Gross Margin 200bps Above Guide, FY25 Raised to Mid-30s, Overseas Dilution Cut in Half — Maintain Outperform

Results vs. Consensus

Quality of Beat

• Revenue: Clean and operational. USD revenue of $33.10B (+40.8% YoY) actually accelerated from Q2's +44.4% YoY pace when we strip out the favorable FX tailwind Q2 enjoyed from the weak NT$. On a constant-currency basis, underlying growth is roughly flat-to-slightly-accelerating from already elevated levels. The monthly revenue cadence through Q3 (August and September stronger than July) is the opposite shape of customer pull-forward; this is demand-driven, not inventory-driven.
• Margins: The blowout. Gross margin at 59.5% is +90bps sequentially on top of Q2's 58.6%, and came in 200bps above the high end of the Q3 guide range (55.5-57.5%). Management had explicitly flagged 180-200bps of NT$ appreciation-driven compression in Q3 — that evidently did not materialize, because NT$ stabilized around 30-31 per USD after the late-June/early-July spike. Separately, the "cost improvement efforts and higher capacity utilization rate" language from CFO Huang implies that fab throughput is delivering ahead of plan. Operating margin of 50.6% (+1.0pp QoQ) is the highest in TSMC's history.
• EPS: NT$17.44 represents +50.5% YoY EPS growth on +40.8% revenue growth — a ratio of 1.24x, slightly below Q2's 1.36x but still signature operating leverage. The +12.3% EPS beat versus consensus is the largest in the ADR series for two years and reflects both the margin beat and the NT$ stabilization that favored translation.
• FX optics: The big change from Q2 to Q3 is that the Q3 margin worry proved a phantom. NT$/US$ averaged ~30.4 in Q3 vs. the end-of-Q2 panic level of ~29.3. That 1%+ NT$ depreciation versus the scare-case contributes ~40bps of the margin outperformance. The rest is genuine operational improvement — cost reductions, yield maturation on N3, and incremental utilization on HPC.

Segment Performance

The platform mix optically shifted away from HPC in Q3 — but this is the healthy kind of shift: smartphone, IoT, and automotive all grew 18-20% QoQ into the Q4 Apple/flagship-Android ramp, while HPC held flat sequentially at record levels. The rest of the portfolio finally woke up to join the party.

HPC: Still the Plant, Not Just the Flower

The surface read on HPC is "share down 3pp from 60% to 57% — is AI peaking?" The correct read is: on a $33.10B revenue base, HPC revenue was approximately $18.9B in Q3, up roughly +60-65% YoY and flat-to-slightly-up sequentially from Q2's ~$18.0B. The 3pp share decline is an entirely arithmetic result of smartphone, IoT, and automotive growing 18-20% sequentially while HPC held near record levels. Wei was unequivocal about the demand signal itself.

"The AI demand actually continue to be very strong. It's more — more stronger than we thought three months ago." — C.C. Wei, Chairman & CEO

"We are receiving very strong signals from our customers' customers" for additional capacity. — C.C. Wei, Chairman & CEO

The "customers' customers" framing is the critical tell. TSMC's direct customers are Nvidia, AMD, Broadcom, Apple, Marvell — but "customers' customers" means the hyperscalers (Microsoft, Google, Meta, Amazon) and emerging sovereign AI buyers. Wei is saying the end-demand signal has strengthened, not just the channel fill. This is a second-derivative statement that matters more than the share print.

Assessment: HPC is the through-cycle driver. The optical dip in Q3 share is a feature (diversification across platforms), not a bug (AI weakness). With CoWoS still capacity-constrained despite doubling and 2026 expansion committed, HPC dollars will grow through 2026 even if share sits at 55-60%.

Smartphone: The Expected Q3 Ramp, Bigger Than Expected

Smartphone revenue at 30% of mix represents approximately $9.9B in Q3, up +19% QoQ from Q2's ~$8.1B. This is the pre-iPhone 17 production ramp running at the usual cadence — but the sequential growth rate is slightly above typical pattern, suggesting both flagship Android and Apple are pulling harder than typical. Smartphone-HPC duality matters because smartphone uses N3 heavily, which continues to keep 3nm utilization full on the Apple A19 series.

Assessment: Q3 smartphone strength is bullish for Q4 revenue guidance and for N3 margin profile. Watch Q4 for confirmation that smartphone strength extends through the full Apple launch cycle. The Q4 revenue guide of $32.2-33.4B implies only a modest sequential decline (~-1% to +1% QoQ), which is consistent with smartphone normalizing post-ramp but HPC absorbing the slack.

Automotive and IoT: Finally Moving

The 18-20% sequential growth in automotive and IoT is the cleanest data point in the release that demand is broadening beyond hyperscale. Auto has been "bottoming, not breaking" for three quarters; the Q3 ramp is the first meaningful sequential move and suggests 2026 auto silicon is re-engaging. IoT's 20% QoQ jump is a combination of edge-AI inference chips ramping and industrial automation customers increasing orders as macro stabilizes.

