FY25 Closes With a 62.3% Margin Print and a $56B Capex Bet on 2028: The AI Foundry Thesis Is Now a Multi-Year Capacity Commitment — Maintain Outperform

Results vs. Consensus

Q4 2025 Scorecard

FY 2025 Full-Year Summary

FY25 Guide Cadence — The Story of a Compounding Beat

The compounding nature of the FY25 guide raises is itself the narrative. Two mid-year upward revisions in a single fiscal year is unprecedented at TSMC, and the actual delivery landed at the top of the third (highest) guide. Management exited 2025 at +35.9% USD growth from a starting consensus of ~20%, meaning TSMC delivered approximately $14B of incremental revenue versus where the Street stood at the beginning of the year. For an industry that obsesses over book-to-bill ratios and 90-day visibility, this is a structural rather than cyclical signal.

Quality of Beat

• Revenue: The +1.8% Q4 revenue beat understates the strength of the underlying business. The monthly revenue disclosures meant the Street had visibility into the ~$33.7B trajectory by early January, so the headline beat was always going to be modest. The more important data is the +25.5% USD YoY growth on an already elevated base — the four-quarter sequence (Q1 +35%, Q2 +44%, Q3 +41%, Q4 +25%) shows growth normalizing to the high-20s in Q4 but only because FX translated less favorably (NT$ stable in Q4 vs. weaker in Q2). On a constant-currency basis, the underlying growth trajectory is consistent throughout the year.
• Margins: The 62.3% gross margin and 54.0% operating margin are the unambiguous story. Three converging factors: (1) cost improvements at N3 as the node has matured, (2) capacity utilization at or near peak across N3/N5/N7, (3) NT$/US$ stability around 30.6 (TSMC's guide assumption) versus the late-Q2 panic-spike to 29.3. CFO Huang explicitly cited cost improvements at N3 as a margin driver. Most importantly, N3 is finally crossing the corporate gross margin average in 2026 — ending its two-year run as a margin-dilutive ramp node and adding tailwind to the structural margin profile.
• EPS: NT$19.50 represents +35.0% YoY EPS growth on +25.5% revenue growth — a ratio of 1.37x, consistent with the operating leverage seen in Q1-Q3. ADR EPS of $3.14 beat consensus of $2.98 by $0.16 (+5.4%). The beat was overwhelmingly operational. Net income of NT$505.74B (+35.0% YoY) and net profit margin of 48.3% are levels you would expect from a pure-play IP licensing company, not a capital-intensive foundry running $40+ billion in annual capex.
• FY25 in aggregate: The full-year delivery is the cleanest validation possible of the initiating thesis. Revenue +35.9%, EPS +46.4%, gross margin +380bps to 59.9%, free cash flow crossing NT$1 trillion for the first time, capex +37% to $40.9B, and the foundry industry grown by more than 2x. Every line item exceeded the Q1 2025 framework when we initiated the position.

Segment Performance

Q4 saw a continuation of the Q3 pattern: HPC absolute dollars grew but share dipped slightly because smartphone volumes ramped through the seasonal Apple A19 cycle and broader holiday demand. The full-year picture is the more important read — HPC at 58% of FY25 revenue, up +48% YoY, with smartphone, IoT, and automotive all delivering double-digit growth on a base that finally absorbed the long auto inventory digestion.

HPC: 58% of the Year, +48% Growth

The full-year HPC numbers are extraordinary. On a $122.4B revenue base, HPC contributed approximately $71B of FY25 revenue — up roughly $23B from FY24's ~$48B and accounting for nearly the entirety of TSMC's incremental revenue. This is the second consecutive year that HPC has been responsible for essentially all of TSMC's top-line growth. The 4% sequential Q4 growth in HPC came on top of a record Q3 base, making it the highest quarterly HPC dollar print in TSMC's history (roughly $18.5B in a single quarter).

"Looking ahead, we observe increasing AI model adoption across consumer, enterprise and sovereign AI segments... our conviction in the multiyear AI megatrend remains strong, and we believe the demand for semiconductors will continue to be very fundamental." — C.C. Wei, Chairman & CEO

The expanded customer engagement across consumer, enterprise, and sovereign AI is a meaningful framing change. In Q1 2025, the HPC narrative was almost entirely about hyperscaler training infrastructure (Nvidia Blackwell, hyperscaler ASICs). Wei's Q4 commentary explicitly broadens this to inference (consumer / on-device AI), enterprise deployments, and sovereign AI projects (which are typically multi-year multi-government contracts). This three-pillar structure for AI demand reduces the cyclicality of the demand base and supports the sustained capacity-constrained dynamic that has held since 2024.

Assessment: HPC is no longer a segment within TSMC — it is the dominant business model. With 58% share at the company level and 70-80% of FY26 capex going into the nodes that serve HPC (≤7nm), TSMC's cost structure is increasingly optimized for AI accelerator production. Wei's customer-engagement window of "two or three years before production" implies HPC visibility extends into 2028 already. We continue to see 2026-2027 HPC growth in the +25-30% range with continued capacity-constrained dynamics.

Smartphone: The Quiet Surprise of FY25

Smartphone grew +11% in FY25 and Q4 sequentially +11% — both well above the low-single-digit growth that consensus had baked in at the start of the year. The N3 node continues to be filled by Apple A19 series and select flagship Android demand. Wei's Q4 commentary specifically called out high-end smartphone resilience to memory price inflation: "TSMC's high-end smartphone customers are less sensitive to memory price rises." This is the right call — flagship smartphones operate with materially higher gross dollar margins, where memory cost is a smaller share of BOM and price-elasticity is lower.

Assessment: Smartphone +11% on a $122B base is meaningful incremental dollars. The mix-shift toward N3 and toward higher-ASP advanced-process content per phone is a structural support for smartphone revenue even as unit growth is modest. The memory-shortage narrative will be a 2026 watch item but management's framing — high-end resilience — aligns with what Apple's premium positioning supports.

Automotive: The Recovery That Materialized

Automotive grew +34% in FY 2025, the cleanest reversal of any segment. After three quarters of "bottoming, not breaking" through 2024 and H1 2025, auto inventory finally cleared and customers re-engaged. The 18% sequential Q3 print was followed by Q4 stability, suggesting the recovery is durable into 2026 rather than a one-quarter snapback. Auto remains 5% of mix but is now growing materially faster than the corporate average for the first time in two years.

Assessment: Auto is a genuine 2026 tailwind. EV silicon content per vehicle continues to rise, and TSMC's mature-node and selected advanced-node capacity is benefiting from BMW, Tesla, NIO, and Chinese EV-maker ramp programs. Watch H1 2026 commentary for whether auto re-engages further or stabilizes; either way, auto stops being a swing factor and becomes a steady contributor.

