FY26 Goes to 44% and Q4 Lands 6% Above the Street; the Stock Fell With the AI Tape, Not the Quarter — Maintain Outperform

Results vs. Consensus

Q3 2026 Scorecard

Year-Over-Year Comparisons

Quarter-Over-Quarter Comparisons

Quality of Beat

Revenue: The headline beat against consensus is small by construction, because TSMC publishes revenue monthly and the September report on October 8 had already put the quarter's NT$ revenue on the table (January–September NT$3,898.73B). The information is in the composition. USD revenue of $46.72B finished $0.92B above the guide ceiling, and it got there on volume growth of only 3.2%: 4,476K 12-inch-equivalent wafers versus 4,336K in Q2. Revenue per shipped wafer rose to roughly NT$334K from NT$293K, up about 14% sequentially and about 38% year-over-year. That is N2 entering the mix at a premium, N3 taking share from N5, and firm pricing. A foundry that grows revenue 17.6% on 3.2% more wafers is selling a richer product, not just more product.

Margins: The two margin lines tell different stories. Gross margin of 66.4% fell 130bp sequentially and, for the first time since Q2 2025, landed inside the guide range instead of above it. The cause is the one management flagged in July: the steep N2 ramp dilutes gross margin before yields mature. The exchange rate was not the cause: the average USD/NT$ rate rose about 1.2% sequentially, a small tailwind. Operating margin, by contrast, beat the ceiling by 70bp, because operating expenses held at 7.8% of revenue while the guide had implicitly allowed more. The gross-margin step-down is the cost of the node transition, and the operating line shows the cost discipline around it is intact.

EPS: The ADR EPS beat of about 3% is a fraction of Q2's 12.5%, and that is the better news. Q2's surprise leaned on a non-operating line running at 3.3 times its normal rate. Q3's surprise rests on operations: operating income grew 75.0% year-over-year and 14.3% sequentially. Net income grew only 8.0% sequentially, the arithmetic cost of lapping Q2's non-operating windfall, which confirms that our decision to anchor estimates to the operating result in July was correct. Year-over-year, ADR EPS grew 57.5% against NT$ EPS growth of 68.7%; the gap is the 6.9% rise in the average USD/NT$ exchange rate, which flatters NT$-reported growth. USD revenue growth of 41.2% is the cleaner read on the business.

Segment Performance

Platform Revenue Mix: Q3 2026

USD values are approximations: reported share of revenue multiplied by total USD revenue. Shares are rounded, so the approximations do not add exactly to $46.72B.

HPC / AI: 68% of Revenue and Most of the Growth

HPC grew 22% sequentially to 68% of revenue, its third consecutive record share (57% a year ago, 61% in Q1, 66% in Q2). On the approximations above, HPC contributed about $5.2B of the quarter's $6.5B USD revenue increase, or roughly 80%, and grew about 68% year-over-year while the rest of the company grew modestly. Management described the drivers as accelerators plus two newer vectors: CPUs for agentic AI workloads and "the continued demand for networking chips." Wei also cited a new factor, the broad adoption of open-source AI models, which he argued lowers the barrier to AI usage and therefore adds to leading-edge silicon demand rather than reducing it.

"Thus, AI-related demand is being supported by strong underlying demand for AI accelerators. In addition, the emergence of agentic AI CPUs and the continued demand for networking chips are driving further silicon demand." — C.C. Wei, Chairman & CEO

Assessment: HPC is now two-thirds of the company and still the fastest-growing part, which makes it both the thesis and the concentration risk. The incremental datapoint is breadth: networking and agentic CPUs now sit beside accelerators as named demand drivers, so the AI wafer pool no longer depends on one chip category. We expect HPC to hold 67–70% of revenue in Q4 as N2 ramps for both smartphone and HPC customers.

Smartphone: A Normal Reload, Flat on the Year

Smartphone revenue grew 10% sequentially, the normal September-quarter reload ahead of flagship launches. The year-over-year picture is weaker: on the share approximations, smartphone revenue was roughly flat in USD versus Q3 2025, when it was 30% of revenue. That is the segment management had in mind when it said consumer and price-sensitive end markets are "being challenged" by rising component prices. N2 demand from smartphone customers is real, and management named it alongside HPC as a driver of the 2027 N2 capacity doubling, but the unit backdrop for handsets is soft.

Assessment: Smartphone is a stable, high-value base, not a growth engine, and the 2027 caution is aimed squarely at it. We model a flat-to-low-single-digit 2027 for the segment, with the N2 transition (a richer wafer per phone) offsetting soft units.

Automotive: The Recovery Holds

Automotive grew 23% sequentially after +15% in Q2, two consecutive double-digit quarters after roughly two years of inventory digestion. At 4% of revenue it is not a model driver, but the direction confirms that the specialty-node correction that weighed on TSMC's automotive customers is over. IoT grew 5%, and DCE declined 12% on a 1% base.

Assessment: A small positive. The non-HPC, non-smartphone segments are no longer a drag, which lets the HPC growth reach the consolidated line almost unhindered.