Assessment: Auto and IoT went from immaterial swing factors to small positive contributors. Collectively they are now ~10% of revenue and growing ~18-20% QoQ; if that pace sustains into 2026, they become a second growth leg behind HPC.

Technology Node Mix

The node mix is remarkably steady at the consolidated level (74% advanced vs. 74% in Q2), but this hides the underlying dollar growth: total revenue at 74% advanced-node share on a $33.10B base means $24.5B of advanced-node revenue in Q3, up roughly +55% YoY from approximately $15.8B in Q3 2024. The 74% share will likely step up to 75-76% in Q4 as N3 continues ramping behind Apple A19. And N2 volume production is confirmed for Q4 2025 — ahead of the half-year frame most analysts had modeled.

Key Topics & Management Commentary

Overall Management Tone: The tonal arc from Q1 to Q3 traced a clean progression — from Q1's defensive maintenance ("we have not seen any change in customer behavior") through Q2's offensive confidence (raising the FY guide to ~30%) to Q3's almost triumphant framing. Wei explicitly told the Street demand is stronger than he thought 90 days ago, used the word "exponential" to describe token growth, and committed that "next year looks to be a healthy year." The defensiveness of Q1 is gone. Huang remained measured and quantitative but did offer the biggest structural positive of the call — revising FY25 overseas fab dilution DOWN from 2-3% to 1-2%. Tariffs got a single measured mention on consumer and price-sensitive segments — a small signal shift worth noting but not alarming.

1. The FY25 Guide Raise — Second in a Row

This is now the second consecutive quarter in which TSMC has raised its FY25 revenue guide. The trajectory: Q1 "close to mid-20s%" → Q2 "approximately 30%" → Q3 "mid-30s percent." On a FY24 base of ~$90B, "mid-30s%" implies FY25 revenue of approximately $121-124B — a ~$4-6B incremental revenue upgrade embedded in just this quarter's revision, on top of the ~$3-5B upgrade from Q2. The cumulative FY25 upgrade from April's "close to mid-20s" outlook is approximately $11-13B, or roughly 10-12% above where consensus entered Q2.

"Full-year 2025 revenue to grow in the mid-30s percent year-over-year in US dollar terms." — Wendell Huang, CFO

This cadence — two consecutive mid-year guide raises — is without precedent at TSMC. Historically the company preserves optionality at the high end and lets monthly revenue data speak for itself. Deliberately stepping up the guide in both July and October signals management has concrete visibility into Q4 and H1 2026 and is comfortable enough to codify that view for the Street.

Assessment: The guide raise is the most important forward indicator in the release. It does two things: (1) locks in FY25 earnings upgrades that will flow through to FY26 base effects; (2) implicitly pre-commits management to a confident 2026 framework at the January 2026 print. If management were seeing demand weaken, they would be pocketing upside rather than adding to the guide.

2. The Margin Surprise: Everything You Worried About at Q2 Went the Other Way

At Q2, the single biggest bear data point was the Q3 gross margin guide of 55.5-57.5%, implying 100-250bps of compression primarily driven by NT$ appreciation. We argued in the Q2 recap that this was "cosmetic FX headwind, not structural." The Q3 print vindicates that view emphatically. Gross margin came in at 59.5% — above the 58.6% Q2 delivery, and 200bps above the guide high end of 57.5%.

"Gross margin increased 0.9 percentage points sequentially to 59.5%, primarily due to cost improvement efforts and a higher capacity utilization rate." — Wendell Huang, CFO

The drivers of the margin beat are clean: (1) NT$ stabilized around 30-31 per USD instead of appreciating to 28 or below as the bear case required; (2) capacity utilization held at elevated levels; (3) cost-improvement programs (which TSMC consistently underplays in guidance) delivered more than flagged.

The Q4 gross margin guide of 59-61% is the more important forward signal — it confirms that the Q3 margin level is not a one-quarter outlier. Underlying unit economics are at or near cycle highs and still improving.

Assessment: The "margin peaking at 58%" bear narrative that bubbled up at Q2 has been decisively buried. Gross margin is expanding, not peaking. The Q4 guide midpoint of 60% is above where the Street was modeling FY25 full-year margin just 90 days ago.

3. Overseas Fab Dilution: Cut in Half for FY25

This is the quietest but most structurally meaningful announcement in the release. Through Q1 and Q2, management held the FY25 overseas fab dilution guide at 2-3%. At Q3, Huang revised this DOWN to 1-2% — effectively cutting the expected dilution in half.