IoT: Edge AI Beginning to Lift

IoT grew +15% in FY 2025, including a 20% sequential jump in Q3. Edge AI inference chips and industrial automation are the primary drivers. As more devices embed AI inference capability (consumer cameras, security systems, industrial sensors, automotive ADAS), IoT demand for advanced and specialty silicon is structurally rising.

Assessment: IoT is too small (5% of mix) to move the model materially, but its growth rate matching HPC's underlying organic rate is itself meaningful — it shows AI demand is broadening beyond hyperscale into edge applications. This is a multi-year tailwind that should compound in 2026-2028.

DCE: Immaterial and Continuing to Decline

Display and consumer electronics fell 22% sequentially to 1% of revenue. This is now an essentially immaterial segment. The decline reflects continued mature-node weakness in display drivers and consumer-grade legacy products. Not a thesis issue.

Technology Node Mix — The Advanced Node Tightening

The 5-percentage-point jump in N3 share from Q3 to Q4 (23% to 28%) is the largest single-quarter advance in N3 share since the node launched. With the Apple A19 cycle absorbing premium N3 capacity and the HPC customer base continuing to migrate to the N3 node, N3 is now scaled beyond its initial dilution phase. Management explicitly stated N3 will cross the corporate gross margin average in FY 2026 — the structural significance of which is that for the first time, TSMC will have three high-volume nodes (N7, N5, N3) all running at corporate-average or accretive gross margin. The advanced-node tightening across all three is the structural margin engine for 2026.

N2 entered high-volume manufacturing in Q4 2025 at both the Hsinchu and Kaohsiung sites, with management characterizing yields as "good." Critically, Investor Relations Director Jeff Su disclosed that TSMC will stage a "very fast ramp" of N2 in 2026 and expects it to deliver more revenue, faster, than the equivalent N3 effort — a meaningful acceleration vs. the N3 ramp pattern. N2 wafers are priced approximately 50% higher than N3, which means the dollar contribution of even modest N2 share is material. The 2-3pp gross margin dilution from N2 ramp in 2026 is real but already incorporated in the Q1 2026 guide of 63-65%, which still represents an expansion from Q4's record level.

A16 (with backside power delivery / Super Power Rail) remains on track for H2 2026 volume production for HPC applications with complex signal routing and dense power delivery. The combination of N2 + N2P + A16 collectively forms what Wei now describes as another "large and long-lasting node family," supporting margin tailwinds through 2027-2028.

Key Topics & Management Commentary

Overall Management Tone: The arc from initiation through Q4 has been linear in confidence trajectory. Q1's defensive posture ("we have not seen any change in customer behavior") gave way to Q2's offensive raise ("approximately 30%"), then Q3's almost triumphant "stronger than three months ago" framing. Q4 represents a different kind of confidence — structural and committed. The defining act is not just the revenue or margin beat but the $52-56B 2026 capex commitment, which Wei explicitly linked to customer verification: "If we don't do it carefully, that'd be a big disaster." This is a CEO who is willing to put $50B+ behind his customer signal because the signal is verified through 2-3 years of forward orders. The tone is no longer "we believe AI demand is strong" — it is "we have already deployed capital to serve the demand we know is coming." Tariffs and memory shortage got first-time material acknowledgments but were framed as risks to consumer/price-sensitive segments, not to HPC.

1. The $52-56B Capex Commitment: A Statement, Not a Forecast

The 2026 capex guide of $52-56 billion is the single most important data point in this release. The math: 2024 capex was $29.8B, 2025 was $40.9B, and 2026 is now guided to $52-56B — a roughly 30% YoY step-up after a 37% YoY step-up in 2025. Past three years total: ~$101 billion. Next three years: "significantly higher," per CFO Huang. This is the largest single-year semiconductor capital commitment in industry history.

"If we don't do it carefully, that'd be a big disaster." — C.C. Wei, Chairman & CEO (on capex justification)

"This year's planned capex is expected to contribute little to 2026 output and only marginally to 2027." — C.C. Wei, paraphrased; capex sized for 2028+ output

The most important framing detail: Wei explicitly stated that the 2026 capex contributes "little to 2026 output and only marginally to 2027." New high-volume fabs require 2-3 years to construct and ramp, meaning the $52-56B being deployed in 2026 is fundamentally about 2028 output. This is not a quarterly demand chase; it is a multi-year capacity commitment based on contracted demand visibility. Combined with the "two or three years before production" customer engagement framing, TSMC is effectively telegraphing that AI accelerator demand is contracted through 2028.

The capex allocation is informative: 70-80% goes to advanced nodes (≤7nm), with the balance split between advanced packaging (CoWoS / SoIC), mask-making, and other support investments. The advanced-packaging share specifically ties capacity expansion to AI accelerator production, which depends on TSMC's CoWoS capacity for HBM stacking. Advanced packaging revenue at 8% of FY25 revenue is guided above 10% in FY26 — meaning the dollars committed to advanced packaging capex have a direct, near-term revenue line of sight.

Assessment: The capex commitment is the de-risking event for the 2027-2028 thesis. Sell-side concerns about post-Blackwell AI cycle fade are materially weakened by a CEO willing to spend $52-56B on capacity that won't produce until 2028. The near-term cost is rising depreciation (high-teens % YoY in 2026), but the margin guide of 63-65% in Q1 2026 absorbs that and still expands. The capex commitment is also a competitive moat: at $52-56B in a single year, no peer can replicate this scale of advanced-node capacity buildout. Samsung's foundry capex is roughly $15-20B annually, Intel's is split across foundry and product, and even SMIC operates at a fraction of this scale on legacy nodes.

2. Margin Trajectory: 63-65% Guide Above Q4's Record

The Q1 2026 gross margin guide of 63-65% is the second most important data point in this release. At Q3, the bear narrative was "margins peak at 58%." We argued in the Q3 recap that this was wrong, and Q4 vindicated that view with a 62.3% print. At Q4, the bear narrative shifted to "62% is the cycle peak, dilution from N2 + overseas + tariffs will compress 2026." Q1 2026's guide of 63-65% structurally rebuts that — management is guiding higher despite known headwinds.

"Gross margin increased 2.8 percentage points sequentially to 62.3%, primarily due to cost improvements at N3, higher capacity utilization, and favorable FX translation." — Wendell Huang, CFO (paraphrased from Q4 commentary)

The drivers are mechanical and durable: (1) N3 yield maturation continuing — cost per acceptable wafer falling as defect density improves; (2) capacity utilization at or near peak across N3/N5/N7; (3) FX stable around the 30.6 NT$/USD assumption; (4) advanced packaging (10-20% of capex) commanding premium pricing; (5) wafer ASP increases continuing per CFO Huang's "such price rises will continue going forward" commentary.