Technology Node Revenue Mix: Q3 2026

The node table shows the transition in one quarter. N2 rose from 3% to 8% of wafer revenue, N3 from 30% to 33%, and the two leading nodes together reached 41% of wafer revenue, up from 33% in Q2 and 23% a year ago. N5 lost five points of share, partly because TSMC is converting 5nm tools to support 3nm capacity in Taiwan. Advanced technologies reached a record 79%. The gross-margin step-down follows directly: N2 carries a premium price but immature yields, so every point of N2 share costs gross margin today and adds to it once the node matures, which is the pattern N3 completed over the past two years.

Assessment: An 8% share in N2's second quarter as a disclosed revenue line makes this the fastest N2 ramp TSMC could plausibly deliver, and the 2027 capacity plan (more than double) says the ramp is only at its start. The dilution is the price of the node, and it is running as guided.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: Management was more categorical on AI demand than in July and more specific on execution. Where Q2's defining act was a capital commitment, Q3's was evidence: fab-by-fab timelines, a capacity number for N2, and a new practice of sending its own staff to inspect US data-center construction and power supply, which reads as a direct answer to the AI-financing worry the market was trading that week. The one hedge was a 2027 caveat for consumer and price-sensitive segments, delivered as planning discipline rather than as a warning. On pricing, the 2027 capex number and the AI growth rate, management again declined to put figures on the table.

1. FY26 Raised Again, From "Slightly Above 40%" to "Around 44%"

"Having said that, demand for our leading-edge technologies continue to be extremely robust, supported by our strong technology differentiation and broader customer base. Therefore, we now expect our full year 2026 revenue growth to be around 44% year-over-year in U.S. dollar term." — C.C. Wei, Chairman & CEO

This is the third full-year raise of 2026: "close to 30%" in January became ">30%" in April, "slightly above 40%" in July, and "around 44%" now. On FY25 USD revenue of $122.42B, 44% implies about $176B, which reconciles closely with the year to date ($122.82B across Q1–Q3) plus the $53.3B Q4 midpoint. Our July estimate of $172–176B now sits at its top end. The raise is mostly a function of Q3 overshooting by about $1.5B against its midpoint and Q4 being guided well above our ~$50B expectation.

Assessment: A four-point raise at the end of the third quarter is smaller than July's ten-point jump, but it is better information. With nine months reported, there is little room left for the full-year number to be wrong, so "around 44%" is close to a statement of fact. The more important number is the Q4 exit rate, discussed below.

2. A Q4 Guide That Is a Statement About 2027

The Q4 guide of $52.3–54.3B implies 14% sequential growth at the midpoint, 58% year-over-year growth, and a quarterly run rate above $210B annualized. The Street was at roughly $50.2–50.5B, so even the low end of the range beats it by about 3.6% to 4.1%. Management attributed the step-up to "the steep ramp-up of our 2 nm technology." A Q4 of this size also sets a high base for the 2027 growth debate: if TSMC merely holds a $53B quarterly rate through 2027, FY27 revenue grows about 21%. Every point of 2027 growth above that has to come from sequential growth on top of the record.

Assessment: The Q4 guide is the most bullish number in the release because it leaves a high run rate going into a year management has just described as one for prudent planning. Read together, the message is that the prudence applies to consumer segments, while leading-edge AI capacity remains sold out.

3. Gross Margin: N2 Dilution Arrives on Schedule

"We have just guided our fourth quarter gross margin to decrease by 40 basis points to 66% at the midpoint, primarily due to the continued margin dilution from the steep ramp-up of our 2 nm technology, as well as dilution from overseas fabs." — Wendell Huang, Senior VP & CFO

Q3 gross margin fell 130bp to 66.4%, "primarily due to margin dilution from 2 nm ramp," and Q4 is guided down another 40bp to 66%. Two offsets limit the decline: cost improvement, including productivity gains and cross-node capacity optimization, and overseas-fab dilution running better than planned. Management now expects overseas dilution "to be closer to 2%," versus the 2–3% framework for the early years, but said that dilution will widen to 3–4% in the next few years as more US fabs ramp. Asked whether margins can expand further, Wei offered a framing worth remembering:

"Let me say that, there is no ceiling for TSMC's profitability, but a dramatic increase will not happen overnight either." — C.C. Wei, Chairman & CEO

Assessment: The margin story has moved from "beating the ceiling" to "delivering the guide," and investors should not confuse the two. A 66% gross margin with 8% of wafers on a node in early yield, and with overseas fabs ramping, is a very strong result. It is still well above the 56% through-cycle target, which management has not revised. For 2027, the overlapping headwinds (more N2, wider overseas dilution) argue for a gross margin in the low-to-mid 60s, not a return to 67–68%.

4. N2: 8% of Wafer Revenue, With Capacity to "More Than Double" in 2027

"We expect our N2 family technologies annual capacity to more than double in 2027, supported by strong demand from both smartphone and HPC AI applications." — C.C. Wei, Chairman & CEO

In July, Wei would say only that the 2nm capacity plan was "bigger" than the Technology Symposium roadmap. This quarter he put a number on it: N2-family annual capacity more than doubles in 2027, ramping in multiple phases at Hsinchu and Kaohsiung. N2 reached 8% of wafer revenue in its fourth quarter of volume production and its second quarter as a disclosed revenue line. Management also said it is extracting more N2 output from existing tools through manufacturing efficiency, and that N2 serves smartphone and HPC customers at the same time, a dual demand base that N3 did not have at the same stage.