"Expect gross margin dilution from overseas fabs ramp to be closer to 2% in 2025... 1-2% for full year." — Wendell Huang, CFO

The long-term dilution framework (2-3% early stage, 3-4% later stage) was maintained, but the near-term pressure is materially lower than previously indicated. The implication: Arizona Fab 21 Phase 1 is running at better unit economics than previously modeled, likely driven by N4 yield maturation comparable to Taiwan, disciplined operating expense, and perhaps accelerated utilization. This also reframes the "Arizona is a structural margin wound" bear argument: at 1-2% dilution on a 60% gross margin, the absolute margin drag is ~60-120bps — easily absorbed by the mix shift into HPC.

Assessment: This is one of the most consequential updates in the release. The Arizona margin drag was the single most-cited structural bear argument going back to 2023. Cutting it in half for FY25, with long-term framework unchanged, means the margin story is less about "can we absorb the dilution" and more about "how high can margins go." The answer, based on Q4 guide, is higher than 60%.

4. CoWoS and Advanced Packaging: Now ~10% of Revenue

Wei disclosed that advanced packaging (CoWoS plus related back-end services) revenue is approaching 10% of total revenue. On a $33.10B Q3 revenue base, that is roughly $3.3B of packaging revenue in a single quarter — up from perhaps $2-2.3B a year ago. The capacity has approximately doubled from 2024 as committed, and management confirmed further expansion into 2026. On the critical 2026 capacity question, Wei was candid about the gap still existing but declined specifics.

"We are still working to increase the capacity in 2026. The real number, we probably update you next year... everything related, like front-end and back-end capacity, is very tight." — C.C. Wei, Chairman & CEO

"We are working very hard to narrow the gap between the demand and supply." — C.C. Wei, Chairman & CEO

The "we are working to narrow the gap" language repeats the Q2 framing but in a context where 2025 capacity has already doubled — meaning demand has grown proportionally if not more. The decision to withhold a specific 2026 capacity number is consistent with TSMC's pattern of under-promising and over-delivering on capacity; it is not a signal of demand softness.

Assessment: CoWoS capacity remains THE leading indicator for AI accelerator revenue. Doubling in 2025 already absorbed, further expansion coming in 2026, and demand still outstripping supply. The Q2 thesis that "another year of capacity-constrained AI accelerator growth in 2026" has been confirmed. AI accelerator revenue for 2025 is now tracking to $45-50B, with 2026 likely another ~20-30% growth year.

5. The "Foundry 2.0" Framework and System-Level Thinking

One of the more interesting strategic disclosures was Wei's introduction of "Foundry 2.0" framing — integrating front-end wafer manufacturing with back-end advanced packaging and system-level optimization. This is partly a narrative positioning (responding to "Moore's law is dead" commentary from customers like Nvidia), but it has concrete business implications: the advanced packaging revenue line is now large enough to matter, and TSMC is moving from a wafer-only value proposition to a system-performance value proposition.

"N2, N2P, A16 and its derivatives will propel our N2 family to be another large and long-lasting node." — C.C. Wei, Chairman & CEO

The important commercial implication: as customers care more about system performance than single-chip performance, they become more willing to pay for integrated back-end services. This is why advanced packaging revenue is growing faster than front-end wafer revenue, and why gross margin is expanding rather than compressing. TSMC is moving up the value stack.

Assessment: Foundry 2.0 is the narrative version of a margin tailwind. The moat extends beyond process technology into the back-end, which is harder for Samsung or Intel to replicate than node-level process development. Expect this framing to be central to the January 2026 call when 2026 guidance is given.

6. N2 on Track for Q4 Volume Production; A16 for H2 2026

N2 is confirmed for volume production "later this quarter" — meaning Q4 2025. This is slightly ahead of the "H2 2025" framing used at Q1 and consistent with Q2's positive tape-out commentary. A16 (backside power delivery) remains on track for H2 2026 with HPC focus. The operational cadence on both nodes is one of the clearest competitive moats TSMC has: Samsung's 3nm yields have not stabilized commercially, and Intel 18A remains an unproven commercial proposition at scale.

Assessment: Node leadership is widening, not narrowing. The N2 family (N2, N2P, A16) will be a 3-4 year workhorse, and the revenue contribution from N2 in 2026 H2 and 2027 will drive another leg of margin expansion at the mix level. TSMC's process-technology moat is the safest bull thesis point in the stack.

7. Tariff Commentary Returns — Modestly

After Q2's near-total absence of tariff discussion, Q3 had one measured acknowledgment from Wei regarding "uncertainties and risk from the potential impact of tariff policies, especially in consumer-related and price-sensitive end-market segments." This is a small signal shift worth flagging: it is the first time in four quarters Wei has explicitly mentioned tariff risk as a current-quarter factor, rather than a hypothetical that has not materialized.

