This transition to localized strategic infrastructure, however, comes with significantly higher costs—TSMC’s U.S.-made chips are estimated to cost 20% to 50% more than those from Taiwan—leading to margin dilution and increased pricing pressures on customers.

TSMC’s ramped-up capital expenditure for 2026, now projected between $60 billion and $64 billion with 70% to 80% allocated to advanced process technologies, underscores the scale and urgency of capacity expansion driven by a multi-year AI megatrend. This massive investment not only supports cutting-edge 2-nanometer and below fabs but also reflects a broader industry shift toward a geopolitically managed, capital-intensive semiconductor ecosystem. The company’s strategy balances the need to secure long-term capacity agreements with major customers like Nvidia, AMD, Apple, and cloud providers, while navigating operational complexities such as labor productivity, yield ramp, and cost control to transform Arizona from a symbolic foothold into a commercially scalable manufacturing hub.

Government policies and geopolitical risks are central drivers behind TSMC’s U.S. expansion, compelling the company to geographically duplicate supply chains despite the resulting higher costs and extended timelines. This regionalization strategy, supported by federal, state, and local governments, aims to capture a geopolitical premium, mitigate tariff and policy risks, and defend market share against competitors like Intel and Samsung. However, these benefits come with trade-offs: increased manufacturing costs, margin dilution estimated at 2% to 4%, and a more complex investment landscape where governments and customers play larger roles in decision-making, ultimately passing higher expenses down the semiconductor value chain to end users.

Despite TSMC's record-breaking revenue and profit growth fueled by surging AI chip demand, investor skepticism persists regarding the sustainability of this AI-driven boom. Market participants question whether the semiconductor giant and broader tech industry are overbuilding capacity that may not be fully utilized, especially as much of the aggressive capital expenditure—approaching $60 billion in 2026—is debt-funded and its long-term returns remain uncertain. This cautious sentiment is reflected in TSMC's stock performance, which has dipped despite strong earnings, underscoring doubts about whether the AI surge represents a durable growth driver or a transient bubble.

Investor concerns about margin dilution are mounting as TSMC ramps up production of advanced nodes like the 2nm technology and expands costly U.S. manufacturing facilities. CFO Wendell Huang has acknowledged that gross margin dilution from overseas fab ramp-ups could reach 3% to 4% over the next several years, a significant headwind given that U.S.-produced chips cost 20% to 50% more than those made in Taiwan. While TSMC reported a robust 67.7% gross margin in Q2 2026, these elevated production costs and ramp-up expenses are tempering investor enthusiasm and fueling worries about the long-term profitability of the AI investment surge.

TSMC’s capital expenditure strategy reveals a deliberate balance between aggressive growth and cautious spending, which may be contributing to investor skepticism. Although the company plans historic investments, including a $265 billion U.S. expansion, its CapEx-to-revenue ratio of 34-36% during the AI supercycle is notably lower than the 53% seen during the pandemic-era semiconductor boom. This restraint suggests management’s preference for maintaining a comfort level around 35% spending, possibly reflecting lessons learned from prior cycles or doubts about the long-term durability of AI demand, thereby signaling a more measured approach amid market uncertainties.