Assessment: This is the most important capacity disclosure of the quarter for the 2027 model. Doubling N2 capacity against two demand bases (smartphone and HPC) supports both revenue growth and a continued ASP mix tailwind, at the cost of another year of node dilution. It also widens the gap with rivals that have not yet shown a commercially proven 2nm alternative at scale.

5. "Prudent" Into 2027: The Consumer Caveat

"Looking ahead, we continue to observe consumer and price-sensitive end market segment are being challenged due to the impact of rising component prices and macroeconomic uncertainties. As such, we are being prudent in our business planning going into 2027 while focusing on the fundamentals of our business to further strengthen our competitive position." — C.C. Wei, Chairman & CEO

The consumer caution is not new. Management flagged rising component costs and soft price-sensitive consumer segments in January, and in April Wei said TSMC was "very mindful" of rising component prices. What is new is the horizon: the caveat is now attached to 2027 planning, and it arrived on the same call as a record Q4 guide. The cause is the cost squeeze in handsets and PCs from rising component prices (memory most prominently), which pressures unit volumes in exactly the segments where TSMC has the least pricing leverage. The same paragraph then pivots: "Having said that, demand for our leading-edge technologies continue to be extremely robust."

Assessment: We read this as a two-speed 2027: AI-driven leading-edge demand above supply, consumer-driven demand flat to soft. It does not threaten the thesis, but it caps the upside to 2027 smartphone and mature-node revenue, and it is why we model FY27 growth at 25–30% rather than extrapolating 2026's 44%. We add it to the standing thesis as an explicit watch item.

6. CapEx Reiterated at $60–64B, Not Raised

"Given the continued robust multi-year demand from our customers, including the emerging agentic AI market, we reiterate our full year 2026 capital budget to be between $60 billion and $64 billion." — Wendell Huang, Senior VP & CFO

After moving up at each of the previous two calls, the 2026 capital budget was held at $60–64B, and management repeated that 2026–2028 capex will be "even more significantly higher than the past three years." Q3 capex was NT$443.21B ($13.8B), down from NT$496.00B in Q2. By our arithmetic, year-to-date capex is about $40.6B (Q1 $11.1B, Q2 about $15.7B at the quarter's average exchange rate, Q3 $13.8B), so Q4 must reach roughly $19–23B to land in the range. Wendell Huang acknowledged that "the tool supply is tight," and on that evidence the low half of the range looks more likely than the high half. On capital intensity, he restated the rule that matters for free cash flow: "if do our job right, the increase in revenue will outpace the increase in CapEx."

Assessment: Neutral for the thesis today, and the key test for 2027. Q3 capex consumed only about 30% of revenue, and free cash flow rose 66% sequentially to NT$476.56B, so funding is not the issue. The issue is the growth-rate rule. A $90B-plus 2027 budget, the level some sell-side estimates now assume, would be 40–50% capex growth on the 2026 range, well above the 25–30% revenue growth we model. If the January guide looks like that, management's rule breaks on a growth-rate reading, and that is one of our Hold triggers below.

7. AI Demand: "No Sign of Slowdown," Now Backed by Site Checks

"In addition, we are now going one step further and sending our own teams of people to check on the data center build-out progress and power supply situation in the U.S. The feedback is so far so good, and the CSPs has been well-prepared." — C.C. Wei, Chairman & CEO

The call came hours after an AI-sector selloff driven by credit concerns at a GPU-cloud provider, and management's demand language was the strongest of the year. Wei said the AI market "continues to be very strong," that "recent developments are very positive" for long-term AI demand, and in Q&A that TSMC did "not see any sign of slowdown." The new element is process. Capacity planning has always combined customer forecasts with the forecasts of customers' customers. TSMC now adds its own physical checks on US data-center construction and power availability, and it lists financing, power and component supply among the planning inputs. Management also repeated that its narrow AI-accelerator definition excludes CPUs, so the agentic-CPU growth it describes is additional to the accelerator figures.

Assessment: TSMC's demand evidence and the market's AI-financing fear point in opposite directions, and TSMC's evidence is closer to the order book. The site-check practice does not remove the risk that a financing shock at the customer level reaches wafer orders with a lag, but it shortens the lag in which TSMC would detect one. For now, the supply gap management describes ("the demand actually is still far above our capability today to supply") is the binding fact.

8. An Unprecedented Build: Nearly 20 Fabs and a Fourth N3 Fab

"TSMC is expanding at unprecedented pace with a nearly 20 fabs worldwide in progress. We are building 13 leading-edge and advanced packaging fab in Taiwan over the next several years, and another five to six fabs overseas." — C.C. Wei, Chairman & CEO

Management updated the N3 expansion that broke its historical policy in April. The new Tainan N3 fab remains on track for volume production in the first half of 2027. TSMC has added one more N3 fab in Tainan, scheduled for the first half of 2028. Arizona's second fab, an N3 fab, is in tool move-in and installation, and volume production moves to the third quarter of 2027, "slightly ahead of our previous schedule." Japan's second fab remains on plan for N3 in 2028. On top of the new fabs, TSMC continues to convert N5 tools to N3 and to share capacity flexibly across N7, N5 and N3. Wei summarized the intent in a line that will be quoted back to him:

"We have no intention to leave food on the table for others, and we want to capture every piece of business possible." — C.C. Wei, Chairman & CEO

Assessment: A fourth new N3 fab, nearly four years after the node entered volume production, is the clearest evidence in this release that leading-edge demand is still running ahead of the plan. The "food on the table" language is also a competitive statement: TSMC intends to leave no overflow demand for rival foundries, and it is willing to accept capital intensity and overseas dilution to do so.

