Earnings Forecast | Is TSM.US (TSM.US) on its way to a market capitalization of $3 trillion right around the corner?
The Zhitong Finance App learned that Taiwan Semiconductor Manufacturing Company (TSM.US) will announce the latest full quarterly results before the US stock market on Thursday. Despite many negative macroeconomic and industry factors, TSMC's excellent third-quarter revenue performance highlighted the strong resilience of AI infrastructure demand.
The increasing restrictions on advanced process nodes and packaging technology in 2027 highlighted the expansion of production capacity as a key driver of TSMC's growth and increased profitability. While overseas expansion poses the risk of profit margin dilution, higher utilization, operating leverage, and rising wafer prices may offset the impending disadvantages.
Revenue has been “filed ahead of schedule”: US$46.7 billion, surpassing financial estimates
TSMC has yet to release its full quarterly report for the third quarter; revenue figures are already out first. According to the revenue report released by TSMC on October 8, consolidated revenue for September was approximately NT$511.857 billion, a monthly decrease of 0.6% and an annual increase of 54.6%, a record high for the same period of the previous year; 514.806 billion yuan in August was still the highest record in a single month. Consolidated revenue for the third quarter reached NT$1,494.243 billion, a quarterly increase of 17.62%, a record high in a single quarter. Cumulative revenue for the previous September was 3,898.727 billion yuan, an annual increase of 41.1%.
Based on the exchange rate of 1 US dollar to NT$32 based on the company's financial estimates, the third-quarter revenue was about 46.7 billion US dollars, which not only exceeded the company's previously estimated financial range of 44.6 billion US dollars to 45.8 billion US dollars, but also higher than the average market consensus value of 45.8 billion US dollars. According to the analysis, this record revenue provided a strong rebuttal to recent market concerns about “whether AI infrastructure spending will be sustainable” — concerns previously accumulated due to rising AI-related debt, the Federal Reserve's restarting interest rate hikes, and increased industry scrutiny of AI security.
TSMC has decided to hold a performance conference call on October 15 (Thursday); as is customary, the 10 days (October 5 to 14) before the announcement of the financial report will be a period of silence. The Taiwanese corporation expects revenue to rise further in the fourth quarter, benefiting from Vera Rubin's new platform and 2nm demand. The focus is on capital expenditure, progress in advanced processes and advanced packaging, gross margin outlook, and global production capacity layout.
AI infrastructure concerns, and a reverse signal that has already been implemented
Over the past few months, the overall performance of AI semiconductors has weakened against industry headwinds. This sensitivity was revealed in early September: Anthropic CEO Dario Amodei (Dario Amodei) issued an open letter calling for a slowdown in cutting-edge model development on the grounds of AI safety, which immediately triggered a sharp drop in semiconductor indicators. Both the VanEck semiconductor ETF (SMH) and the triple-multiplied semiconductor ETF (SOXL) were clearly damaged. Related concerns have since eased — Nvidia (NVDA.US) Wong In-hoon and AMD (AMD.US) Su Zifeng both insisted on continuing but responsibly advancing AI; the industry's expectations for incremental computing power to assess and monitor workloads have also strengthened the sustainability of infrastructure spending.
However, the latest setback on projects that have already started has revived doubts: Oracle (ORCL.US) issued a force majeure notice to developers regarding Project Jupiter, a 2.45 gigawatt data center campus in New Mexico, hoping to defer part of payment obligations when delivery of the park is delayed due to external factors such as electricity and approval. According to reports, the crux is that the right to pass through the gas supply pipeline was denied twice by the state of New Mexico, and the commencement of pipeline construction was delayed until February 1, 2027. The transaction price of the $18 billion syndicated construction loan for the project has already fallen below 90% of face value. This also shows that the restrictions on the implementation of AI computing power are shifting from chip supply to electricity and approval.
Main line 1: Shortage of production capacity in 2027 and “can demand be turned into revenue”
The drive logic on the demand side has changed gears. AI infrastructure requirements are increasingly dependent on overall token consumption, and intelligent asana AI and its continuous inference cycle between task planning, execution, and evaluation amplify this consumption. Industry estimates show that post-training loads may require up to 30 times the computing power of the original basic model; workloads with higher inference intensity may consume 100 times more computing power than simple tasks such as email summaries. Dell (DELL.US) expects inference tokens to grow 87 times to 360 trillion dollars by 2030, and this outlook was later echoed by AMD.