"Uncertainties and risk from the potential impact of tariff policies, especially in consumer-related and price-sensitive end market segment[s]." — C.C. Wei, Chairman & CEO

The specific call-out of "consumer-related and price-sensitive" segments is notable. This would include some PC, some low-end automotive, and some smartphone — NOT HPC, NOT AI, NOT flagship smartphones (which have been consistently ordered). Wei is essentially saying: our top-end business is fine, but there may be some tariff passthrough issues at the mass-market edge.

Assessment: This is a small bear data point, but it is a first-time acknowledgment of tariff-related customer hesitancy in specific pockets. Worth monitoring in Q4; not thesis-breaking. The HPC, N3, N2, and advanced packaging stories — the drivers of the earnings power — were not flagged as at risk.

8. Intel Relationship: Competitor and Customer

An analyst asked about the Intel relationship (Intel is both a foundry competitor and a TSMC customer for leading-edge CPUs). Wei's response was straightforward without dodging the duality.

"That competitor happened to be our customer, very good customer." — C.C. Wei, Chairman & CEO

Management declined further comment on whether any formal TSMC-Intel partnership or JV arrangement was under discussion — the same answer given at Q1 and Q2. Given that Trump administration pressure on Intel has quieted since the April 2025 Arizona commitment announcement, the JV rumor has largely faded.

Assessment: No news is good news here. TSMC is extracting revenue from Intel as a customer while maintaining process-technology distance. The JV scenario that would have been dilutive has receded as a tail risk.

Guidance & Outlook

The guide-by-guide cadence is the clearest evidence of an accelerating business. For FY25 revenue growth:

• January 2025 initial: ~20% USD growth (pre-Q1 framework)
• Q1 2025 guide: "close to mid-20s percent" — modest upward bias
• Q2 2025 guide: "approximately 30%" — first explicit raise
• Q3 2025 guide: "mid-30s percent" — second explicit raise

On a FY24 base of ~$90B, "mid-30s%" growth implies FY25 revenue of approximately $121-124B. Through Q3, TSMC has delivered approximately $88.7B USD revenue for H1+Q3 combined ($25.5B Q1 + $30.1B Q2 + $33.1B Q3). The Q4 guide midpoint of $32.8B puts FY25 at approximately $121.5B — cleanly at the midpoint of the new "mid-30s%" framework.

Implied FY26 setup: If FY25 lands at $121-124B, and if management's "healthy year" 2026 framing implies 15-20% USD growth, FY26 revenue would be approximately $139-149B. At the midpoint, that is $145B — above current consensus of ~$140B. Management has structurally moved the 2026 consensus higher without formally guiding.

Q4 revenue guide of $32.2-33.4B — flat QoQ: The bear reading is that sequential growth stalling at $33B is an early sign of AI cycle maturing. The bull (and correct) reading is that Q3's $33.1B already captured the smartphone Q4 ramp, and Q4's flat sequential is consistent with HPC holding while smartphone normalizes post-iPhone launch. Management's FY25 mid-30s framing means Q4 could land closer to $33.4B (high end), which would imply H2 acceleration rather than deceleration.

Gross margin setup: The Q4 gross margin guide of 59-61% is the new cycle high by a wide margin. At the midpoint of 60%, this is 140bps above Q3's actual and 350bps above where the Q3 guide implied. The implicit FY25 gross margin will land around 59%, up from 56-58% modeled at Q1. The raised margin profile drops $3-4B of incremental operating income to the bottom line versus January's starting framework.

Guidance style: TSMC has now beaten the high end of its own revenue guide for seven consecutive quarters and the high end of its gross margin guide in most of them. Historical pattern suggests the FY25 "mid-30s%" framework contains another 100-200bps of upside; our base case is FY25 lands at 35.5-37% USD growth.

Analyst Q&A Highlights

Quality of the FY25 Guide Raise and 2026 Visibility

• Morgan Stanley: Probed whether the mid-30s FY25 target was driven by customer pull-forward or true organic demand. Wei emphasized the "stronger than three months ago" demand signal and the "customers' customers" direct engagement, explicitly stating he sees no pull-forward pattern.
  Assessment: Same answer as Q2 but with stronger affirmation. The monthly revenue data supports this: August and September were the strongest months of Q3, not the weakest, consistent with demand-driven rather than inventory-driven revenue.

2026 AI Accelerator Growth Rate

• Bernstein: Pressed Wei on specific 2026 AI accelerator revenue growth given the high 2025 base. Wei declined to quantify but confirmed "next year looks to be a healthy year and we are confident on the megatrend." He referenced token volume "exponentially increasing" every 3 months.
  Assessment: "Healthy year" plus "exponential" token growth implies at least mid-teens AI accelerator revenue growth in 2026 and likely high-teens-to-20s. This is more bullish than the Q2 commentary on 2026.