9. Beyond Arizona: Texas and a Customer's Fab

"Recently, we did expand our capacities consideration to see outside Arizona whether we can get another chance to expand our capacity. Texas is one." — C.C. Wei, Chairman & CEO

Management confirmed that it is evaluating US sites outside Arizona, with Texas named, and that it is in discussion with a customer building its own fab (Wei named it as Terafab) about using that capacity. The terms are open, but Wei set one condition: "the fab has to be run by TSMC." That rules out a customer-operated model and keeps TSMC's yield, process and customer-trust control intact.

Assessment: Positive optionality with a sensible guardrail. Running a fab that a customer finances would let TSMC add US capacity with less of its own capital, though it would also add to the overseas-fab dilution. Nothing is signed, so we treat it as a 2027 watch item and do not model it.

10. A14, and Why Adopting High NA in 2030 Is Not a Disadvantage

"In addition, we have already begun preparing the capacity for our A14 technology. Risk production is well on track for 2027, and volume production will start in 2028." — C.C. Wei, Chairman & CEO

A14 risk production is on track for 2027 and volume production for 2028. In Q&A, Wei said A14 has drawn more customer engagement than N2 had at the same stage, and he expects it to be a bigger node than N2. On High NA EUV, which TSMC plans to put into production in 2030 (later than some competitors), Wei's answer was about maturity and cost: TSMC adopts a tool when it is "the best technology with a very reasonable cost," and it is working with ASML on a 12-inch photomask by 2031 to make that cost reasonable.

"Whether that our technology, because we are using it later than our competitor, so disadvantage to TSMC? No, because we have confidence that we are still the best company to offer the most advanced technology to our customers." — C.C. Wei, Chairman & CEO

Assessment: The roadmap extends the monopoly argument to 2028 and beyond. TSMC was early with the original EUV; with High NA it is choosing to wait for cost and maturity. A competitor that adopts High NA first gains a tool, not a yield-proven process with a customer base.

11. Advanced Packaging: 5x Reticle in Production, and a Backup Offer

"We are in the real-world production of 5x reticle size. I don't believe that anybody can do that today except at TSMC, and the yield is excellent." — C.C. Wei, Chairman & CEO

TSMC is in production on 5x-reticle CoWoS packages, working on 9x, and moving toward 14x. The back end is still short of capacity, and two disclosures address that. First, TSMC struck an interposer supply arrangement with GlobalFoundries, which Wei described as part of its back-end strategy because "we have a little bit shortage on the back end." Second, asked whether TSMC would provide CoWoS capacity to a customer whose first-choice competitor packaging fails to deliver, Wei said he would avoid a direct answer and then gave one: "yes, we will be."

Assessment: Packaging remains the most constrained part of the AI supply chain, and TSMC is now using outside partners to relieve it rather than letting the shortage cap front-end wafer shipments. The backup offer is a competitive signal: customers who try a rival's packaging still have a route back to CoWoS, which reduces the risk of trying an alternative and keeps TSMC as the default.

12. Cash Generation and Shareholder Returns

"Even as we invest for the future growth with this level of CapEx spending in the next few years, we remain committed to delivering profitable growth to our shareholders. We also remain committed to a sustainable and steadily increasing cash dividend per share on both an annual and quarterly basis." — Wendell Huang, Senior VP & CFO

Cash from operations reached NT$919.77B, up 17% sequentially, and free cash flow reached NT$476.56B (about $14.9B). Cash and marketable securities ended at NT$4.02 trillion (about $126B), and long-term interest-bearing debt declined to NT$818.38B. The board approved a NT$7.00 quarterly dividend for Q2 2026 in August, payable in January 2027.

Assessment: The balance sheet absorbs the capex cycle easily. With cash and securities exceeding long-term interest-bearing debt by more than NT$3 trillion, and quarterly free cash flow of nearly $15B even at a 30% capex-to-revenue ratio, the capital-intensity bear is about returns on capital, not funding. A dividend increase in 2027 is likely.

Guidance & Outlook

The Q4 guide carries the quarter. A $53.3B midpoint is a $6.6B sequential step, slightly larger than Q3's record $6.5B, and puts the exit run rate above $210B annualized. The margin guide steps down modestly: gross margin to 66% at the midpoint and operating margin to 57.5%, as N2 dilution deepens and overseas fabs add a smaller drag. The exchange-rate assumption of 31.8 implies a slightly stronger NT$ than Q3's 31.98, a minor headwind already inside the range.

Implied FY26: Q1–Q3 revenue was $122.82B. The Q4 range takes FY26 to $175.1–177.1B, or +43.0% to +44.7% on FY25's $122.42B, which brackets "around 44%." In NT$, FY26 revenue lands near NT$5.59 trillion, above the NT$5.42 trillion our model's consensus band carried in early October. We model FY26 gross margin at about 66.5%, operating margin at about 58.5%, and net margin at about 51.5%, for ADR EPS of $17.30–17.70 (Q1–Q3 actual $12.40 plus a Q4 estimate near $5.05).