The headwinds are equally specific but smaller in 2026: N2 ramp dilution at 2-3pp (offset by N3 finally turning margin-accretive); overseas fab dilution maintained at 2-3pp early stage and 3-4pp later years (tracking expectations from prior quarters); rising depreciation as the 2024-2025 capex cohort begins flowing through cost; effective tax rate ticking up from 16% to 17-18%. None of these is a surprise; all are absorbed in the 63-65% Q1 2026 guide.

Assessment: The structural gross margin profile of TSMC is settling at a higher cycle level than even bulls had modeled at the start of 2025. A 60%+ gross margin at this scale, with 70-80% of capex going to higher-ASP advanced nodes and continued wafer ASP increases, suggests through-cycle gross margin lands at 58-60% (versus the long-term ≥56% management target). The dollar implication is meaningful: every 100bps of gross margin at $160B FY26 revenue equals $1.6B of incremental gross profit that drops largely to the bottom line.

3. AI Demand Expansion: Three Pillars Now

Wei's framing of AI demand has structurally evolved over the year. In Q1, the AI narrative was hyperscaler training infrastructure. In Q2, it was hyperscaler training + early ASIC waves. In Q3, Wei added "tokens exponentially increasing" and "customers' customers" as direct demand signals. In Q4, Wei explicitly identified three distinct AI demand pillars: consumer AI, enterprise AI, and sovereign AI. Each represents a different demand cycle with different elasticity properties.

"Looking ahead, we observe increasing AI model adoption across consumer, enterprise and sovereign AI segments... our conviction in the multiyear AI megatrend remains strong." — C.C. Wei, Chairman & CEO

"After extensive discussions with customers and their customers, [I am] confident in the AI demand's sustainability. Customers have shown evidence of AI's positive impact on their business." — C.C. Wei (paraphrased Q&A)

The "evidence of AI's positive impact on their business" framing is the strongest claim Wei has made in any earnings call we have covered. It is not a forecast; it is an observation about already-realized customer ROI on AI deployments. The implication is that hyperscaler capex is no longer purely speculative buildout — it is being pulled by demonstrated returns at the customer level. This dramatically strengthens the through-cycle AI demand narrative.

The AI accelerator revenue specifics: high-teens % of total wafer revenue in 2025 = $20-22B in FY25 alone. Management projects this growing at a mid-to-high 30s percent CAGR through 2029 (some sources cite higher; we use the conservative end). On a 2025 base of $20-22B, a 35% CAGR through 2029 implies AI accelerator revenue reaching $70-80B by 2029 — over half of TSMC's projected $200-250B revenue by then.

Assessment: The three-pillar AI framing is the most significant strategic update in this release. It reduces concentration risk relative to a hyperscaler-only narrative and supports the 2027-2028 visibility that the capex commitment requires. Sovereign AI specifically is a new revenue category that management did not flag at all in Q1 — it has now become a recognized demand driver, which speaks to how rapidly the customer landscape is evolving.

4. N2 Ramp: Ahead of Schedule with "Good Yields"

N2 entered high-volume manufacturing in Q4 2025 at both the Hsinchu and Kaohsiung sites with yields characterized by management as "good." This is a meaningful timing acceleration: at Q1 2025, the consensus expectation was for N2 volume production to begin in H2 2025 with limited Q4 contribution. Q4 actual delivery represents the high end of what was modeled.

"Two TSMC plants are already producing 2nm parts with good yield." — C.C. Wei, Chairman & CEO

"TSMC will stage a 'very fast ramp' of its two-nanometer process in 2026, and that it will deliver more revenue, faster than its three-nanometer efforts." — Jeff Su, Director of Investor Relations

The "very fast ramp" framing combined with N2 ASPs ~50% above N3 has material near-term implications. If N2 reaches 5-7% of wafer revenue by end-2026 (consistent with Su's "faster than N3" pace), the dollar contribution would be approximately $8-12B in FY26 alone — a meaningful new revenue stream from a node that wasn't producing volume a year ago. The 2-3pp gross margin dilution from N2 ramp is the cost of admission, but the dollar contribution and forward margin profile (N2 will become accretive within 18-24 months based on historical node-cycle patterns) is the asymmetric reward.

Assessment: N2 is one of TSMC's safest bull-case data points. Samsung's 3nm yields remain commercially uncompetitive; Intel 18A is producing in volume but with limited customer commitments; N2 is in volume production with multiple customers (Apple, AMD, Nvidia, MediaTek confirmed across the 12-month roadmap). The competitive landscape at N2 remains a one-horse race, and the ASP economics support both revenue growth and a long-term margin tailwind. A16 (H2 2026) extends this for HPC-specific applications.

5. The 2026 Wafer ASP Conversation: Pricing Power Is the Quiet Tailwind

An analyst question on the Q4 call referenced a 20% wafer ASP increase that has been speculated about for 2026. CFO Huang's response was unusually direct: such price rises "will continue going forward." This is the most explicit pricing-power statement TSMC has made in years.

"Each new generation manufacturing process requires higher capital expenditure than its predecessor, so prices for finished products must rise." — Wendell Huang, CFO

The economic logic is straightforward and increasingly visible. Capex per wafer has approximately doubled from 5nm to 3nm and will roughly double again from 3nm to 2nm. Tool costs, mask costs, process complexity, and yield-development overhead all scale exponentially with node advancement. Customers face a stark choice: pay materially more per wafer for advanced nodes, or accept architectural disadvantage in their end markets. For the AI accelerator, smartphone, and HPC customers that drive 90%+ of TSMC's revenue, paying for advanced nodes is the only viable strategy.

Assessment: Wafer ASP is the quiet structural tailwind that the bears systematically underweight. With N2 wafers priced 50% above N3, every quarter of N2 share growth in 2026-2027 mathematically expands corporate ASP. The 20% wafer ASP increase narrative is consistent with this trajectory and supports both revenue growth and gross margin expansion. Management's "such price rises will continue going forward" commentary is rare explicit pricing guidance.

6. Arizona: From Hedge to Strategic Asset

The Arizona expansion has fully transitioned from "expensive geopolitical hedge" to "strategic capacity asset." The most recent disclosure is a $197M land parcel purchase in Arizona — a second large parcel signaling plans for what the company is internally describing as a Gigafab cluster. Combined with the prior $165B commitment announced March 2025, the Arizona footprint is approaching scale where it can absorb 30%+ of TSMC's N2 capacity by late-decade.

The Arizona roadmap as updated: Fab 1 Phase 1 (N4) in HVM since Q4 2024; Fab 2 construction complete with tool move-in 2026 and HVM in H2 2027; Fab 3 under construction; Fab 4 in permitting; first US advanced packaging facility in permitting. The pace of construction has accelerated meaningfully over the past 12 months.

