TSMC (TSM) — Q3 FY2026 Earnings Briefing
Source: TSMC Q3 2026 earnings call transcript (transcripts/TSM/Q3_2026.txt), October 15, 2026 · Fiscal period: Q3 FY2026 · Report date: 2026-10-15
Executive Summary
- A beat-and-raise quarter powered by AI leading-edge demand: revenue of $46.7B cleared the high end of guidance, and full-year 2026 growth was lifted to around 44% YoY in USD terms.
- Profitability is the counterweight: the 2nm ramp began diluting margins this quarter, overseas fabs widen their drag in the years ahead, and Q4 gross margin is guided down again.
- Management’s Q&A posture was one repeated sentence — demand exceeds supply — while declining to re-quantify the AI accelerator growth outlook or name a capacity-growth percentage.
- The build-out is unprecedented: a first-ever break of the no-post-target-expansion rule on 3nm, 13 leading-edge and advanced-packaging fabs in Taiwan, five to six overseas, and 2026–2028 capex guided “even more significantly higher” than the prior three years.
- Prudence sits underneath: consumer and price-sensitive end markets are challenged, and planning into 2027 is deliberately conservative.
Highlights and Lowlights
Highlights
- Revenue rose 16.2% QoQ in USD (17.6% in NT) to clear the high end of the company’s own guidance, on strong leading-edge demand.
- HPC grew 22% QoQ to carry 68% of total revenue, with smartphone up 10% QoQ to a 21% share.
- Fourth-quarter guidance of $52.3–54.3 billion, up 58% YoY at the midpoint, implies a 14% sequential step-up led by the steep 2nm ramp.
- N2 reached 8% of wafer revenue in its high-volume-manufacturing ramp quarter.
- Management reiterated the 2026 capital budget at $60–64 billion, with 2026–2028 spending pledged “even more significantly higher” than the past three years.
- TSMC broke its own node rule and authorized three new 3nm fabs — two in Tainan, one in Arizona — plus ongoing 5nm-to-3nm tool conversion in Taiwan.
- Arizona’s second 3nm fab was pulled slightly ahead of schedule, with volume production now set to begin in Q3 2027.
Lowlights
- Gross margin fell to 66.4%, down 130 bps QoQ, on margin dilution from the initial 2nm ramp.
- Q4 gross margin is guided to 65–67%, a further 40 bps decline at the midpoint, as continued 2nm dilution is joined by overseas-fab dilution.
- Overseas-fab margin dilution widens to 3–4 percentage points in the latter part of the coming five-year period, per the CFO’s clarification when Sunny Lin (UBS) read the widening as arriving from next year.
- Consumer and price-sensitive end markets are “being challenged” by rising component prices and macroeconomic uncertainty — C.C. Wei’s own words — and 2027 planning is being kept deliberately prudent.
- Tool-price inflation is “real” — tools, materials and labor, per the CFO — compounding tight tool supply and raising the cost of the record build-out.
- DCE declined 12% QoQ to a 1% revenue share, the only platform to shrink.
- Pressed twice for an updated AI accelerator CAGR — by Laura Chen (Citi) and Arthur Lai (Macquarie) — management offered no number beyond “probably will be higher” than the prior mid-to-high-50s% framing (2024–2029).
1. Q3 Scoreboard and Q4 Guidance
All revenue figures in USD unless noted. The quarter landed above the top of the guided range on revenue; gross margin landed inside its guide band.
| Metric | Q3 2026 | QoQ change |
|---|---|---|
| Revenue | $46.7B | +16.2% USD (+17.6% NT) |
| Gross margin | 66.4% | −130 bps |
| Operating margin | 58.7% | — |
| Operating expenses | 7.8% of net revenue | — |
| EPS | TWD 29.43 | — |
| ROE | 45.5% | — |
Node contribution to wafer revenue:
| Technology | Share of wafer revenue |
|---|---|
| N2 | 8% |
| N3 | 33% |
| N5 | 28% |
| N7 | 10% |
| Advanced (7nm and below) | 79% |
Platform contribution:
| Platform | QoQ change | Share of revenue |
|---|---|---|
| HPC | +22% | 68% |
| Smartphone | +10% | 21% |
| IoT | +5% | 4% |
| Automotive | +23% | 4% |
| DCE | −12% | 1% |
Fourth-quarter 2026 guidance:
| Item | Q4 2026 guide |
|---|---|
| Revenue | $52.3–54.3B |
| Sequential growth at midpoint | +14% |
| YoY growth at midpoint | +58% |
| Gross margin | 65–67% |
| Operating margin | 56.5–58.5% |
| Exchange rate assumption | TWD 31.8 / USD |
2. Margin Mechanics: 2nm Dilution Now, Overseas Dilution Later
The sequential margin decline this quarter is attributed entirely to the initial 2nm ramp (see Lowlights). For Q4, the same driver continues “from the steep ramp-up of our 2 nm technology” and is joined by dilution from overseas fabs — partially offset by cost-improvement efforts including productivity gains and cross-node capacity optimization (see the guidance table).
