TSMC Breaks Its Own Capacity Rule With Another 3-Nanometer Fab and Opens the Door to Texas, While Guiding to Wider Overseas Margin Dilution
Third-quarter 2026 earnings call, October 15, 2026
TSMC reported third-quarter revenue of $46.7 billion, up 16.2% sequentially in US dollar terms and above the high end of guidance. The more important news was a change in capacity strategy. Management said it will add 3-nanometer capacity even though the node has already reached its targeted size. Chairman and CEO C.C. Wei acknowledged the break with precedent: "Historically, we do not add additional capacity to a node once it reaches the targeted capacity. However, based on our assessment, we are executing our global capacity plan to add 3 new fabs to support a robust multiyear pipeline of demand for 3-nanometer technologies."
A 3-Nanometer Build-Out Across Three Regions
The new Tainan Science Park 3-nanometer fab remains on track for volume production in the first half of 2027. TSMC has now decided to add a further 3-nanometer fab in Tainan, scheduled for the first half of 2028. In Arizona, construction of the second fab is complete and tool move-in is under way. Volume production is now slated for the third quarter of 2027, slightly ahead of the previous schedule. The second Japan fab will also use 3-nanometer, with volume production planned for 2028. TSMC is also converting 5-nanometer capacity to 3-nanometer in Taiwan and shifting capacity flexibly among N7, N5 and N3.
Wei said the company has about 25 fabs in progress worldwide, including 13 leading-edge and advanced packaging fabs in Taiwan over the next several years and another 5 to 6 overseas. N2 family annual capacity is expected to more than double in 2027. A14 risk production remains on track for 2027, with volume production in 2028. Wei said A14 customer engagement is running ahead of where N2 stood at the same stage, and he expects A14 to be a bigger node than N2. He declined to say how an IDM competitor's in-sourcing plans factor into that view, saying only that TSMC treats every customer fairly while taking internal capacity into account.
Management did not give a wafer capacity growth figure. Wendell Huang said it was "too early" and noted that advanced-node additions are the main driver of capacity growth. Wei said that after cross-checking customers, cloud service providers and data center readiness, demand still exceeds supply. He added that he could not say whether the gap will widen or narrow because customers keep raising their requirements.
Overseas Dilution Is Better This Year but Set to Widen
Overseas fab dilution to gross margin is now expected to be closer to 2% for full-year 2026, helped by scale and cost-improvement efforts. That benefit is temporary. Management expects the drag to widen to 3% to 4% in the next few years as more US fabs ramp. When pressed by UBS on whether this implied a faster third phase in Arizona, Huang said the 3% to 4% range applies to the later part of the five-year period, "and that still remains the case." He described the acceleration as happening "everywhere, including in the U.S."
Third-quarter gross margin fell 1.3 percentage points sequentially to 66.4%, which management attributed to the initial 2-nanometer ramp. The fourth-quarter guide of 65% to 67%, with a 66% midpoint, reflects another 40 basis points of dilution from the steep N2 ramp and from overseas fabs. Productivity gains and capacity optimization partly offset this. Operating margin is guided to 56.5% to 58.5%, versus 58.7% in the third quarter. N2 contributed 8% of wafer revenue, against 33% for N3, 28% for N5 and 10% for N7. Advanced nodes made up 79% of wafer revenue.
On pricing, Wei offered little detail on whether increases apply to contracted volumes or only to incremental capacity. He described pricing as "very strategic" and dependent on market segment, location and technology. He noted that US production will be priced differently from Taiwan. TSMC's stated aim is to have its value recognized. Asked about long-term profitability, Huang said: "There is no ceiling for TSMC's profitability, but a dramatic increase will not happen overnight either."
Texas Under Evaluation, With a Customer-Linked Fab Under Discussion
Wei confirmed that TSMC is looking beyond Arizona for US expansion, naming Texas as one location under evaluation. He also confirmed discussions with a customer that is building its own fab, to see whether that capacity could be used. Questions on the call referenced SpaceX and Tesla and the Terafab project, but Wei did not name the customer. He said no firm conclusion has been reached and that TSMC would share one if it does. His only constraint was that the fab "got to be run by TSMC." He added that the structure could take more than one form. Wei said US customers and the US government are both pushing for more domestic capacity.
AI Demand: No Revision to the Narrow Framework, but a Higher Bias
Management did not formally update its AI accelerator revenue CAGR, which stands at mid-to-high 50s for 2024 to 2029. Huang said, "We did not see any sign of slowdown, whether it's high 50% CAGR, it doesn't matter, it probably will be higher." Wei pointed to agentic AI, open-source model adoption and demand for AI accelerators, networking chips and agentic AI CPUs.
Agentic AI CPUs are not captured in the narrow AI accelerator definition. Wei said TSMC cannot reliably distinguish AI server CPUs from conventional server CPUs, and the definition will stay narrow. That means CPU strength will show up in reported revenue without moving the AI CAGR.
On power and other bottlenecks, Wei said TSMC now sends its own teams to check data center build-out and US power supply. He said the feedback is "so far so good." Power constraints have not led customers to change their adoption of backside power rail. Wei said that choice depends on product needs, with some accelerators using it and mobile products not.
Advanced Packaging and the GlobalFoundries Arrangement
Wei defended TSMC's packaging position against alternatives. TSMC is in production at 5x reticle size, with 9x and 14x in development. "I don't believe that anybody can do that today except TSMC," he said, adding that yields are excellent. CoWoS remains customers' first choice, and TSMC is also developing panel-based approaches to lower costs. Asked whether TSMC would serve as a backup if a competing packaging solution fell short, Wei would not answer directly but said the priority is that customers' products launch smoothly, which "gives you enough hint."
Wei acknowledged a shortage in the back end, which explains the agreement with GlobalFoundries to supply silicon interposers. He called it part of TSMC's back-end supply strategy and said the aim is to ensure customers are fully supported.
CapEx Stays Elevated, With Tool Inflation Acknowledged
TSMC reiterated its 2026 capital budget of $60 billion to $64 billion, after spending $13.8 billion in the third quarter. It continues to expect 2026 to 2028 CapEx to be significantly higher than in the prior three years. Huang acknowledged that "the tool price inflation is real," with material and labor costs also rising. He said the majority of spending is tied to long-term demand and that revenue growth should still outpace CapEx growth. Wei said TSMC has "no intention to leave food on the table for others." Operating cash flow was about TWD 920 billion in the quarter, against TWD 443 billion of CapEx, and cash and marketable securities stood at about $126 billion.
Consumer Weakness Tempers 2027 Planning
The call was not uniformly bullish. Wei said consumer and price-sensitive segments are being squeezed by rising component prices and economic uncertainty, and TSMC is being "prudent" in 2027 planning. Smartphone revenue still rose 10% sequentially to 21% of the total, while HPC grew 22% to 68%. Automotive rose 23%, and DCE fell 12%. For the fourth quarter, TSMC guided revenue to $52.3 billion to $54.3 billion, up 14% sequentially and 58% year over year at the midpoint, assuming 31.8 TWD per US dollar. Full-year 2026 growth is now expected at about 44% in US dollars.
High-NA Timing and Wafer Allocation
Asked whether adopting high-NA EUV around 2030 leaves TSMC behind competitors, Wei said TSMC will deploy the tool only when it offers the best technology, reasonable cost and the best maturity. He said TSMC is working with partners toward a 12-inch photomask by 2031. He described high-NA as "a very good tool" with advantages at smaller geometries, but said TSMC does not see a competitive disadvantage.
On the secondary wafer market, Wei said TSMC does not allow customers to sublet or auction capacity. Limited collaboration is possible under master licensing agreements, but the volumes are "very, very small."