On margins, the FY25 overseas dilution was revised down from 2-3% (Q1/Q2) to 1-2% (Q3) and FY25 actuals tracked the lower revised range. The long-term framework (2-3% early stage, 3-4% later years) was maintained, but the near-term execution is meaningfully better than initially modeled.

Assessment: Arizona is no longer a thesis risk. The combination of yield parity with Taiwan, accelerating fab construction, and political insulation under the Trump administration has converted what was a 2024 bear point into a 2025-2028 capacity asset. The next watch item is whether N2 wafer production in Arizona Fab 2 (HVM H2 2027) achieves the same yield parity that N4 has at Fab 1.

7. Tariffs and Memory Shortage: First Material Risk Acknowledgments

For the first time across four prints in 2025, Wei meaningfully acknowledged two macro risks: tariff policies and memory price inflation. The framing was disciplined — both were flagged as risks to "consumer-related and price-sensitive end-market segments" rather than HPC or flagship smartphone. But the explicit acknowledgment is a small tonal shift worth noting.

"Potential changes in tariff policies, rising component costs — especially memory price inflation — and softness in price-sensitive consumer segments could weigh on unit demand in PCs and smartphones." — C.C. Wei, Chairman & CEO

"TSMC's high-end smartphone customers are less sensitive to memory price rises." — C.C. Wei, Chairman & CEO

The memory shortage is real and well-documented — HBM and select DRAM/NAND categories have been in tight supply since H2 2025 as memory makers struggle to add capacity at pace. For TSMC, the second-order effect would be reduced unit demand for memory-paired SoCs (PCs, low-end smartphones) if memory prices rise enough to compress system pricing. Wei's "high-end smartphone customers are less sensitive" point is the right empirical observation: Apple, Samsung's flagship, and high-end Android pricing leaves room to absorb $20-50 of incremental memory cost per device.

The tariff acknowledgment continues to be cautious rather than alarming. Section 232 investigation remains technically open, and the 2026 election cycle may re-activate policy brinksmanship. But Wei's framing — tariffs as a "risk" rather than a "current factor" — suggests no current order book impact.

Assessment: These are real risks but they are smaller in magnitude than the bear cases imply. Memory shortage compresses unit demand at the consumer edge but the consumer edge is ~20-25% of TSMC's revenue (smartphone + IoT + auto + DCE excluding HPC). A 5% unit decline in those segments equates to ~1pp of revenue at most — absorbed easily by HPC outperformance. Tariffs remain unquantifiable but Wei's posture suggests no current impact. We continue to view both as monitor items, not thesis-changing.

8. Foundry 2.0 and Industry Share Gain: 30% vs. 14%

Management disclosed that TSMC expects the Foundry 2.0 industry to grow approximately 14% in 2026 — while TSMC itself expects to grow ~30% in USD terms. The implication is that TSMC plans to take roughly 16 percentage points of industry share for the second consecutive year. In FY25, TSMC grew 35.9% USD vs. industry +17% — approximately 19 percentage points of share gain.

The Foundry 2.0 framework Wei introduced at Q3 has now been integrated into management's growth forecasting language. Foundry 2.0 includes wafer manufacturing, advanced packaging, mask-making, testing, and system-level integration — effectively a broader value chain than traditional pure-play foundry. The growth differential between TSMC (30%) and the industry (14%) reflects: (1) TSMC's dominant position at the leading edge where customer concentration is highest; (2) advanced packaging revenue being TSMC-led (CoWoS); (3) the inability of Samsung or Intel Foundry to capture leading-edge customers at competitive yield; (4) wafer ASP increases benefiting TSMC disproportionately as the price-leader.

Assessment: Two consecutive years of 15-20pp share gain is structural rather than cyclical. The competitive landscape is unlikely to materially change before 2027 at earliest, and the customer engagement model (2-3 years pre-production) means 2026-2028 share is largely locked in. The mathematical implication: by 2027, TSMC's share of leading-edge foundry production will be approaching 95% — effectively a monopoly at advanced nodes.

9. Capital Return: Steady Dividend Growth, No Buyback Surprise

Management committed to a 2026 dividend of NT$23/share with explicit intent to raise dividends steadily over time. With FY25 free cash flow crossing NT$1 trillion for the first time and year-end cash & marketable securities of NT$3.1T (~$98B), TSMC has structural cash generation capacity well above current capital return levels. No expanded buyback was mentioned.

The dividend math: at NT$23/share annualized and a current ADR price in the $300-330 range, the dividend yield is approximately 0.7-0.8% — modest by US large-cap standards but reasonable for a capex-intensive growth company. Given $52-56B of 2026 capex deployment, retaining cash to fund advanced-node expansion is the right capital allocation.

Assessment: Capital return remains a legitimate frustration for value-oriented investors but is correctly subordinated to the growth investment cycle. With cash generation accelerating and capex visibility through 2028, there is room for more aggressive dividend growth in 2027-2029. We would expect the dividend to compound at 15-20% annually through this period, with the structural ability to support a meaningful buyback if growth opportunities slow — an option-value worth noting.

Guidance & Outlook

The guide cadence speaks for itself. TSMC just delivered the highest gross margin in its history (62.3%) and is guiding the next quarter to 63-65% — expanding from a record level despite known dilution headwinds. The Q1 2026 revenue guide of $34.6-35.8B represents +38% YoY at the midpoint, which is an acceleration from Q4's +25.5% YoY pace. The full-year 2026 framework of ~30% USD growth implies revenue approaching $160B, compared to FY25's $122.4B.

Implied FY26 quarterly cadence: If Q1 2026 lands at the midpoint of $35.2B, the implied Q2-Q4 average is approximately $41-42B per quarter to hit a $158B FY26 midpoint. This is plausible but aggressive — it would require continued sequential expansion through the year. More likely, FY26 lands at $155-158B with Q1 at $35.2B and a moderate sequential ramp through the year, with seasonal smartphone strength in Q3-Q4 providing the seasonal boost. Either path is consistent with the ~30% USD growth framework.

Implied FY26 EPS: At ~30% USD revenue growth from a $122.4B base, FY26 revenue lands at ~$159B. At a 60% gross margin (versus Q1 guide of 63-65% midpoint of 64%), gross profit is ~$95B. Operating margin at 50% (Q1 guide midpoint 55% but conservatively assuming some moderation) implies operating income of ~$80B. After 17-18% tax (assume 17.5%) and minor net interest/other, net income lands at ~$66B. With ~25.9B shares outstanding (post-split), this implies FY26 EPS of approximately $2.55 (NTD ~NT$80) on a per-share basis. Versus FY25 EPS of $2.11 (NT$66.25), this represents +21% EPS growth. Note: ADR EPS is per-share, with each ADR equal to 5 ordinary shares; actual ADR EPS for FY25 was $3.14 in Q4 alone, suggesting full-year ADR EPS of ~$10-11.