Increased token consumption means more resilient wafer demand for TSMC, and also combines two forces: first, switching to next-generation commercial processors — Nvidia's Vera Rubin system, AMD's Instinct MI450X accelerator, and Zen 6 EPYC Venice server CPU, all designed around increasingly complex intelligent inference loads; second, the adoption of custom ASICs is expanding, and hyperscale manufacturers are increasingly inclined to customize architectures according to workloads to optimize inference units at scale Economical.
As a result, supply-side tension has intensified. TSMC management has previously stated that demand will remain “very strong” in 2029 and 2030; Wei Zhejia emphasized at last quarter's French conference that customers and their customers (that is, AI data center operators) continue to release strong prospects, advanced process production capacity is still in short supply, and the company is not worried about Intel and Samsung taking orders.
The bottlenecks are concentrated on the most sought after and most scarce N3 nodes. According to the original article, N3 accounts for about 30% of TSMC's wafer revenue portfolio. It is a technology that supports next-generation AI processors such as Nvidia's latest Vera Rubin architecture, and incremental production capacity is still significantly limited. According to the French press conference last quarter, advanced manufacturing processes of 7 nm and below together accounted for 77% of wafer revenue (30% for 3 nm, 33% for 5 nm, 11% for 7 nm), and 2 nm contributed 3% for the first time. The company is speeding up the expansion of N3 production capacity in Taiwan, Arizona and Japan, and is transforming N5 equipment to supplement production capacity.
The climb of N2 forms another catalyst, and its adoption is expanding from smartphones to high-efficiency computing. Demand drivers include Apple (AAPL.US)'s newly launched A20 Pro and M6 chips (equipped with the latest iPhone 18 Pro and Mac mini, respectively), and the climb of AMD's next-generation EPYC Venice server CPUs. The Taiwanese corporation estimates that if 10 million AI GPUs are equipped with HBM4e customized base crystals, they will consume 45,000 to 50,000 3-nanometer wafers in a single month, which is equivalent to eating 20% to 25% of the 3-nanometer production capacity; even if the 3-nanometer monthly production capacity expands from 160,000 tablets at the end of 2026 to 220,000 tablets by the end of 2028, the capacity utilization rate will still maintain an overload of over 120% in 2027. Because of this, TSMC broke new ground this year by building three new 3-nanometer plants.
As a result, the speed of implementation of production capacity expansion became a key variable in this law conference — it determines whether TSMC can turn strong terminal demand into incremental revenue that exceeds expectations reflected in current stock prices.
Main line 2: Dilution of gross profit from overseas production expansion, and planned price increases
Another problem hanging over the head is the dilution of profit margins due to overseas production expansion. TSMC's gross margin guidance for the third quarter was 65% to 67%. Even if taken as a margin, it implied a reduction of at least 70 basis points compared to 67.7% in the second quarter. Management continued to warn: In the early stages of the decline in overseas production capacity, the company's overall gross margin was diluted by 2 to 3 percentage points each year, and later expanded to 3 to 4 percentage points. The company stated that the main reason for the decline was the high initial cost of mass production of 2 nanometers and rapid slope dilution by about 3 to 4 percentage points. Expanding capital expenditure has further exacerbated this concern. TSMC invested an additional $100 billion in Arizona in July, with a cumulative commitment scale of US$265 billion; the company's capital expenditure forecast for the full year 2026 was also raised from US$52 billion to US$56 billion to US$60 billion to US$64 billion. There are also rumors in the market that they may go to Texas to establish a manufacturing base and sub-lease production capacity at Tesla (TSLA.US) and SpaceXAI's future Terafab facilities. In an environment where capital costs are rising, this more capital-intensive expansion cycle will introduce the risk of incremental return compression.
However, revenue continued to exceed expectations, which only strengthened confidence in the large-scale expansion of advanced production capacity to improve operating leverage — this may outperform overseas dilution, thus forming a profit inflection point that exceeds market expectations. The supporting evidence comes from the second quarter: revenue of $40.2 billion reached the upper financial margin, with a gross profit margin of 67.7% and a profit margin of 60.3% above the financial ceiling. Net profit after tax was NT$706.6 billion, an annual increase of 77%, a surplus of NT$27.25 per share (equivalent to an ADR of approximately US$4.31), and an annualized shareholders' equity return of 45.9%.