Street at: Into the print, Q4 revenue consensus sat at roughly $50.2–50.5B. The guide midpoint is about 6% higher, so Q4 and FY26 estimates move up across the Street, and FY27 revenue estimates, which compound off the Q4 exit rate, should follow.

Guidance style: Conservative, as usual. Q3 revenue landed $0.92B above the ceiling of its own range, at least the sixth consecutive quarter at or above the top of the revenue guide. Operating margin beat the ceiling again. The exception is gross margin, which landed inside the range for the first time in five quarters, an indication that management is now guiding gross margin with less cushion while N2 dilution is in progress.

Analyst Q&A Highlights

Is the AI-Accelerator Growth Rate Being Raised Again?

For the second consecutive quarter, analysts pressed for an updated AI-accelerator CAGR beyond the mid-to-high-50s rate for 2024–2029. Management again declined a number but said the rate "probably will be higher." Later in the call, management confirmed that data-center CPUs remain outside that narrow definition, because TSMC cannot tell an AI-server CPU from a conventional one.

Q: "So far from our comprehensive study of the supply chains, we are still quite positive on the demand in the longer term run. In that case, can you also give us more color about that? Any update on your AI CAGR growth? I recall previously we talking about like a high 50%."
— Laura Chen, Citi

A: "Laura, all I can say right now is the demand is extremely robust, okay? We did not see any sign of slowdown. Whether it is a high 50 CAGR, it does not matter. It probably will be higher."
— C.C. Wei, CEO

Assessment: Two quarters of "higher, but no number" now amount to a de facto upgrade. The CPU exclusion matters as much as the CAGR: the agentic-CPU growth management keeps describing sits outside the published AI figure, so the reported AI share of revenue understates the AI-driven share of TSMC's business.

How TSMC Guards Against an AI Overbuild

The question the market was asking that week came up first on the call: with power availability, financing, high-bandwidth memory, packaging and optical components all constrained, could chip capacity outrun the AI build-out it serves? Management described its planning inputs and concluded that demand still exceeds supply.

Q: "If these kind of trends create a mismatch between semiconductor supply and also actual AI development, how would TSMC adjust your capa-minimize the potential overbuilding in the supply chain?"
— Laura Chen, Citi

A: "We have a very sophisticated capacity planning process, and we also have a very disciplined on our consideration. Let me share with you, we put every possibility we can, talking about the CSPs, CapEx, data center build-out, financial, power supply, and also other component supply, et cetera, into our consideration. We make our own judgment, and then based on our assessment the demand, we still think the demand is higher than the supply, and we are working very hard to narrow the gap."
— C.C. Wei, CEO

Assessment: The answer names financing and power as inputs, which is the right list, but the conclusion is an assertion rather than a figure. Combined with the new physical site checks, it is the best available rebuttal to the overbuild thesis. It is not proof, and an AI-financing shock would reach TSMC's order book after its customers' budgets, not before.

Capital Intensity Beyond 2026 and Tool-Price Inflation

A questioner asked whether 2027 capex could grow in line with or faster than revenue, as it did in 2026, and whether tool-price inflation would add to it. Management declined a 2027 figure, acknowledged the inflation, and restated its capital-intensity rule.

Q: "Recall a few quarters ago, you mentioned, if you are doing your job well, then you may be able to drive CapEx to grow a bit less than top line. ... Meanwhile, obviously the tool price is also increasing, given a very tight supply, and therefore, going to 2027, will the higher tool price also be a driver for higher CapEx as well?"
— Sunny Lin, UBS

A: "The tool price inflation is real. The tool price, material, labor cost, et cetera. However, the majority of the CapEx is always related to the long-term demand. That will be the composition going forward. Capital intensity-wise, as we always said, if do our job right, the increase in revenue will outpace the increase in CapEx, and that is still the case from what we are looking at."
— Wendell Huang, CFO

Assessment: "Revenue will outpace CapEx" is the commitment that matters for free cash flow, and it is a testable one. If 2027 capex rises toward the $90–105B range some estimates now assume, revenue needs to grow at a similar or faster pace to honor it. We will grade the January guide against this sentence.

A US Fab Run by TSMC, Financed by Someone Else?

A questioner followed reports of a possible Texas site and of talks with a customer building its own fab, and asked whether TSMC would join that project or run the capacity itself. Management confirmed the discussions and set one condition.

Q: "But all these are going to be TSMC-run fabs, right? I think you are not becoming part of the Terafab project or anything like that."
— Gokul Hariharan, JPMorgan

A: "We are still in discussion, let me say that. It's not either way, it would be possible, but the fab has to be run by TSMC."
— C.C. Wei, CEO

Assessment: The operator condition is the important part. TSMC will consider capacity it does not wholly finance, but it will not license its process to a customer-run fab. That protects yields and customer trust, the two assets the "five years, no shortcut" argument rests on.

An IDM Customer's Insourcing, and Why A14 Is Bigger Than N2

A questioner asked whether a large integrated device manufacturer that is expanding its own fabs would take work back in-house, and whether TSMC's expectation that A14 will be a bigger node than N2 depends on that customer outsourcing server CPUs. Management declined to discuss the customer and pointed to broader A14 engagement.