Capex framework: The 2026 capex of $52-56B is the headline number. Past three years (2023-2025) total capex was ~$101B. CFO Huang's commentary that the next three years (2026-2028) will be "significantly higher" implies cumulative 2026-2028 capex of ~$160-200B+. Combined with Wei's commentary that 2026 capex contributes "little to 2026 output and only marginally to 2027," this is a signal that 2027-2028 capex will likely be in the $60-70B range each year — supporting 2028+ output.

Guidance style: TSMC has now beaten the high end of its own gross margin guide in eight consecutive quarters. The Q1 2026 guide of 63-65% is therefore likely conservative — we model actual delivery at 64-65% and expect quarterly upside of 50-150bps versus guide throughout 2026. The revenue guide structure has been tighter (typically beating midpoint by 1-3%) but consistently above-midpoint. We model FY26 USD revenue growth at 30-32% on the upside (vs. ~30% guide).

Analyst Q&A Highlights

AI Demand Sustainability and Capex Justification

• Multiple analysts: Asked whether the $52-56B capex commitment is supported by sufficiently durable demand or risks oversupply if AI demand moderates. Wei was unambiguous — the capex hike comes after rigorous verification with customers; customers engage 2-3 years pre-production; AI demand sustainability is supported by demonstrated customer ROI ("evidence of AI's positive impact on their business").
  Assessment: This is the most direct customer-verification statement Wei has made in any earnings call. The "evidence of positive impact" phrasing implies hyperscaler ROI on AI deployments is being measured and is positive, which fundamentally rebuts the "AI capex is speculative buildout" bear narrative.

2026 AI Accelerator Trajectory

• Bernstein / Morgan Stanley: Pressed Wei on specific 2026 AI accelerator growth rate given the high 2025 base. Wei confirmed a 25% company CAGR through 2029 and characterized AI accelerator revenue as growing at "mid-to-high 30s percent CAGR" through the period (some sources cite even higher 50s%). 2026 specifically expected to be "another strong growth year."
  Assessment: A 35%+ AI accelerator CAGR through 2029 implies AI accelerator revenue of $70-80B+ by end-decade — over half of TSMC's projected total revenue. This is more bullish than the Q3 commentary on 2026.

Margin Trajectory and Long-Term Profitability

• Goldman Sachs / JPMorgan: Asked about the 63-65% gross margin guide and whether 60%+ is the new structural level. Huang attributed the elevated margin to N3 cost improvements, capacity utilization, and stable FX. Long-term gross margin target maintained at ≥56%, but Q1 2026 guide implies through-cycle landing zone is materially higher.
  Assessment: This is the most important answer in Q&A. If Q1 2026 GM lands at 64-65% (midpoint of guide), FY26 consensus gross margin and EPS estimates need to move 10-15% higher. Several sell-side firms appear to be in the process of doing this within 24-48 hours of the print.

2026 Capex Magnitude

• Citi / Barclays: Asked Huang to reconcile the $52-56B 2026 capex (vs. $40.9B in 2025) with 2027-2028 capacity needs. Huang said past three years totaled $101B, next three years will be "significantly higher" without specific guide. 70-80% allocated to advanced nodes; 10-20% to advanced packaging and mask-making.
  Assessment: The implication is 2027-2028 capex will likely each exceed $60B. Combined with Wei's commentary that 2026 capex contributes "little to 2026 output and only marginally to 2027," this telegraphs an enormous 2028 supply ramp. Either AI demand stays strong through 2028 (likely, per customer engagement framework) or TSMC will face material capex underabsorption from 2028 onwards (the bear case).

Wafer ASP and Pricing Power

• Multiple analysts: Asked about 20% wafer ASP increase reports for 2026 and whether TSMC could comment. Huang's response: "Such price rises will continue going forward." Each new generation requires higher capex per wafer, supporting necessary price increases. Wafer ASP "is not the main profit driver" — utilization and mix matter equally.
  Assessment: Direct pricing-power commentary is rare. Combined with N2 wafers priced ~50% higher than N3, this supports the structural ASP expansion narrative through 2026-2028. Pricing power is the underrated structural tailwind.

N2 Ramp Speed and Yield

• UBS / TD Cowen: Asked about N2 yield specifics and ramp pace. Wei said two TSMC plants are producing 2nm parts with "good yield." Jeff Su (IR Director) added that N2 will see a "very fast ramp" in 2026 and deliver more revenue, faster, than the equivalent N3 effort.
  Assessment: N2 is 1-2 quarters ahead of where the Street had modeled at the start of 2025. The "faster than N3" framing is concrete and quantitatively meaningful — N2 share could reach 5-7% of wafer revenue by late-2026 vs. consensus assumptions of 3-5%. The ASP uplift (50% above N3) makes this a material 2026 revenue contributor.

Arizona Capacity Acceleration

• JPMorgan / Citi: Asked about the pace of Arizona capacity ramp and the recent $197M land parcel purchase. Wei said Arizona expansion is being accelerated due to AI demand. Fab 1 in HVM since Q4 2024; Fab 2 tool move-in 2026 with HVM H2 2027; Fab 3 under construction; Fab 4 + first US advanced packaging facility in permitting.
  Assessment: Arizona has fully shifted from "geopolitical hedge" to "strategic capacity asset." The Fab 4 + advanced packaging permits suggest TSMC is committing to a US Gigafab cluster on the same scale as a major Taiwan campus. This both supports US tariff insulation and provides multi-year US capacity for hyperscaler customers.

Memory Shortage and Tariff Risks

• Needham / Mizuho: Asked about memory shortage impact on smartphone/PC demand and Wei's tariff acknowledgment. Wei clarified: high-end smartphones less sensitive to memory prices; tariffs flagged as a 2026 risk for consumer/price-sensitive segments only; HPC and flagship demand are not affected.
  Assessment: First material tariff acknowledgment Wei has made in four prints, but framing is contained to 5-10% of revenue (low-end consumer). Worth monitoring; not thesis-breaking. Memory shortage is a real second-order risk but absorbed by high-end customer mix.

Power Supply for AI Data Centers

• Goldman Sachs: Asked how TSMC views data center power constraints. Wei said TSMC focuses on Taiwan electricity for its own operations; for global AI data centers, customers have planned power supply for years. The bottleneck is silicon, not power.
  Assessment: This is an important clarification. The "power is the bottleneck" narrative that has dominated AI-bear commentary in late 2025 is being explicitly contested by TSMC's CEO. Customers (hyperscalers) have been planning power supply at multi-year horizon and continue to commit to silicon. Reduces the credibility of the "AI capex constrained by grid capacity" bear thesis.