More importantly, the monthly revenue growth rate for each month of the third quarter accelerated month by month — reaching 54.6% in September. As a result, the Taiwanese corporation estimates that with 2 nm initial dilution and product portfolio optimization offsetting each other, TSMC's gross margin for the third quarter is expected to rise 68%, which is better than the financial target of 65% to 67%. Goldman Sachs and Morgan Stanley are relatively conservative, with gross margins estimated to reach 67.5% for the third quarter, respectively.
The variable that can really change the profit trajectory is price increases. According to reports, TSMC expects to launch a comprehensive price adjustment in 2027: basic price increases of 5% to 10% for advanced manufacturing processes of 7 nm and below, 10% to 15% for advanced packaging, and 3% to 5% for mature processes; if customers add high-efficiency computing capacity in addition to existing orders, they must pay an additional 10% to 15% premium on the standard increase, and the total increase for some additional orders after adding up is about 25%. According to another review, 2nm wafers may increase by an additional 6% to 8%, and some mature nodes may increase by up to 10%. According to reports, the price adjustment is intended to “offset rising costs of materials, manufacturing equipment, and overseas factory construction”; TSMC has yet to confirm it, but it is consistent with its pricing philosophy of “the higher the better,” the premise is that it does not sacrifice customer success or “push customers out of the market.”
According to Morgan Stanley's judgment, the price increase for advanced manufacturing processes of 7 nm and below by 5% to 10% in 2027 and no more than 3% for mature processes will drive TSMC's gross margin close to 70% in 2027. The industry-level pricing environment also confirms this: According to previous reports, Samsung Electronics has raised the price of new orders for some advanced processes by as much as 15% (4-nm manufacturing increased 10% to 15% for customers in the US and mainland China, the same increase of 5 nm, and nearly 10% increase for 8 nm); according to Counterpoint Research data, the global pure wafer foundry market grew 29% year-on-year in the second quarter of 2026, and TSMC was at the top for two consecutive quarters with a 73% share, equivalent to Samsung (7%) in second place ten times.
As a result, further explanation of pricing strategies at this conference will become one of the core highlights: the supply-driven rise in average sales prices can provide a key safety cushion for TSMC's profit trajectory to hedge against overseas dilution and the cost of 2nm climbing and A14 research and development.
Valuation: Discrepancies in institutional target prices
According to Summit Research's valuation framework, TSMC will achieve a five-year compound annual revenue growth rate of 28% by 2030, driven by an increase in the share of advanced manufacturing processes to support AI high-efficiency computing opportunities with higher unit value, and the compound revenue growth rate of related platforms is expected to be 35%. In the discounted cash flow model, a 3.5% sustainable growth rate was used to estimate the final value of 2030 EBITDA and discounted at a 10.3% weighted average cost of capital, and the benchmark target price was 498 US dollars per share; if production capacity is more profitable, the optimistic scenario is 535 US dollars, thus making the $3 trillion market value threshold “within reach”.
The seller had already set off a wave of target price increases before the French conference. Gokul Hariharan (Gokul Hariharan), head of Taiwan's research department at J.P. Morgan Chase (Xiaomo), raised the target price from NT$3,200 to 3,300 yuan, rated “superior to the market”; Zhan Jiahong, head of Taiwan research at Morgan Stanley (Damo), raised the rating from $2,988 to $3,088; Goldman Sachs maintained a “buy” target price of NT$3,300; UBS maintained its target price of NT$3,300; and indicated that the stock price had a significant “re-rating”” Potential.
Among them, Goldman Sachs semiconductor analyst Yu Chengjing predicts that TSMC's 2026 and 2027 US dollar revenue will increase by 42% and 36.9%, respectively, gross margins will be 67.2% and 67.5%, respectively, and the capital expenditure forecast for 2027 and 2028 will be raised to 85 billion and 98 billion US dollars. Damo Zhan Jiahong, on the other hand, expects that the company will raise the annual revenue growth rate guide to a low position in the 40% to 45% range, increase 10% to 15% in the fourth quarter, maintain capital expenditure of 54 billion US dollars in 2026, and expand to 75 billion to 80 billion US dollars in 2027; he also listed three key topics: US production expansion (it may be necessary to build a factory in Texas within the next five years), Intel outsourcing trends, and large chip packaging support (CoPOS or CoPOS) for MediaTek's large TPU chips .
What investors need to hear at this law conference is essentially two things: one is the execution visibility of production capacity expansion, that is, whether terminal demand can be monetized; the other is the resilience of gross margin under overseas dilution pressure, which determines whether free cash flow can continue to increase.