Q: "As far as I know, one of your IDM customer is expanding capacity aggressively. There are some concern about this customer may insource more than outsource to you."
— Charlie Chan, Morgan Stanley

A: "Well, Charlie, we treat every customer fairly. Whether it is IDM or it is purely fabless, we treat them the same. But of course, we take into consideration of their internal capacity. We calculate our support, we calculate our capacity build-up according to our own judgment. Actually, in our A14, we have seen more and more customer actually higher than the 2 nm activities so far. We expect that the A14 will be a bigger node than our current N2 technologies."
— C.C. Wei, CEO

Assessment: The A14 claim stands without the IDM customer: TSMC already plans that customer's internal capacity into its own build, and A14 engagement is broader than N2's at the same stage. Insourcing by one large customer is therefore a known input, not an unpriced risk.

Can Customers Trade Wafer Allocations?

A questioner asked about reports that customers pass unused wafer allocation to other customers, in effect a secondary market for TSMC capacity. Management said the practice is not permitted, with a narrow exception.

Q: "First one, C.C., some customers talk about a secondary wafer market where one customer passes their wafer allocation to another. Is this a very small portion of the business?"
— Krish Sankar, TD Cowen

A: "The principle of TSMC is we do not allow our customer to sublet or auction off the capacity. But in certain cases, customer can collaborate through a mask licensing agreement, but this is not our common practice. The amount is very, very small."
— C.C. Wei, CEO

Assessment: A secondary market would let customers capture the scarcity premium on TSMC's capacity, so the prohibition protects TSMC's pricing power. The question itself is a sign of how tight leading-edge supply is: allocation is now valuable enough that customers want to trade it.

Do Price Increases Reach the Contracted Book?

With reports of significant price increases across mature and advanced nodes, a questioner asked whether new pricing applies to existing commitments or only to incremental capacity. Management did not confirm any increase and described pricing as strategic and location-dependent.

Q: "On the offset side, there have been reports of significant price increases across your mature and advanced nodes. Could you provide some color into perhaps some of that conversation? Where pricing does move, does it typically apply across the existing contracted book, or is it mainly being applied to incremental capacity above what customers have already committed to?"
— Sahej Singh, Stifel

A: "Well, we have discussed this kind of thing before, that our pricing strategy is very strategic. It also depends on the market segment, depends on the location, and depends on the technology and something. Once we offer the flexibility of the physical location, like in the U.S., it will be a little bit different from the one we produce in Taiwan. No matter what, we try to get our value to be recognized, and we will give our shareholders the maximum value of our technologies."
— C.C. Wei, CEO

Assessment: A non-answer, as expected, but with one useful detail: US-made wafers are priced differently from Taiwan wafers. That is the mechanism by which TSMC intends to pass overseas-fab costs to customers, and it is the main offset to the 3–4% overseas dilution management expects in later years.

Ramping Smartphone and AI Chips on the Same New Node

A questioner noted that new nodes traditionally ramp on small mobile dies first, and asked whether running smartphone and large AI dies at the same time (on different power-delivery architectures) makes the yield-learning curve harder and the margin dilution deeper. Management argued the opposite.

Q: "But for the future node families supporting both a conventional power delivery for the applications like mobile and consumer applications and Super Power Rail technology for HPC and AI chipsets, would operating two process architectures at the same time make the learning curve more challenging or increase the manufacturing complexity?"
— Haas Liu, Bank of America

A: "Well, today's situation actually is a little bit different from previous years. Yes, we now have a very strong demand from smartphone and also AI HPC applications. So the smaller die size per se or the big die size in the accelerator, they happen at the same time. This is to TSMC's advantage, let me say that."
— C.C. Wei, CEO

Assessment: This exchange bears on the 2027 margin model. A simultaneous ramp means N2 dilution is concentrated in a shorter window instead of spread over a mobile-first then HPC-later sequence. Management says yields are progressing well ("we are doing a reasonable job"), but it gave no yield figures, so the claim rests on the Q4 gross margin delivering the guided 66%.

What They're NOT Saying

1. A 2027 capex figure: "It's a bit too early to share the CapEx number beyond this year's." With sell-side estimates for 2027 now ranging up to about $105B, the outer-year capital bill is the largest open input for free cash flow and depreciation models, and it will not be addressed until January.
2. A 2027 growth framework: The call offered "prudent" planning for consumer segments and "extremely robust" AI demand, but no quantification of how the two net out. Investors have a record Q4 run rate and an explicit caution, with nothing in between.
3. An updated AI-accelerator CAGR: For a second consecutive quarter, management said "higher" without a number. The mid-to-high-50s framework is now stale, and the excluded CPU category is growing.
4. Confirmation or denial of price increases: Asked directly about reported price increases across nodes, management described a "very strategic" pricing approach. The revenue-per-wafer arithmetic (about +14% sequentially) shows that price and mix are working, but the split between them was not given.
5. The N3 margin crossover: In April, management said N3 would cross the corporate-average gross margin in the second half of 2026. A questioner described N3 as already accretive, but management's prepared remarks did not confirm the crossover or say whether N3 is now offsetting part of the N2 dilution.
6. The terms of the Texas and Terafab discussions: Management confirmed the talks and the operator condition, but said nothing about who would fund the capacity, which nodes it would run, or the timing.
7. Customer-level AI financing: Financing appeared once, as one item in a list of planning inputs. Management did not address whether any customer's funding conditions have changed, which is the specific risk the market was trading on the day of the call.