Foundry 2.0 Industry Growth

• Mizuho: Asked Wei about the Foundry 2.0 framework and 2026 industry growth. Wei: industry expected to grow 14% in 2026; TSMC expects ~30% growth — continuing to take share for second consecutive year. Capacity planning uses "top-down and bottom-up" approach — TSMC builds to contracted demand, not speculative forecasts.
  Assessment: Two consecutive years of 15-20pp share gain is structural. Combined with the moat at advanced nodes and CoWoS, TSMC's leading-edge foundry share should approach 95% by 2027 — effectively a monopoly at the leading edge.

What They're NOT Saying

1. Specific 2026 quarterly cadence beyond Q1: The Q1 2026 guide is specific ($34.6-35.8B) but no Q2/Q3/Q4 framework was given. With the FY26 guide at ~30% growth and Q1 already at $35.2B midpoint, Q2-Q4 needs to average $41-42B per quarter to hit the FY26 midpoint. Management did not address this implied seasonal acceleration explicitly. We model a more moderate Q2-Q4 trajectory averaging $40B per quarter, implying FY26 lands at $155-158B (slightly below management's $159B implied midpoint).
2. 2027-2028 specific capex framework: Huang said "significantly higher" but declined to specify magnitude. Given the Wei commentary that 2026 capex is for 2028 output and that capacity expansion will "remain elevated as long as the AI-driven demand cycle continues," 2027-2028 capex is likely $60-70B each year — a number management is not yet ready to formalize.
3. CoWoS 2026 specific capacity targets: Wei acknowledged CoWoS expansion but did not provide specific capacity numbers. CoWoS doubled in 2025 from 2024 and continues to be tight; the implied 2026 capacity expansion is likely 30-50% based on the customer mix and capex allocation, but management would not commit to specifics.
4. HPC absolute revenue and AI accelerator specifics: Management does not break out HPC revenue dollars or AI accelerator revenue specifically. We estimate HPC at ~$71B FY25 and AI accelerator at $20-22B FY25 from inferred share figures, but management's reluctance to provide concrete numbers leaves uncertainty around the underlying growth dynamics.
5. Specific Section 232 scenario planning: Tariffs were acknowledged as a risk but no scenario analysis or impact range provided. The investigation remains technically open; management continues to deflect specific scenario questions. The 2026 election cycle could re-activate policy brinksmanship; absence of scenario commentary leaves analysts to infer.
6. Buyback authorization or expanded capital return: With cash reserves of NT$3.1T (~$98B) and free cash flow exceeding NT$1T, TSMC has structural capacity for a meaningful share repurchase. No buyback was mentioned. Dividend commitment of NT$23/share is incremental but modest. Capital return remains a legitimate frustration relative to peer averages.
7. Pricing actions and ASP by node: CFO Huang's "such price rises will continue" was directionally clear but no specific N2/N3/N5 ASP framework provided. With wafer ASPs being a major margin driver, the absence of explicit pricing visibility forces analysts to model on inference.
8. China revenue trajectory: China share continues to decline (~9% of revenue in Q3-Q4) under export controls. Management does not discuss China-specific trends. Given the structural decline of China exposure, the "TSMC as China-risk stock" narrative is increasingly stale — but management's silence leaves the topic unaddressed for investors who continue to focus on it.
9. Specific N2 ASP and yield numbers: Wei said N2 yields are "good"; Su said N2 ramp will be faster than N3. Neither quantified ASP or yield specifically. With N2 economics being a major 2026-2027 margin driver, more specificity would have been helpful.
10. Specific competitive positioning vs. Samsung Foundry / Intel 18A: Wei maintains the neutral-supplier posture and will not comment directly on competitors. With Samsung's 3nm yields still struggling commercially and Intel 18A in production but with limited customer commitments, TSMC's competitive position at leading edge is the strongest it has been in years — but management will not say so directly.

Market Reaction

• Pre-earnings ADR (Jan 14, 2026 close): ~$300-310 area (estimated from Q3 close of $295 plus modest November-December rally and early-January stability)
• Pre-market Jan 15 reaction: Positive — "TSMC's Strong Earnings Sparked a Tech Stock Rally Thursday" (Investopedia)
• Earnings day intraday (Jan 15): Stock estimated to gain +3-5% with semi-equipment names (ASML, AMAT, KLA, LRCX) and AI-adjacent names (NVDA, AMD, AVGO, MU) rallying in sympathy
• Estimated Jan 15 close: $310-325 area
• T+1 (Jan 16, report date): Modest positive follow-through; estimated Jan 16 close $315-330 area
• Volume: ~1.5-2x average daily volume on Jan 15
• Analyst actions within 24-48 hours:
  • Needham: Reiterated Buy; PT $410 (Charles Shi)
  • Barclays: Reiterated Overweight; PT $450 (Simon Coles)
  • TD Cowen: Continued Outperform / Buy; PT $370 (Krish Sankar)
  • Bernstein: PT $330 (pre-print baseline; revisions in motion)
  • Median sell-side PT post-print: $410 (per QuiverQuant aggregation)
  • Range: $330 (low) to $450 (high)
  • FY26 EPS estimates: +8-12% upward revisions clustering across the sell side
  • FY26 gross margin assumptions: Moving from 56-58% (pre-print) to 60-62% (post-print)
  • FY26 revenue estimates: Moving toward $155-160B from $140-145B pre-print
  • No downgrades reported within 48 hours

The +3-5% reaction is constructive but not extreme — consistent with the pattern we've seen across TSMC prints in 2025. Three dynamics explain the magnitude: first, monthly revenue disclosures had pre-positioned much of the top-line beat; second, the stock had rallied substantially (~80-100% from April lows) heading into the print, leaving less room for explosive upside; third, the capex hike to $52-56B was the most attention-grabbing data point but it cuts both ways — it telegraphs management confidence in multi-year demand AND signals near-term margin pressure (depreciation rising high-teens % in 2026, N2 ramp dilution 2-3pp).

The more informative signal is the analyst community's reaction. Median sell-side PT now sits at $410 with the high at $450 (Barclays / DA Davidson), which compares to current trading in the $310-330 area — implying ~25-35% upside to the consensus PT. Historically, when prices dip or trade flat while analyst targets advance, the gap resolves in favor of fundamentals over 6-12 months. That mean-reversion has played out twice already (after Q2 and Q3 2025 prints). We expect a similar pattern through Q1-Q2 2026.

The Jan 15 pre-market commentary highlighted the $52-56B capex as the central narrative driver. App Economy Insights captured the interpretation: "If the AI era is an arms race, TSMC is the arsenal builder everyone depends on... You don't spend $56 billion on a hunch." This framing — capex as customer-verified demand commitment rather than speculative buildout — is the right read and supports the constructive multi-year setup.

Street Perspective

Debate: Is the $52-56B 2026 Capex Commitment Bullish or Concerning?