CoWoS 2026 Capacity Specifics

• Goldman Sachs: Asked for specific 2026 CoWoS capacity growth numbers given the committed expansion. Wei acknowledged ongoing expansion but declined to quantify, saying "the real number, we probably update you next year."
  Assessment: The refusal to quantify is consistent with TSMC's historical under-promise/over-deliver pattern on capacity. The positive signal is the explicit acknowledgment of tightness continuing into 2026. CoWoS capacity will likely grow another 30-50% in 2026 based on industry checks.

Q3 Margin Beat Sustainability

• JPMorgan: Asked Huang whether the 59.5% gross margin reflects a new structural level or a one-quarter FX benefit. Huang attributed the beat to cost improvements and utilization, with FX as a secondary factor. The Q4 guide of 59-61% confirms the structural nature.
  Assessment: This is the most important answer in the Q&A. If the 59%+ gross margin is structural (as the Q4 guide implies), then FY26 consensus EPS estimates need to move 10-15% higher. Several sell-side firms appear to be in the process of doing this.

Overseas Fab Dilution Revision

• Citi: Asked why the FY25 overseas dilution was revised DOWN from 2-3% to 1-2%. Huang attributed the revision to (a) Arizona Fab 21 Phase 1 yields comparable to Taiwan, (b) better-than-expected operating expense management, (c) higher utilization than planned.
  Assessment: All three factors are structural positives. Yield parity at Arizona especially was a 2024 risk that has now been definitively resolved. The long-term dilution framework (2-3% early stage, 3-4% later stage) was unchanged, but the near-term drag is materially smaller.

N2 Ramp and ASP Economics

• UBS: Asked about N2 pricing and whether early N2 wafers would be margin-accretive or dilutive. Wei said N2 is on track for Q4 volume production and customer tape-out pace is strong; Huang declined specific pricing commentary.
  Assessment: N2 volume production in Q4 2025 is actually 1-2 quarters ahead of where the Street had modeled at the start of the year. If ASPs are a step-function above N3 (as expected), N2 will be margin-accretive from early 2026 onward, providing another margin tailwind.

Tariff Comments on Consumer/Price-Sensitive Segments

• Needham: Flagged Wei's tariff callout on consumer segments and asked if it signals weakness in specific categories. Wei clarified this was a general risk-awareness statement, not a specific customer observation; HPC and flagship segments are fine.
  Assessment: Small information gain. The tariff risk exists at the consumer edge but is not currently visible in order books. Worth monitoring; not thesis-breaking.

2026 Capex

• Barclays: Asked whether 2026 capex will exceed 2025's $40-42B. Huang said "a higher level of CapEx is always going to be correlated with a higher growth opportunity" and reaffirmed "next year looks to be a healthy year." Specific 2026 capex framework deferred to January 2026.
  Assessment: This is a soft pre-commit to higher 2026 capex. A flat or declining capex trajectory would have been deeply bearish (suggesting demand concerns); continued growth in capex is consistent with a multi-year capacity expansion cycle.

What They're NOT Saying

1. Specific 2026 revenue framework: Wei confirmed 2026 will be "a healthy year" with "confidence on the megatrend" but declined to quantify. This is procedural (TSMC typically gives formal FY guide in January) but notable given how assertive the rest of the commentary was. A specific "15-20%" or "high-teens" framework would have been even more bullish.
2. Specific 2026 capex numbers: Huang said "healthy year" and implied higher than 2025, but no range. With CoWoS expansion continuing and N2 ramping, 2026 capex could plausibly exceed $45-50B — a meaningful step-up.
3. Specific CoWoS 2026 capacity numbers: Wei acknowledged tightness continuing and expansion ongoing, but no capacity targets. For a constraint this important to the AI supply chain, the vagueness is deliberate; but it leaves an information gap that some analysts will fill with assumption rather than fact.
4. Pricing actions and ASP by node: As with Q2, management declined to quantify N2 pricing, any N3/N5 ASP actions, or the per-wafer uplift from Foundry 2.0 services. This is consistent with TSMC's historical commercial confidentiality, but leaves the margin walk impossible to forecast precisely.
5. China revenue trajectory: China held at 9% of revenue in Q3 (same as Q2). Management did not discuss direction, particularly in light of the Nvidia H20 situation and ongoing export controls. The structural point — China is small and stable as a share — increasingly makes "China-risk" a stale narrative for TSMC.
6. Specific Section 232 scenario planning: Unchanged from prior quarters. The investigation is still technically ongoing, and management continues to deflect scenario questions. The policy environment has been quieter through Q3, but a formal Section 232 finding could still be delivered into 2026.
7. Buyback authorization or capital return expansion: With cash generation of ~$60B+ at current earnings power and a narrowed capex range of $40-42B, there is room for expanded capital return. No mention of expanded buyback. Huang reiterated "steady dividend growth" but declined a forward framework. Given TSMC's capex intensity, a major buyback is unlikely, but the absence of any signal continues to frustrate investors.
8. Specific 2026 HPC share target: After HPC dipped 3pp optically in Q3 to 57%, management did not commit to where share normalizes. The FY25 average will likely land at 58-59% and FY26 consensus assumes 60%+. Any explicit 2026 HPC share commitment would have been material; the silence leaves room for upside if HPC share re-expands in H1 2026.