Across the five-year overseas ramp, the shape is explicitly re-drawn: thanks to company-wide scale and cost work, early-stage dilution now comes in “closer to 2%,” then widens — the figure flagged in Lowlights — as more U.S. fabs ramp. When UBS pressed on what drives the widening, the CFO answered that TSMC is “accelerating everywhere, including in the U.S.” and that the widening endpoint “still remain[s] the case.” Management’s counterweight claim, in the CFO’s words: “TSMC continues to be the most cost-efficient manufacturer in every region that we operate.”
On the longer arc, C.C. Wei told Citi: “there is no ceiling for TSMC’s profitability, but a dramatic increase will not happen overnight either” — with the stated mechanism being customer success first, value capture after.
3. Demand: AI Intensity Versus Consumer Frailty
The demand case broadened rather than just deepened. Three new structural vectors were named: the transition from chatbots to AI assistants (agentic AI), broader adoption of open-source AI models that “help lower the barrier to future AI usage,” and — new this quarter — the emergence of agentic AI CPUs alongside continued strong networking-chip demand. The conclusion management drew: “all the AI players are working with TSMC,” and semiconductors’ share of value in the AI supply chain rises.
The quantified frame for this — the prior mid-to-high-50s% AI accelerator CAGR through 2029 (see Lowlights) — was left standing but labeled as understated rather than updated. Note its narrow construction: agentic AI CPUs are excluded from the accelerator definition, because TSMC says it cannot identify which CPUs are conventional-server versus AI-server usage; management calls that exclusion “a narrow definition, but it is certainly part of our business.”
At the edges, the story is choppier. Consumer and price-sensitive end markets are being challenged (see Lowlights), which is the stated reason 2027 planning is being kept prudent while “the fundamentals of our business” are strengthened. On AI-side supply checks, management sent its own teams to verify U.S. data-center build-out progress and power availability; the feedback so far: “so far so good, and the CSPs has been well-prepared.” The aggregate verdict repeated across the Q&A to three different analysts: demand is “far above our capability today to supply.”
4. The Build-Out: Nearly Twenty Fabs and a Broken Node Rule
Scale first: TSMC is building nearly 20 fabs worldwide — a plan of 13 leading-edge and advanced-packaging fabs in Taiwan over the next several years, plus five to six fabs overseas.
On N2: high-volume manufacturing began in Q4 2025 “with good yield,” the technology is ramping in multiple phases at both Hsinchu and Kaohsiung, and the N2 family’s annual capacity is set to more than double in 2027, supported by both smartphone and HPC AI demand; manufacturing-excellence output gains are layered on top. Beyond N2, capacity preparation for A14 has begun: risk production is “well on track for 2027,” with volume production in 2028.
On N3, the company did something it says it has never done — historically it adds no capacity once a node “reaches its targeted capacity.” The Q3-2026 plan adds three new fabs on top of converting 5nm tools to 3nm:
| Fab | Location | Volume production |
|---|---|---|
| New 3nm fab | Tainan, Taiwan | 1H 2027 |
| Additional 3nm fab (newly decided) | Tainan, Taiwan | 1H 2028 |
| Fab 2 | Arizona, USA | Q3 2027 (slightly ahead of prior schedule; tools now in installation) |
| Fab 2 (3nm) | Japan | 2028 |
The posture around it is maximalist, in C.C. Wei’s words: “We have no intention to leave food on the table for others, and we want to capture every piece of business possible.” Flexibility is explicitly engineered across N7, N5 and N3.
5. Capital, Cash and the Balance Sheet
The corollary framing for the elevated outlay, in the CFO’s words: “a higher level of capital expenditures is always correlated with higher growth opportunities in the following years.” The full-year budget sits where the Highlights state it, and the 2026–2028 step-up is framed as a shareholder-item rather than a concession: even at this spend, TSMC commits to “profitable growth” and a “sustainable and steadily increasing cash dividend per share” on annual and quarterly bases. Tool supply is tight — management says it collaborates with tool suppliers “well in advance,” mirroring how customers book TSMC capacity early — and tool-price inflation is real but treated as a secondary variable: “the majority of the CapEx is always related to the long-term demand.” On capital intensity, the standing claim was reaffirmed: “if [we] do our job right, the increase in revenue will outpace the increase in CapEx, and that is still the case.”
Third-quarter cash-flow and balance-sheet detail (NT$):
| Item | Amount |
|---|---|
| Cash from operations | ~TWD 920B |
| Capital expenditures | TWD 443B ($13.8B) |
| Dividends distributed (for Q4 2025) | TWD 156B |
| Cash balance change | +TWD 488B, to TWD 3.6T |
| Cash + marketable securities | TWD 4.0T ($126B) |
| Current liabilities | +TWD 110B (incl. +TWD 46B bond/bank-loan current portions; +TWD 26B dividend payable) |
| Accounts receivable turnover days | 28 days (−1) |
| Inventory days | 82 days (−5, on higher wafer shipments) |
6. Competitive Posture: Pricing, Packaging, Tools and Adjacent Partners
Pricing. Asked about reported price increases across mature and advanced nodes (an analyst-reported figure), the CEO called pricing “very strategic” — varying “by the market segment, by the location, and by the technology” — with U.S.-fab output priced differently from Taiwan-fab output: “we try to get our value to be recognized.” He did not specify whether increases apply to the contracted book or only incremental capacity.
Advanced packaging. The lead claim: TSMC is “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.” Development runs 9x toward 14x reticle sizes, CoWoS carries a stated cost-reduction roadmap and remains “the number one choice” of customers. Asked whether TSMC would take customers stranded by a competitor’s (EMIB-T) shortfall, the answer: “I try to avoid to answer it directly, but let me say that, yes, we will be.”
Capacity hygiene. On the secondary wafer market some customers discuss: “we do not allow our customer to sublet or auction off the capacity,” with mask-licensing collaborations the only sanctioned workaround and amounts “very, very small.”
Partnered supply. The GlobalFoundries interposer arrangement (an analyst-reported announcement, confirmed by the CEO) is framed as “part of our overall strategy in the back-end supply” — an admission that TSMC has “a little bit shortage on the back end” and prefers partners to unserved demand.
Adjacent build-out. Teams are evaluating Texas as a second U.S. site, and ongoing discussions aim at TSMC utilizing capacity for a customer’s own “Terafab” — with the non-negotiable already stated: “the fab has to be run by TSMC.” No framework (à la JASM or ESMC) has been settled.
Tools and the roadmap. High-NA EUV is timed for 2030, and only then, on the triple gate of best technology, reasonable cost and — “most important” — best maturity; a 12-inch photomask pipeline with ASML is targeted for 2031. Being later than competitors on High-NA is dismissed as a disadvantage: “we have confidence that we are still the best company to offer the most advanced technology.”
IDM treatment. On insourcing risk from an aggressively-expanding IDM competitor, TSMC “treat[s] every customer fairly,” accounts for customers’ internal capacity in its own plans, declines customer-level commentary — and adds a demand datapoint: A14 customer activity is running above N2’s at the same lifecycle stage, and “we expect that the A14 will be a bigger node than our current N2 technologies.”
7. Conclusion
This quarter, supply — not demand — became the binding constraint, and the report reads that way end to end: management’s unanimous claim is that orders exceed everything the company can build, and every action (a broken node rule, three new fabs, a bigger multi-year capex pledge, partner-supply deals) is shaped by that single fact. The cost of acting at this speed is visible and un-hidden: margins are absorbing the steepest 2nm ramp and a widening overseas-fab drag simultaneously, and management chose honesty about that cadence over a tidy bridge. The genuine tension is between the conviction embedded in the build-out and the careful caution being applied to consumer-facing demand into 2027 — one of those will have to give in the next two quarters. For the investment view, the franchise’s earning power now hinges less on AI demand existing (management treats that as settled) and more on whether disciplined cost execution can hold the margin line while the U.S. footprint multiplies.
8. Artifacts and Follow-Ups
Key Artifacts
- Q4 gross margin versus the guided band (65–67%, midpoint further down QoQ). Management claims productivity gains and cross-node capacity optimization partially offset 2nm and overseas dilution; the print either validates those offsets or shows the dilution compounding faster than absorbed.
- Full-year 2026 delivery against the ~44% YoY revenue guide. Also the lens for the next print’s Q1 2027 guide: the “prudent” consumer-end framing should show up either in the first 2027 guide or in platform mixes first.
- N2 ramp proof. Baseline: N2 at 8% of wafer revenue this quarter, with the N2 family’s annual capacity pledged to more than double in 2027 — watch wafer-mix progression and yield commentary as the steep ramp continues.
- Overseas-fab dilution trajectory. Early-stage dilution is now described as closer to 2%, widening to 3–4 points later in the five-year period; watch whether realized gross-margin drag stays below the endpoint as Arizona scale builds.
- The 2027 capital-budget disclosure. Baseline: 2026 budget reiterated at $60–64B, and 2026–2028 spending guided “even more significantly higher” than the past three years — the next call’s budget line tests that pledge.
Secondary Tracking Items
- Texas/Terafab site evaluation — a conclusion or settled framework, with the TSMC-run precondition intact.
- Progress on the three-fab N3 expansion and the five-to-six-fab overseas plan (see the build-out table).
- A14 capacity preparation and risk-production progress toward its stated window.
- Any re-quantification of the AI accelerator CAGR beyond “probably will be higher.”
- Advanced-packaging roadmap execution (9x toward 14x reticle work) and CoWoS availability.
- Consumer-side platform stabilization (DCE, IoT) against rising component-price commentary.
- High-NA adoption-gate milestones (see the tools-and-roadmap note).
- Evidence on dual-architecture (Super Power Rail vs conventional) learning curves.