Bull view: The capex commitment is the most explicit customer-verified demand signal the semiconductor industry has ever received. Wei's "if we don't do it carefully, that'd be a big disaster" framing makes clear this was sized after rigorous customer engagement. The 70-80% allocation to advanced nodes and the explicit linkage to 2028+ output extend AI accelerator demand visibility through end-decade. No competitor can match this scale of capex; the moat widens.

Bear view: $52-56B in a single year is enormous, and the depreciation flow-through (rising high-teens % YoY in 2026) is a real near-term margin headwind. If AI capex moderates in 2027-2028, TSMC will face capex underabsorption that compresses gross margins toward the 56% long-term target. Past large capex cycles in the semiconductor industry have ended in oversupply; this could too.

Our take: Bull view wins decisively. The historical analogy to past semiconductor capex cycles is misleading because the customer engagement model is structurally different — TSMC builds to contracted demand from customers who engage 2-3 years pre-production. Combined with the three-pillar AI demand framework (consumer, enterprise, sovereign), the demand sustainability through 2028 is well-supported. The depreciation headwind is real but absorbed in the Q1 2026 guide of 63-65% gross margin. The capex commitment is a moat-widening event.

Debate: Is 60%+ Gross Margin Sustainable Through Cycle?

Bull view: Q4 GM at 62.3% and Q1 2026 guide at 63-65% codify the new structural margin level. N3 turning margin-accretive in 2026 + N2 ASP premium + advanced packaging mix + wafer ASP increases all support sustained 60%+ gross margins through 2027. The 2-3pp dilution from N2 ramp and 2-3pp from overseas fabs is absorbed by underlying mix and yield improvements.

Bear view: Long-term GM target remains ≥56%, suggesting management anticipates dilution returning to that range. Depreciation rising high-teens % YoY in 2026 + N2 ramp dilution + overseas fab dilution + tax rate ticking from 16% to 17-18% all compound. The 60%+ margin in 2025 is cycle-peak, supported by FX stability and N3 yield maturation timing — not a structural floor.

Our take: Bull view has stronger empirical support through 2026, with the bear case becoming more credible in 2027-2028 if N2 ramp dilution exceeds expectations or if overseas fabs scale aggressively. Through-cycle GM is likely 58-60% rather than the management's stated ≥56% — meaning consensus EPS estimates need to move structurally higher to reflect the elevated margin profile.

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

Bull view: CoWoS continues to expand and remain capacity-constrained; Wei explicitly cites multi-year AI megatrend conviction; three-pillar AI demand (consumer, enterprise, sovereign) reduces concentration risk; customer engagement window of 2-3 years pre-production provides forward visibility. AI accelerator growth of 25-30% in 2026 is achievable on a $20-22B 2025 base.

Bear view: 2025's roughly 2x growth in AI accelerator revenue is unrepeatable. Hyperscaler capex is at or near peak. Sovereign AI is overhyped and concentrated with specific governments. Inference workload expansion has been promised for two years but the enterprise scale-out has been slow. 2026 AI accelerator growth in the mid-teens is more realistic, and the deceleration triggers multiple compression.

Our take: Bulls win 2026; bears have a clearer case for 2027 if hyperscaler capex moderates. CoWoS capacity expansion combined with N2 ramp acceleration and the three-pillar demand framing supports 25%+ AI accelerator growth in 2026. The more interesting question is whether sovereign AI scales quickly enough to absorb potential hyperscaler training capex moderation in 2027 — that is a Q2-Q3 2026 watch item, not a current concern.

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

Bull view: At ~$315 ADR and FY26 EPS estimates moving toward $11-13, TSM trades at ~24-28x forward P/E. For a business growing revenue 30% and earnings 25-30% with 60%+ gross margins, widening process-technology leadership, and a $52-56B capex commitment de-risking 2027-2028, this multiple is below the sector average for AI beneficiaries (NVDA at 30-35x, AVGO at 30-35x). Sell-side median PT of $410 implies ~30% upside.

Bear view: The stock has more than doubled from the April 2025 lows ($134 → $315). At ~25-28x forward EPS, multiple expansion is largely done. Any FY26 disappointment (margin compression, AI demand moderation, tariff escalation) triggers sharp re-rating lower. Geopolitical tail risk is never zero. The asymmetry has shifted from extraordinary (April) to merely favorable (today).

Our take: Valuation is fair-to-attractive but not exceptional. Our base case is FY26 ADR EPS of $11-12, with 26-28x target multiple implying $290-340 price target range. Bull case (EPS $13, 28x = $365, +15-20% upside) and bear case (EPS $9.50, 22x = $210, -33% downside) frame the asymmetry at roughly 1.0-1.5x up-to-down. The risk/reward is less extreme than April or July 2025 but remains favorable for investors who can underwrite multi-year AI demand.

Debate: Does the Tariff / Memory Acknowledgment Signal a Cyclical Top?

Bull view: Wei's tariff and memory shortage acknowledgments are framed strictly to consumer/price-sensitive segments — explicitly excluding HPC and flagship smartphone. The HPC, advanced node, and AI accelerator drivers (representing 70%+ of earnings power) were not flagged as at-risk. Capex commitment of $52-56B is fundamentally inconsistent with a cyclical top. Wei would not commit to record capex if he saw demand softening.

Bear view: First material tariff acknowledgment in four prints is a tonal shift worth noting. The 2026 election cycle could re-activate policy brinksmanship. Memory shortage is a real second-order risk that could compress consumer device demand. Combined with an already-stretched valuation and cycle-peak margins, even small bear data points matter at the margin.

Our take: Small bear signals worth monitoring but not thesis-changing. The HPC/advanced node portion of the portfolio (70%+ of earnings power) was explicitly not flagged. Mature-node mix softness in tariff-sensitive categories would be a 30-60bps margin headwind at most, easily absorbed by HPC mix and N2 ASP uplift. Worth noting in the 2026 watch items; not worth repositioning for.

Model Update & Valuation Framework

Valuation framework: At a ~$315 ADR price, TSM trades at approximately 28-30x our FY25 EPS estimate of $10.80-11.00 and 22-24x our FY26 EPS estimate of $13.00-14.50. For a business growing revenue ~30% in 2026 and earnings 25-30%+ with 60%+ gross margins, widening process-technology leadership, and a multi-year capex-de-risked customer demand signal, this multiple is fair but not stretched. A 25-27x multiple on FY26 EPS of $13.50 implies a 12-month price target range of $338-365 — upside of 7-15% from current levels. A more bullish re-rating to 28-30x on $14 implies $390-420 — upside of 24-33%, consistent with the median sell-side PT of $410.

Bull-case scenario: AI accelerator revenue in 2026 grows 30%+; gross margin holds 62-64% as N3 fully turns accretive and N2 yields scale faster; FY26 revenue lands at $160-165B; 2027 capex commitment at $60-65B confirms 2028+ visibility. FY26 ADR EPS lands at $14.50-15.50. Target multiple of 28x implies $405-435 — upside of ~28-38%.

Bear-case scenario: 2026 AI accelerator growth moderates to 15-20% as hyperscaler capex flattens; Taiwan dollar appreciates meaningfully, pushing gross margins to 56-58% in H2 2026; N2 ramp dilution exceeds 3pp; tariff escalation cuts consumer/PC demand 5-10%. FY26 ADR EPS lands at $10-11; at a compressed 20x multiple, implies $200-220 — downside of ~30-37%.

Risk-reward calculus: Upside of ~25-30% (bull) versus downside of ~30-35% (bear) implies a roughly 0.85-1.0x asymmetry — less favorable than April 2025 (3:1) or July 2025 (1.5:1) but still reasonable for a stock at the center of a multi-year AI capex super-cycle. The capex commitment de-risks 2027-2028 fundamentals but does not eliminate cyclical risk if AI demand moderates faster than the customer engagement model suggests.

Through-cycle earnings power: Codifying the structurally higher margin profile and capex visibility, we now estimate TSMC's through-cycle EPS power at $11-12 (vs. our prior $9-10 estimate). This increases our intrinsic value range and supports a higher long-term price target trajectory.

Thesis Scorecard: Q3 Watch Items Revisited

Our Q3 2025 recap laid out ten specific signposts for Q4 2025 / January 2026. Every single one tripped favorably, with several exceeding the bullish threshold by material margins.

Thesis Scorecard: Full Bull / Bear Matrix Through FY25

FY25 Overall: Five primary bull points all confirmed or strengthened. Two new bull points added (multi-year capex de-risking 2027-2028; wafer ASP as structural tailwind). The Q2 FX bear point was definitively resolved. Two new bear points added at modest intensity (memory shortage, rising depreciation). The original tariff bear remains active but contained. Net-net, the FY25 thesis was vindicated comprehensively, and the FY26 outlook adds new confirming bull points while introducing only modest new headwinds.

The initiation thesis (April 2025) was four-for-four: AI demand structural — confirmed; Process-technology leadership widening — confirmed; Operating leverage dropping to bottom line — confirmed; Arizona as geopolitical hedge not margin wound — confirmed. We initiated at Outperform on a $155 ADR with the stock down 23% YTD on tariff fears. The stock has approximately doubled since while the operational thesis has strengthened across every dimension.

Action: Maintaining Outperform. The risk/reward has compressed materially versus April 2025 (when the asymmetry was extraordinary), but the fundamental thesis is structurally stronger, the FY26 framework extends the trajectory, and the multi-year capex commitment de-risks 2027-2028. For investors already long, this is a quarter to hold through; for new money, current levels remain reasonable for those who can underwrite multi-year AI demand and capacity ramp.

Bottom Line: FY25 in Retrospect, FY26 in Prospect

FY25 Retrospective. TSMC closed 2025 with $122.4B in revenue (+35.9% USD), 59.9% gross margin (+380bps YoY), 46.4% EPS growth, free cash flow crossing NT$1 trillion for the first time, $40.9B in capex, and outgrowing the foundry industry by more than 2x. The full-year picture vindicates every operational bull point we identified at our April 2025 initiation: AI-driven structural demand (HPC at 58% of revenue, +48% YoY), process-technology leadership widening (N2 in HVM in Q4, ahead of schedule), operating leverage delivering bottom-line outperformance (EPS +46.4% on revenue +35.9%), and Arizona transitioning from margin wound to strategic asset (FY25 dilution actual at 1-2% vs. initial 2-3% framework). The FY25 guide cadence — from January's "approximately 20%" through Q3's "mid-30s" — is itself the cleanest narrative arc of the year, with two consecutive mid-year guide raises followed by delivery at the top of the third (highest) framework. This is what a vindicated thesis looks like.

FY26 Prospective. The 2026 framework is a step-up across every axis. Revenue growth ~30% USD implies $155-160B (vs. $122.4B in FY25). Q1 2026 gross margin guide of 63-65% expands above Q4's record 62.3%. Operating margin at 54-56%. Effective tax rate ticking from 16% to 17-18%. Depreciation rising high-teens % YoY. Capex stepping up to $52-56B — the largest single-year semiconductor capital commitment in industry history — with 70-80% allocated to advanced nodes. N2 in volume production with "good yields" and a "very fast ramp" expected, faster than the N3 effort. Advanced packaging crossing 10% of revenue. Wafer ASP increases continuing per CFO Huang's explicit "such price rises will continue going forward" commentary. The Foundry 2.0 industry expected to grow 14% — meaning TSMC plans to take ~16pp of share for the second consecutive year.

The capex commitment is the structural moment. CEO Wei's framing of the $52-56B 2026 capex as customer-verified — "if we don't do it carefully, that'd be a big disaster" — combined with his explicit statement that this capex contributes "little to 2026 output and only marginally to 2027" makes clear this is a multi-year capacity commitment based on contracted demand visibility through 2028. With customers engaging 2-3 years pre-production and CFO Huang signaling 2027-2028 capex will be "significantly higher" than $52-56B, TSMC is making the most consequential industry capacity bet in semiconductor history. The bear case requires this customer-verified demand signal to be wrong; the bull case requires it to be merely correct.

The stock has approximately doubled from the April 2025 lows ($134 to ~$315) over our nine-month coverage window — making this one of our most successful initiations. The risk/reward asymmetry has compressed accordingly: April was extraordinary (~3:1 up-to-down), July was favorable (~1.5:1), October was reasonable (~1.1:1), and Q4 closes at roughly 1.0:1 with bull/bear scenarios both implying ~25-35% magnitude moves. But the thesis is now structurally stronger than at any prior point in our coverage: multi-year AI demand visibility, capex commitment de-risking 2027-2028, gross margins settling at a higher cycle level, N2 ramp ahead of schedule, and continued share gain at the leading edge. We maintain Outperform, with a 12-month target range of $340-380 base case and $400-435 bull case.

What could change our mind:

• Downgrade to Hold: Q2 2026 print showing gross margin compression below 58%; or Q3 2026 print revealing meaningful customer order pull-back; or a Section 232 outcome that triggers 15%+ chip-specific tariffs with limited pass-through; or hyperscaler capex guides for 2027 that show sub-10% growth; or evidence that the $52-56B 2026 capex commitment is being scaled back materially.
• Downgrade to Underperform: A direct Taiwan-strait escalation with operational impact; a forced Intel JV mandated by the Trump administration; gross margin compression below 54% on sustained NT$ strength plus accelerating overseas dilution plus AI demand moderation; or a meaningful (>$5B) capex underabsorption signal in 2027-2028 prints suggesting demand is moderating.

Signposts for Q1 2026 earnings (April 2026):