Market Reaction

• Pre-earnings ADR (Oct 15, 2025 close): ~$304.71
• Pre-market Oct 16: Mixed; some initial enthusiasm quickly faded after guide details digested
• Close Oct 16, 2025 (earnings day): ~$299.84 (-1.6% on the day)
• Close Oct 17, 2025 (T+1): ~$295.08 (-1.6% further, -3.2% two-day)
• YTD performance entering the print: ~+55-60% from start of 2025
• From April 7 low ($134.25): approximately +125% to pre-print peak
• Analyst actions within 24 hours:
  • JPMorgan: Upgraded to Overweight from Neutral; PT raised to $550 from $370 — Street-high target
  • Goldman Sachs: Reiterated Buy; PT raised to $518 from $399
  • Goldman Taiwan-listed PT: NT$2,330 from NT$1,720 (+35%)
  • Broad PT lifts: Most firms moved to $325-$400 range; Morgan Stanley, Bernstein, Citi, Needham, Barclays, UBS, Mizuho all raised targets
  • FY25/26 EPS estimates: +8-15% upward revisions across the sell side
  • No downgrades reported within 48 hours

The modest -3% two-day dip on a print this strong is the single most interesting data point about market reaction. The disconnect is striking: analyst estimate revisions are clustering in the +10-15% range, price targets have been lifted 15-35%, and rating actions are overwhelmingly positive — yet the stock is down. Three factors explain the gap.

First, positioning. TSM had rallied ~90% from the April 7 low to the October print. Long-only managers and hedge funds were crowded long heading in. Any print — even a blowout — that does not include an unambiguously higher 2026 framework invites profit-taking from crowded positioning. The stock's price-action resembles the +2-3% pre-market pop that faded, followed by systematic reduction through the Oct 16 and Oct 17 sessions.

Second, AI-bubble narrative. October 2025 featured several high-profile "AI is a bubble" commentaries, and TSM became the vehicle for expressing that view given its concentration in AI accelerator wafer exposure. The Q4 revenue guide of roughly flat QoQ — technically appropriate given the FX reset and smartphone normalization — was misread as an early deceleration signal.

Third, the tariff callout. Wei's single measured mention of tariff risk in "consumer and price-sensitive" segments was the first such acknowledgment in four quarters, and some tariff-sensitive traders took it as a meaningful signal. It isn't — HPC, advanced nodes, and AI accelerators are not what Wei flagged — but it provided a narrative hook for profit-takers.

The informative signal is the analyst action: broad price target increases, the JPMorgan upgrade to Overweight with a Street-high $550 target, Goldman's PT lifted 30%+, and no downgrades. Historically, when prices dip and analyst targets rise simultaneously, the gap resolves in favor of fundamentals over 3-6 months. This pattern was also evident after the Q2 2025 print, which was followed by another 25%+ rally into October. We expect a similar mean-reversion through Q4 2025 and into the January 2026 FY26 guide print.

Street Perspective

Debate: Is the Modest Post-Print Dip a Sign of Peaking, or Profit-Taking?

Bull view: The stock is down -3% on the best earnings print in two years — clear profit-taking after a 90%+ rally from April. Every analyst metric supports the bull case: estimate revisions higher, price targets up, rating upgrades, no downgrades. The dip is positioning, not fundamental. As flows normalize through Q4 and the Street integrates the raised FY25/FY26 framework, the stock should resume upward trajectory.

Bear view: When a stock can't rally on a clear beat-and-raise, it is a bearish technical signal. The Q4 revenue guide of flat-QoQ is the first sequential deceleration point of the year. AI bubble fears are intensifying, and TSM is the cleanest vehicle for shorting that view given its concentration in AI accelerator wafers. Elevated forward P/E (~27-28x) leaves no room for error.

Our take: The bull view has materially better evidence. The stock's price action is consistent with profit-taking patterns we have seen repeatedly after major rallies (including after the Q2 print, which dipped then rallied). The fundamental signals — guide raise, margin beat, overseas dilution revision, CoWoS commitment — are unambiguously positive. We view the ~-3% dip as an opportunity to add, not a warning sign. The AI-peaking narrative requires Wei's "stronger than three months ago" commentary to be false, which is a high bar given monthly revenue data confirms it.

Debate: Margin — Structural 60%, or Temporary Peak?