Market Reaction

• Pre-print setup: The TSM ADR closed at $430.00 on October 14, down 3.4% that day as an AI-sector selloff began. It entered the print +41.5% year-to-date, +41.1% over the trailing twelve months and +3.9% over the trailing 30 days, about 11.5% below the top of its 52-week closing range of $275.06–$485.80. The S&P 500 was +12.8% year-to-date.
• Reaction session (October 15): TSMC reports before the US open. The ADR opened at $415.17 (−3.4%), traded as low as $407.13 (−5.3%), and closed at $422.25 (−1.8%) on 22.4M shares, 2.3 times its 30-day average volume. The S&P 500 rose 0.3%, the SMH semiconductor ETF fell 1.0%, and NVDA fell 1.9%.
• Next session (October 16): The ADR fell another 2.3% to $412.61 as semiconductors sold off again (the semiconductor ETF −3.4%, AVGO −4.5%, AMD −6.0%, S&P 500 −1.1%). Over the two sessions, TSM fell 4.0% against 4.4% for the semiconductor ETF.
• What moved it: Coverage attributed the move mainly to the sector: an AI selloff that began on October 14 with concerns about the debt load of a GPU-cloud provider, then spread to other AI-infrastructure names and their suppliers, compounded by hotter-than-expected US inflation data. Company-specific attention went to the sequential gross-margin decline and the 2027 consumer caution.

The stock traded with the sector, not against the print. The ADR gapped down 3.4% at the open and touched a 5.3% loss as the market priced the margin step-down and the 2027 caveat against a hostile tape. It then recovered 3.7% from that low to close at $422.25 as the guide was digested. On the reaction day itself, TSM lagged the SMH semiconductor ETF (−1.8% against −1.0%), so the margin step-down and the 2027 caveat drew some selling of their own. Over two sessions, TSM fell slightly less than the ETF (−4.0% against −4.4%). The print added a modest company-specific discount on the first day; the sector set the direction.

A familiar pattern, with a different cause. This is the third consecutive 2026 print to close down on the reaction day: −3.1% in April, −2.3% in July, and −1.8% now. In April the market sold the positioning, in July it sold the capex raise, and in October it sold the AI complex. Each time, the fundamentals moved up. From the July pre-print close to the October 16 close, the ADR is down 1.6%, while our FY26 EPS estimate rose about 3% and FY26 growth guidance rose four points. The multiple, not the business, has absorbed the AI-sentiment shift.

Why it matters for the rating. Valuation was our primary bear in July. A stock that is now about 15% below its 52-week closing high, with a higher earnings base, trades near 23.6x FY26 and about 20x our FY27 estimate. The AI-financing scare is a legitimate risk to monitor, and we treat it as the main bear below. But at this multiple, the stock no longer needs the AI narrative to improve to outperform; it needs TSMC to deliver the Q4 guide and a 2027 plan near our estimates.

Street Perspective

Debate: Is the AI-Financing Scare a Signal TSMC Cannot See Yet?

Bull view: TSMC sits closest to the order book of any company in the AI chain, and it just guided Q4 6% above the Street, added a fourth new N3 fab, and said demand is "far above" its ability to supply. Its own teams are checking US data-center construction and power, and the feedback is "so far so good." A credit wobble at one GPU-cloud provider is a funding-structure problem for that layer, not a demand problem for leading-edge wafers.

Bear view: Foundry orders are the last link in the chain to feel a funding shock. If the neocloud and AI-lab layer cannot refinance at reasonable rates, hyperscaler and GPU orders will be cut first, and TSMC will see it in its monthly revenue two to three quarters later. A record Q4 guide is exactly what a peak looks like.

Our take: Bull, with a defined tripwire. TSMC's monthly revenue disclosure is the earliest public evidence of any order change, and the November and December reports will test the Q4 guide. Until those show a break, a supply-constrained monopoly with a raised guide is the wrong place to express an AI-credit concern.

Debate: Have Margins Peaked?

Bull view: A 66.4% gross margin with N2 at 8% of wafers is a remarkable result, overseas dilution is running better than planned at closer to 2%, and US wafers carry different pricing. Revenue per shipped wafer rose about 14% sequentially, which says pricing and mix are strong enough to absorb N2's dilution. "There is no ceiling for TSMC's profitability."

Bear view: The four-quarter streak of above-ceiling gross margins has ended, Q4 is guided lower again, N2 capacity doubles in 2027 (more dilution), and overseas dilution widens to 3–4%. Add depreciation from $60B-plus capex, and 2027 gross margin heads toward the low 60s. Sell-side summaries after the print said everything looks strong except margins.

Our take: The bears are right on direction and wrong on importance. We model FY27 gross margin at 63–65%, down from about 66.5% in FY26, and still estimate earnings growth of about 20% because revenue growth dominates the margin give-back. Peak gross margin is not peak earnings.

Debate: What Does "Prudent" Mean for 2027?

Bull view: The caveat applies to consumer and price-sensitive segments, which are a minority of revenue and shrinking as a share. AI capacity is sold out, N2 capacity doubles, and the Q4 exit rate alone supports about 21% growth in 2027 if it simply holds. Consensus FY27 estimates are too low.

Bear view: Management chose to introduce 2027 caution on a record quarter, and companies rarely do that without reason. Component-cost inflation hurts handsets and PCs, mature nodes stay soft, and if AI spending pauses even briefly, 2027 growth could fall well below the 30%-plus the Street extrapolates from 2026.

Our take: Closer to the bull view, but we take the caution seriously. We model FY27 USD revenue growth of 25–30%, below a straight extrapolation of 2026 and roughly at or below the FY27 consensus. The caution lowers our confidence in the consumer half of the business, not in the AI half.

Model Update & Valuation Framework

Valuation framework: With nine months of 2026 reported, the 12-month price target now rests on FY27 earnings. At the October 16 close of $412.61, TSM trades at about 23.6x our FY26 ADR EPS midpoint of $17.50 and about 19.6x our FY27 midpoint of $21.00. Our FY27 estimate assumes revenue growth of 25–30%, a gross margin of 63–65% as N2 and overseas dilution overlap, and a net margin near 48.5%. That is a deliberately conservative build relative to a Q4 exit rate that supports about 21% growth on its own. We apply 22–23x to FY27 for the base case, below the 26x we applied to FY26 in July, to reflect slower growth and the AI-sentiment risk now in the multiple.

Revised risk-reward: From $412.61, the base case ($460–485) implies about +11% to +18%, the bull case ($540–565) about +31% to +37%, and the bear case ($295–320) about −22% to −29%. The base-to-bear ratio is similar to July's, but the inputs have improved: the earnings base is higher, the multiple is lower, and the bear case now requires an AI-digestion year that TSMC's own order book does not yet show. That is why we keep the Outperform rating and restore conviction.

Thesis Scorecard: Q2 2026 Signposts Revisited

The Q2 2026 recap set five commitments to watch and four Hold triggers. None of the Hold triggers fired, four of the five commitments were met or beaten, and the fifth (a formal AI-CAGR revision) did not happen.

Scorecard summary: Eight of twelve signposts were bullish or better and a ninth, valuation, de-escalated, including the four that drive near-term estimates: Q3 revenue, the Q4 guide, the FY26 raise and the N2 ramp. Three items need watching rather than action: the AI-CAGR number did not come, customer concentration rose with HPC's record share, and management attached a 2027 caveat to its consumer segments.

Standing Thesis: Pillar Status Through Q3 2026

Overall: The thesis strengthened on fundamentals and on valuation at the same time, which is unusual. Every bull pillar stays on track, and the process-technology and capacity pillars gained concrete 2027 milestones (N2 capacity doubling, a fourth new N3 fab, A14 risk production). On the bear side, valuation moves from emerging to contained as the multiple compressed, and the primary risk becomes AI-spending sentiment, which TSMC's own data do not yet confirm. The consumer and component-cost risk, tracked in our matrix since Q4 2025, moves into the standing thesis because management has now tied it to 2027 planning.

Bottom Line: The Seventh Consecutive Outperform

Rating decision: We maintain Outperform on TSM, the seventh consecutive Outperform in our coverage, and restore conviction to 8 from the 7 we set in July. The July reduction was explicitly about valuation dependence, not fundamentals. Since then, the FY26 earnings base has risen and the stock has fallen, so that reason has largely reversed.

• Revenue above the guide ceiling: $46.72B against $44.6–45.8B. ✓
• Operating margin above the guide ceiling: 58.7% against 56.0–58.0%. ✓
• Gross margin inside the guide, as N2 dilutes: 66.4% against 65.0–67.0%. ✓
• FY26 guide raised, not cut: "slightly above 40%" to "around 44%." ✓
• Q4 guide above the Street: $53.3B midpoint against about $50.2–50.5B. ✓
• N2 ramp with a 2027 capacity number: 8% of wafer revenue; capacity more than doubles in 2027. ✓

What is genuinely different this quarter. Two things. First, the gross-margin beat streak ended, and the next year carries overlapping dilution from N2 and overseas fabs, so earnings growth in 2027 will come from revenue, not margin expansion. Second, the market's AI concern has moved from capital intensity at TSMC to financing stress among TSMC's customers' customers. TSMC's order book shows no sign of it, but the transmission lag means the monthly revenue reports, not the earnings calls, are now the signposts to watch.

What could change our mind:

• Downgrade to Hold: Q4 revenue below the $52.3B low end of the guide, or two consecutive sequential declines in monthly revenue outside normal seasonality; a January FY27 growth guide below 20% in USD; Q4 gross margin below 64%; 2027 capex guided to grow faster than 2027 revenue, breaking management's capital-intensity rule; or a re-rating above ~26x FY27E without a matching estimate revision.
• Downgrade to Underperform: Evidence that a major AI customer has cut or pushed out leading-edge wafer orders; a Taiwan Strait escalation with operational impact on the fabs; gross margin sustained below 57% for two consecutive quarters; or an N2 yield problem that delays the 2027 capacity doubling.

For now, TSMC raised the year, guided Q4 6% above the Street, put a number on its 2027 N2 capacity, and saw its stock fall with the AI complex rather than on its own results. At about 20x next year's earnings, the market is pricing an AI slowdown that the company with the best view of the order book says it does not see. We stay Outperform.