The most significant message from TSMC’s latest earnings conference wasn’t the usual “Artificial Intelligence is doing well,” but something more uncomfortable for those looking for signs of a slowdown: the advanced chip market is hitting physical and capacity limits, and the company believes the issue is no longer measured in quarters, but in years.
On January 15, 2026, the world’s largest contract semiconductor manufacturer reported Q4 2025 figures that reinforce this perspective. TSMC announced consolidated revenues of 1.046 trillion New Taiwan Dollars and a net profit of 505.740 billion New Taiwan Dollars, with a gross margin of 62.3% and a net margin of 48.3%. In U.S. dollars, quarterly revenue reached $33.73 billion, a 25.5% year-over-year increase. The snapshot is further highlighted by a strategic detail: the share of advanced technologies (7 nm and below) accounted for 77% of wafer revenue, with the 3 nm node contributing 28% and the 5 nm node 35%.
Starting from this baseline, the company provided guidance for the first quarter of 2026 that emphasizes a “structural ceiling” in margins: expected revenues between $34.6 billion and $35.8 billion, with a gross margin between 63% and 65%. In other words: TSMC isn’t just selling more; it’s selling the higher-margin products, in an environment where demand for advanced computing has turned into capacity allocation.
The major headline is CapEx: over $50 billion for manufacturing in 2–3 years
The figure that truly shifted the landscape was the projected investment for 2026. The management increased its capital expenditure budget to a range of $52 billion to $56 billion. During the call, the financial team framed this decision within an industry typically accustomed to cycles: in 2025, TSMC already invested around $40.9 billion (up from $29.8 billion in 2024) “anticipating” future growth, and now it’s doubling down.
But the key isn’t just volume—it’s the destination. According to the company, between 70% and 80% of CapEx in 2026 will focus on advanced process technologies; about 10% will go toward specialized technologies; and between 10% and 20% will be allocated to advanced packaging, testing, mask fabrication, and other areas. Simply put: money is flowing where the system is constrained, not where marketing dictates.
This shift is also reflected in another part of the call: TSMC warned that tool costs are rising and complexity is increasing, a reminder that “supply” isn’t just a matter of budget. It depends on clean rooms, equipment, power supply, water, logistics, and increasingly, advanced packaging capacity.
“I’m nervous”: validation is no longer just from customers, but “the customer of the customer”
In an unusual move for a company of this size, CEO C.C. Wei directly voiced the risk: the group is “nervous” because investing $52 billion–$56 billion could become a disaster if demand is misinterpreted. He then explained how they’re trying to mitigate this risk: by engaging not only with their immediate customers but also with the “customers of their customers,” especially major cloud service providers.
This part of the presentation was almost confessional. Wei stated that after those meetings, he was satisfied with the business evidence provided by hyperscalers. He even joked that he reviewed his own financial situation: “they are very rich.” The subtext is clear: TSMC is seeking signals of end-demand, not just orders along the immediate supply chain.
The company also made it clear that operational capacity is “very tight,” and the goal for 2026 and 2027 is to narrow the gap between supply and demand. They also noted that customer commitments are planned 2 to 3 years in advance, with visibility extending toward supply needs for 2028 and 2029. In an industry where building a factory can take several years, this shifts the debate: it’s no longer about “whether demand exists,” but “how quickly supply can catch up—and which actors capture value along the way: advanced nodes, equipment, packaging, and industrial productivity.”
Arizona, energy, and data centers: the slowdown isn’t where many expected
One of the fears surrounding the AI race is that data center deployment may hit energy and cooling limits. During the call, Wei assured that these plans are being considered: he said major operators have been planning their electrical supply for 5–6 years, and that TSMC is reviewing not only energy but also rack supplies and cooling systems. His conclusion, for now, was reassuring: “Everything is going well.”
Meanwhile, the CEO sought to lower the “risk discount” some investors apply to expansion in the U.S. He explained that in Arizona, performance and defect density are “nearly at Taiwan levels,” and the company is accelerating its U.S. expansion to meet AI demand, much of which is concentrated there.
The business mix confirms a shift to a new era
The revenue structure itself illustrates this change. In Q4, high-performance computing (HPC)—where much of the AI-related business is grouped—represented 55% of revenues, compared to 32% for smartphones and 13% for the rest (automotive, IoT, and others). Essentially, TSMC is increasingly less exposed to the “cheap electronics” cycle and more aligned with advanced computing cycles.
Wei also pointed to an important insight for consumer electronics: although memory costs may rise, high-end mobile devices are less sensitive and maintain solid demand. Conversely, more “affordable” products tend to suffer earlier when components increase in price or inventory restrictions occur.
What 2026 means: it’s not a slowdown, but a race for supply
In its long-term outlook, TSMC summarized the moment with two key points. First, that AI accelerators already accounted for a “high double-digit” percentage of revenues in 2025. Second, that it now expects that segment to grow at an “average-high 50%” annual rate from 2024 to 2029, raising its overall long-term growth forecast to around 25% CAGR during this period.
In market language: the AI cycle isn’t seen as a short wave but as a regime change. And so the debate for 2026 shifts from “will demand happen” to “how quickly can supply catch up—and which players capture value along the way: advanced nodes, equipment, packaging, and industrial productivity.”
Frequently Asked Questions
Why is it so important for TSMC to invest $52–$56 billion in 2026?
Because this investment translates into actual manufacturing capacity within 2–3 years. It’s a signal that the company sees sustained demand for advanced chips, especially for AI, and that the bottleneck is in production, not sales.
What does it mean that 70%–80% of CapEx goes to “advanced nodes” (7 nm and below)?
It indicates that the priority is expanding capacity for the most cutting-edge chips (AI, HPC, premium mobiles). This decision also benefits the supply chain for tools and materials needed for high-complexity processes.
Is energy really not the main bottleneck for AI data centers?
TSMC stated that major cloud operators have been planning their electrical supply years in advance, and the company is reviewing racks and cooling systems as well. Their message now is that the immediate bottleneck remains “silicon” (manufacturing capacity).
What role does advanced packaging play in the new AI race?
TSMC said that advanced packaging accounts for around 10% of contributions and will grow faster than the overall group. Additionally, CapEx allocated to packaging, testing, and masks could reach 20%, reflecting that final performance increasingly depends on how chips are integrated and assembled, not just the node technology.
via: Jukan on TSMC