Bull view: Q3 gross margin at 59.5% and Q4 guide at 59-61% codify 60% as the new structural level. Overseas fab dilution revised DOWN from 2-3% to 1-2% for FY25. N2 volume production starting Q4 2025 will be margin-accretive by 2026. Operating leverage combined with mix shift into advanced nodes drives the margin profile higher through 2026.

Bear view: Margins always revert. FX is currently neutral; if NT$ resumes appreciation in 2026, the 40bps-per-1pp sensitivity bites. Overseas fab dilution steps up to 2-3% in 2026 (maintained long-term framework) and 3-4% in later years. Electricity rate hikes in Taiwan are ongoing. The 60% margin level reflects cyclical-peak conditions that are unlikely to sustain through 2027.

Our take: Bulls have the stronger case through 2026. The "margin peaking at 58%" narrative from Q2 has been decisively falsified; that same narrative at 60% will likely be falsified again. N2 mix accretion and continued HPC share should support 60%+ gross margin through 2026. The bear case becomes more compelling in 2027 when overseas dilution ramps more aggressively — but that is a concern for Q2-Q3 2026 recap, not for today.

Debate: 2026 AI Accelerator Growth — Mid-Teens or High-Twenties?

Bull view: Wei's "exponential" framing on tokens, CoWoS capacity continuing to expand in 2026 while remaining tight, and "customers' customers" sending stronger signals, all support high-twenties growth in AI accelerator revenue for 2026. Apple continues scaling its AI silicon; hyperscaler ASICs ramping; sovereign AI creating new customer categories.

Bear view: 2025's doubling is unrepeatable. Hyperscaler capex is approaching peak. Sovereign AI deals are concentrated with specific governments and may not scale linearly. Inference workload expansion has been promised but not fully materialized at enterprise scale. 2026 AI accelerator growth will be in the mid-teens at best, triggering multiple compression on decelerating growth.

Our take: Bulls win the 2026 debate; bears have a clearer case for 2027-2028. Combining the CoWoS capacity signal (continued expansion, still tight) with Wei's "healthy year" framing and the order-book-visible nature of AI accelerator wafers, 2026 growth should be 20-25%+. That is below 2025's ~2x doubling but still exceptional by any historical standard. The more interesting 2026 dispersion is how quickly inference-driven demand expands to absorb an eventual training investment plateau.

Debate: Valuation — Full, Fair, or Still Cheap?

Bull view: At ~$295 and FY26 EPS estimates moving toward $13-14, TSM trades at ~22-23x forward P/E. For a business growing revenue 20%+ with 50% operating margins, 1-2% overseas dilution, and widening process-technology leadership, this multiple is below historical average. Peer-group AI beneficiaries trade at 30-45x, making TSM the cleanest value proposition in the AI supply chain.

Bear view: The stock has nearly doubled from April. Much of the re-rating is done. At ~27-28x trailing or ~22-23x forward, there is limited room for multiple expansion. Any disappointment in the January 2026 FY26 guide triggers a sharp re-rate lower. Geopolitical tail risk is never zero. The asymmetry has shifted from wildly favorable (April) to merely favorable (today).

Our take: Valuation is fair but not stretched. Our base case is that FY26 EPS (ADR) lands at $13-14, with 22-25x target multiple implying a 12-month target range of $290-$350. Upside of 0-20% from current levels with downside (bear case 18x on $11 EPS = $200) of ~30%. The risk/reward is less extreme than April but still attractive at ~1.0-1.5x up-to-down ratio. For patient holders, the risk/reward supports maintaining position; for new money, current levels are reasonable but not exceptional.

Debate: Does the Tariff Callout Signal a Turn?

Bull view: Wei's single mention of tariff risk in "consumer and price-sensitive" segments is a general risk-awareness statement that explicitly excludes HPC and flagship categories — which are the earnings drivers. Section 232 remains technically open but the policy environment has quieted since April. The tariff tail risk is contained.

Bear view: First time in four quarters Wei has acknowledged tariff risk as a current factor. The "consumer-related" language may foreshadow trimmed consumer-device orders in Q4 that flow through to lower mature-node utilization and some mix headwind. Separately, the 2026 US election cycle is now 12 months away — policy brinksmanship could re-escalate.

Our take: Small bear data point but not thesis-breaking. The HPC/N3/N2 portions of the portfolio — which represent 70%+ of the earnings power — were not flagged. Mature-node mix softness in tariff-sensitive categories would be a minor 30-60bps margin issue at most, easily absorbed by HPC mix. Worth monitoring in Q4; not worth repositioning for.

Model Update & Valuation Framework

Valuation framework: At a ~$295 ADR price, TSM trades at approximately 27-28x our FY25 EPS estimate of $10.40-11.00 and 21-24x our FY26 EPS estimate of $12.50-14.00. For a business growing revenue mid-30s in 2025 and mid-teens-to-low-20s in 2026, with 50% operating margins and widening process-technology leadership, this multiple is fair but not stretched. A 24-26x multiple on FY26 EPS of $13 implies a 12-month price target range of $310-$340 — upside of 5-15% from current levels. A more bullish re-rating to 28x on $14 implies $392 — upside of ~33%.

Bull-case scenario: AI accelerator revenue in 2026 grows 25%+; gross margin sustains 60%+ through H1 2026 as NT$ stays stable; N2 volume production ramps to 8-10% of wafer revenue by Q4 2026. FY26 EPS lands at $14.50-$15.00. Target multiple of 26x implies $375-$390 — upside of ~25-30%.

Bear-case scenario: 2026 AI accelerator growth moderates to 12-15% as hyperscaler capex flattens; Taiwan dollar appreciates meaningfully, pushing gross margins to 55-57% in H2 2026; a Section 232 outcome triggers 10-15% consumer-electronics tariffs. FY26 EPS lands at $11-11.50; at a compressed 20x multiple, implies $220-$230 — downside of ~22-25%.

Risk-reward calculus: Upside of ~25-30% (bull) versus downside of ~22-25% (bear) implies a favorable asymmetry of roughly 1.1-1.3x. Less extreme than April or July, but still positively skewed given the fundamental trajectory. The through-cycle earnings power at 58-60% gross margin and 48-50% operating margin is now codified, which compresses the variance of forward estimates.

Thesis Scorecard: Q2 Watch Items Revisited

Our Q2 2025 recap laid out eight specific signposts for Q3. Every single one tripped favorably, with several exceeding the bullish threshold.

Thesis Scorecard: Full Bull / Bear Matrix

Overall: Thesis materially strengthened through Q3. Every primary bull point either confirmed or strengthened; one new bull point added (advanced packaging as a 10% revenue leg). The Q2 FX bear point has been definitively resolved in favor of the bulls. One bear point (tariffs in consumer segments) partially reactivated as a small signal shift. One new bear point added (valuation) but at modest intensity. Net-net, the operational thesis is stronger than at either initiation (Q1) or the Q2 maintained rating. The valuation headwind is real — the stock has done a lot of work — but the fundamental trajectory supports continued upside even at current multiples.

Action: Maintaining Outperform. Risk/reward has compressed modestly versus Q2 (less extreme upside, comparable downside), but the fundamental case is stronger than at any prior point in coverage. For investors already long, this is a quarter to add through the profit-taking dip; for new money, the ~-3% post-print pullback is a reasonable entry into a thesis that has now been vindicated across three consecutive earnings prints.

Bottom Line

TSMC just delivered its strongest earnings beat in two years, with gross margin 200bps above the guide high end, FY25 revenue growth raised to mid-30s percent for the second consecutive quarter, and overseas fab dilution revised DOWN from 2-3% to 1-2%. Every operational question flagged at Q2 has been answered, and the answers are overwhelmingly favorable. AI demand is characterized as "stronger than three months ago" with "exponential" token growth driving "customers' customers" direct engagement. CoWoS capacity has doubled in 2025 and continues to be tight with expansion into 2026 committed. N2 volume production confirmed for Q4 2025, ahead of initial schedule.

The modest -3% two-day stock dip on a print this strong is profit-taking after a roughly 90% rally from April lows, not a fundamental concern. The disconnect between price action (-3%) and analyst reaction (JPMorgan upgrade to Overweight with Street-high $550 PT, Goldman to $518, broad PT lifts of 15-35%, no downgrades, FY25/FY26 EPS revisions of +8-15%) historically resolves in favor of fundamentals over 3-6 months. That mean-reversion has twice played out since April, including after the Q2 print; we expect a similar pattern through Q4 and into the January 2026 FY26 guide print.

The single small bear signal worth monitoring is Wei's first-time (in four quarters) acknowledgment of tariff risk in "consumer and price-sensitive" segments. HPC, flagship smartphones, and AI accelerators — the earnings drivers — were not flagged. The valuation has moved from "extraordinary" in April to "fair" today, compressing the upside asymmetry but not breaking it. We maintain Outperform with a 12-month target range of $310-$340 base case and $375-$390 bull case.

What could change our mind:

• Downgrade to Hold: FY26 preliminary framework at January 2026 print that comes in below mid-teens USD growth; or Q4 2025 gross margin compression below 57% on FX or structural factors; or a Section 232 outcome that triggers 15%+ chip-specific tariffs with limited pass-through; or hyperscaler capex guides for 2026 that show sub-10% growth.
• Downgrade to Underperform: A forced Intel JV or foundry partnership mandated by the Trump administration; direct Taiwan-strait escalation with operational impact; or gross margin compression below 54% on sustained NT$ strength plus accelerating overseas dilution plus AI demand moderation.

Signposts for Q4 2025 / January 2026 earnings: