The Guidance Ceiling Was the Floor: 66.2% Gross Margin, Agentic AI Names the Next Wave, and the FY26 Guide Gets Raised — Maintain Outperform

Results vs. Consensus

Q1 2026 Scorecard

Year-Over-Year Comparisons

Quarter-Over-Quarter Comparisons

Quality of Beat

Revenue: The 0.6% NT$ beat against LSEG consensus understates Q1's quality. Revenue landed at the absolute top of the $34.6–35.8B guidance range ($35.9B vs. $35.8B ceiling) — which is a small beat on the headline but is operationally meaningful in context: TSMC sets conservative guidance ranges and finishing at or above the ceiling for the third consecutive quarter is a consistent pattern of under-promising and over-delivering. The more important signal is the +40.6% YoY USD growth rate on a Q1 2025 base that was already elevated.

Margins: The 66.2% gross margin and 58.1% operating margin are the cleanest beats in the report. Three structural tailwinds converged in Q1: (1) N5 and N3 together at 61% of wafer revenue with both nodes now at or above the company average gross margin; (2) AI demand that is capacity-constrained not demand-constrained, driving peak utilization; and (3) cost improvements at N3 as the node has matured. CFO Huang's explicit attribution to "capacity utilization and cost improvements" confirms the beat is operational, not accrual-timing or FX-driven.

EPS: The 58.3% YoY EPS jump is fully operational. Revenue grew 40.6% YoY; the incremental leverage from margin expansion drove EPS growth roughly 1.4x revenue growth. No significant below-the-line items distort the figure. FX was a mild tailwind as NT$ remained stable around 31 to the USD versus the Q1 2025 period. This is a structurally cleaner EPS beat than many comparable semiconductor results.

Segment Performance

Platform Revenue Mix — Q1 2026

HPC / AI — 61% of Revenue, Record Share

HPC at 61% of wafer revenue represents a record mix share and continues to structurally re-define what TSMC is. Two years ago, mobile chips held plurality. Today, AI accelerators and HPC chips together account for a majority that is still growing. The +20% QoQ growth in HPC came in a quarter where total company revenue grew 6.4% sequentially — meaning the rest of the portfolio (smartphone, auto, IoT, DCE combined) was essentially flat-to-declining while AI drove every unit of incremental growth. Implied Q4 2025 HPC revenue was ~$18.5B; Q1 2026 is ~$21.9B. That is $3.4B of QoQ incremental HPC revenue added in a single quarter on what was already a record base.

Assessment: HPC at 61% is structurally self-reinforcing. Advanced-node capacity is being added almost entirely to serve HPC customers; the customers engaging 2–3 years pre-production are AI hyperscalers and custom silicon designers. The mix shift is not reversible in the near term — there is no force that reallocates capacity from HPC to smartphones or automotive at meaningful scale. Watch for whether HPC sustains 60%+ in Q2 as smartphone typically recovers seasonally.

Smartphone — Seasonal, Not Structural

The −11% QoQ decline in smartphone is standard Q1 seasonality. Mobile OEMs reset inventory between January and March after the Q4 holiday ramp, creating a recurring Q1 trough. No competitive displacement or customer share loss is evident. Smartphone at 26% of Q1 revenue is consistent with the same period in prior years. Q2 typically recovers modestly toward 28–30%.

Assessment: Monitor for Q2 normalization. The smartphone bear case (Apple diversifying to non-TSMC fabs, Samsung flagship recapturing share) remains immaterial in the near-term: Apple's entire A-series ecosystem is on N3, and there is no announced migration timeline.

Automotive — Inventory Correction Persists

Automotive declined 7% QoQ to approximately $1.4B (4% of revenue). The inventory correction that began in 2024 has extended into Q1 2026. TSMC's exposure is primarily through its advanced and specialty-node customers (Infineon, NXP, Renesas, Mobileye) who supply the auto OEM ecosystem. The correction appears to be progressing rather than accelerating, suggesting it is not structurally worsening.

Assessment: At 4% of revenue, automotive has essentially no near-term model impact. Any recovery in H2 2026 would be additive upside. The EV silicon content tailwind remains intact as a 2027–2028 driver.

Technology Node Revenue Mix — Q1 2026

The 74% advanced-node revenue mix is a record and directly drives the 66.2% gross margin result. N5 at 36% is now a fully mature, high-yield node that runs above the company average gross margin. N3 at 25% is approaching that threshold, with management confirming the crossover in H2 2026. The structural picture: as N3 turns accretive and N2 begins its initial ramp, TSMC will for the first time have three high-volume nodes (N7, N5, N3) simultaneously above the corporate gross margin average — while N2 follows the same yield maturation curve.

Key Topics & Management Commentary

Overall Management Tone: The Q1 2026 call has a qualitatively different character than prior TSMC calls. In Q4, the defining moment was capex commitment — a CEO putting $52–56B behind a multi-year demand signal. In Q1 2026, the defining moment is a demand characterization upgrade: Wei explicitly named the agentic AI transition as generating "another step up in the amount of tokens being consumed." This is not a repetition of prior AI commentary; it is a forward-looking observation that a new demand driver is already visible. The overall tone is confident and operationally grounded, not promotional. When Wei says AI demand is "extremely robust," the 66.2% gross margin is the validation.

1. Agentic AI: The Next Token Consumption Step-Up

"AI related demand continues to be extremely robust. The shift from generative AI and the query mode to agentic AI and command and action mode is leading to another step up in the amount of tokens being consumed." — C.C. Wei, Chairman & CEO

This is the most strategically significant statement on the Q1 2026 call. It marks the first time TSMC's CEO has explicitly named agentic AI as a distinct and additive demand catalyst. The distinction matters: generative AI in query mode (ChatGPT-style interactions) processes one request at a time, with token consumption proportional to the prompt and response length. Agentic AI in command-and-action mode executes multi-step tasks, spawns sub-agents, maintains state across extended workflows, and performs actions in the real world. The token consumption per user interaction is orders of magnitude higher.

Wei's framing suggests TSMC is already observing this transition in customer order books — hyperscalers and AI labs are expanding capacity commitments to accommodate agentic workloads on top of generative inference. This is not a 2028 story; it is a 2026–2027 demand catalyst that is layering onto an already capacity-constrained environment. If agentic AI deployments materially scale in 2026, the already-bullish FY27 capacity outlook becomes even more demand-constrained.

Assessment: This is the most important datapoint on the call for the long-term thesis. The AI accelerator CAGR upgrade to 54–56% (from 45%) is the quantitative expression of agentic AI's incremental demand. A company of Wei's credibility and operational visibility does not voluntarily raise a five-year CAGR by 9–11 percentage points without already observing the trajectory in customer commitments.

2. AI Accelerator CAGR Raised to 54–56% (2024–2029)

Management raised the revenue CAGR for AI accelerators (2024–2029) from 45% to 54–56%. The compound math is striking: over five years, the difference between a 45% CAGR and a 55% CAGR on a $20B 2024 base produces revenue that is approximately 65% higher at the 2029 endpoint ($~460B vs. $~280B, illustratively). This is not a marginal adjustment — it is a structural upgrade to the demand forecast.

The revision reflects three observable trends already mentioned by Wei: (1) generative AI inference workload scaling, where more deployed models means more parallel inference capacity needed; (2) agentic AI emergence, as described above; and (3) sovereign AI, where governments committing to multi-year compute buildouts are creating a third and more stable demand pillar alongside hyperscalers and AI labs.

Assessment: At a 55% CAGR from a 2024 AI accelerator revenue base of approximately $20B, AI accelerator revenue reaches ~$450–500B by 2029 — nearly half of TSMC's projected total revenue at that time. This framing represents TSMC's CEO telling the market that AI acceleration is now the company's defining business for the rest of the decade. We have no reason to dispute it given five consecutive quarters of confirmation.

3. Gross Margin: 66.2% Shatters the Guidance Ceiling

"[Gross margin] exceeded the high end of guidance by 120 basis points, mainly due to higher capacity utilization and cost improvements." — Wendell Huang, CFO

The 66.2% gross margin is the operational centerpiece of the quarter. For context: at Q3 2025, management guided Q4 at 59–61%. Actual delivered 62.3%. At Q4, management guided Q1 2026 at 63–65%. Actual delivered 66.2%. The pattern is now five consecutive quarters of gross margin delivery above the high end of guidance — a manufacturing precision story as much as a demand story.

Three structural tailwinds drove the 390bp sequential expansion from Q4 2025's 62.3%: (1) N5 fully mature at above-average margin, running at peak utilization driven by Nvidia Hopper/Blackwell successors and AMD MI-series; (2) N3 approaching but not yet at corporate average — yield improvements continuing to drive cost reductions; (3) overseas fab dilution tracking at the lower end of the 2–3% framework, with the US-Taiwan equipment duty exemption (from the January 2026 trade accord) providing incremental margin support for the Arizona buildout.

The Q2 2026 gross margin guide of 65.5–67.5% (midpoint 66.5%) implies essentially flat sequential margins despite N2 ramp beginning to contribute dilution and seasonal volume mix shift. This is a bullish signal: management is implicitly telling the market that N3 margin maturation in H2 2026 will offset N2 dilution in H1.

Assessment: The structural gross margin floor for TSMC has been repriced by each of the last five prints. The "56% long-term target" phrasing in prior commentary is increasingly a legal/guidance floor rather than an operational expectation. Through-cycle gross margin for a company with 74% advanced-node revenue mix, pricing power confirmed by CFO Huang, and N3 turning accretive is more plausibly 62–65% than the legacy 56% framework. Every 100bp upward revision in the market's through-cycle margin assumption implies $1.6B of incremental gross profit at FY26 revenue scale.

4. N2 Ramp: "Very Fast," Tracking Ahead of N3

N2 (2nm-class) entered high-volume manufacturing in Q4 2025 and is in active ramp at Fab 20 (Hsinchu) and Fab 22 (Kaohsiung) with yields described as "good." The most significant N2 data point from the call is Jeff Su (IR Director) reiterating that N2 will ramp "very fast" in 2026 and deliver more revenue, faster, than the equivalent N3 effort. N2 wafers are priced approximately 50% above N3, so even modest share gains represent substantial dollar contributions.

If N2 follows the historical node trajectory at TSMC — initial 2–3% wafer share, growing to 10–15% within 4–6 quarters — and if the ramp truly outpaces N3, we could see N2 at 6–8% of wafer revenue by Q4 2026. At $35–40B quarterly revenue, that would be approximately $2.1–3.2B of N2-attributed revenue in a single quarter by year-end 2026. Given N2's 50% ASP premium over N3, the dollar contribution grows non-linearly with share gains.

The 2–3pp gross margin dilution from N2 ramp in 2026 is real and already factored into guidance. But the mathematical asymmetry of N2 is compelling: every quarter N2 dilutes margins 50–100bp from its initial ramp phase, it also adds 1.5–2x that amount in revenue-per-wafer versus an equivalent N3 quarter. N2's dilution period is a near-term cost; its accretive phase (likely H2 2027–2028 based on N3's pattern) is a multi-year structural tailwind.

Assessment: N2 is on its best possible trajectory. Multiple customers (Apple for A20 chip, AMD for next-gen CPUs, Nvidia for post-Blackwell architecture) are confirmed in various industry reports for N2 production. The competitive landscape remains a one-horse race: Samsung 2nm yields are commercially unproven; Intel 18A has limited outside customer commitments. TSMC's monopoly at the leading edge strengthens with each node generation.

5. N3 Crossing Company Average Gross Margin in H2 2026

Management confirmed that N3's gross margin will cross the corporate average in H2 2026. This is the elimination of one of the two primary structural margin headwinds that has compressed TSMC's gross margin relative to its fully mature portfolio. N3 has been margin-dilutive since its 2023 introduction due to high initial tooling costs, mask complexity, and yield development overhead. Reaching corporate average removes that drag entirely.

The gross margin expansion math: if N3 (25% of revenue) goes from ~150bp dilutive to neutral in H2 2026, that is approximately 37bp of structural improvement on a blended basis. Combined with N5 already above average and N7 as a high-margin mature node, the H2 2026 portfolio will have the cleanest margin profile in TSMC's history at the advanced-node tier — while N2 incurs its own dilution, which is smaller in initial share and higher in ASP offset.

Assessment: This was a known positive catalyst from the Q4 recap, and Q1 confirms it is on track. It is the structural basis for our conviction that FY26 gross margins land at 65–67% for the full year rather than compressing toward the 63–65% Q1 guide midpoint.

6. Rare N3 Capacity Expansion — Breaking Historical Foundry Policy

The most consequential strategic disclosure on the Q1 2026 call was not the margin beat, the guide raise, or the agentic AI framing — it was TSMC announcing a global N3 capacity expansion in direct departure from its own stated capacity-allocation policy. C.C. Wei was explicit about the policy break:

"Historically, we do not add additional capacity to a node once it's reached its target capacity. However, as a foundry, our first responsibility is to provide our customers with the most advanced technologies and necessary capacity to unleash their innovations." — C.C. Wei, Chairman & CEO

This is not a rhetorical concession. TSMC's capacity-allocation model is built around staged node transitions: each new generation (N7 → N5 → N3 → N2 → A14) is ramped to target capacity, then preserved as a high-margin mature fleet while the next generation absorbs new capex. Adding material new capacity at an N-1 node breaks that model and requires a qualitatively different justification — multi-year contracted customer demand at that node in addition to the leading-edge build-out, not instead of it.

The global N3 build-out — four concurrent vectors

The in-place 5nm-to-3nm tool conversion is the tell that this is demand-pulled, not precautionary. Converting working N5 tools to N3 is only justified if N3 demand is contracted through the depreciation tail of the converted equipment — a multi-year commitment window. A foundry does not convert a productive node to an older generation absent firm customer capacity reservations.

Why this matters — four implications

1. Multi-year AI demand is broader than leading-edge alone. N2 and A14 will absorb the leading-edge AI accelerators through 2028. The N3 expansion is additive — it tells us hyperscalers and custom-silicon designers have contracted N3 capacity through at least 2028–2029 for chiplet architectures, AI peripheral silicon (networking, storage controllers, memory controllers), and mid-tier accelerators. The market has been modeling AI silicon demand as a leading-edge-only story; N3 expansion at scale reframes it as a multi-node story.

2. N3 recasts from decaying node to multi-year margin compounder. The original Section 5 above established that N3 crosses the corporate average gross margin in H2 2026. The capacity expansion extends that thesis: if N3 cross-over is H2 2026 and substantial new N3 capacity comes online in 2027–2028, N3 becomes accretive at an expanding revenue base rather than a declining one. The H2 2026 margin crossover is the start of a multi-year tailwind, not a one-time inflection.

3. Contracted pricing at N3 is confirmed. TSMC does not greenfield fabs or convert working tools at a maturing node absent long-term pricing commitments that support the ROI hurdle. The fact that capital is being committed to N3 capacity is direct evidence that TSMC's pricing power extends through N-1. For FY27–FY28 model assumptions, this supports holding N3 ASPs flat or modestly accretive rather than the historical convention of node-ASP decay.

4. Competitive moat widens at N-1 as well as leading-edge. Samsung Foundry and Intel Foundry have focused their catch-up narratives on leading-edge (Samsung 2nm, Intel 18A/14A). TSMC expanding at N3 — while simultaneously running N2 in HVM and building A14 for 2028 — forces competitors to fight on two fronts. Samsung and Intel cannot credibly offer foundry customers N3-equivalent capacity at TSMC's yields and pricing without a multi-year capex commitment they have neither committed to nor financed. The moat is widening at N-1 in the exact window competitors were hoping to use as a beachhead.

Assessment: Thesis-strengthening. On top of margins, guide, and AI demand framing, the rare N3 capacity expansion is incremental evidence that multi-year demand visibility extends beyond the leading edge. This signal raises the conviction weight on the Outperform rating and extends the FY27–FY28 revenue floor in our model.

7. FY26 Guide Raised: "Close to 30%" Becomes ">30%"

At Q4 2025, management characterized FY26 growth as "approximately 30%" or "close to 30%" USD. On the Q1 2026 call, Wei upgraded this to ">30%." This is a qualitative but meaningful change: it shifts the guidance from a point estimate to a floor, and removes the implicit downside scenario where FY26 growth lands just below 30%.

With Q1 actual of $35.9B and Q2 guided to a $39.6B midpoint, H1 2026 totals ~$75.5B. To hit exactly 30% growth from FY25's $122.4B, TSMC needs $159.1B for the full year — implying H2 averages $41.8B per quarter. The Q2 midpoint of $39.6B shows H1 already running well ahead of pace; a modest sequential expansion into Q3–Q4 (supported by HPC demand and smartphone seasonality) gets FY26 comfortably above $160B. Our estimate of $163–168B (33–37% growth) remains in force.

Assessment: The guide upgrade is the right directional signal. Every quarter of Q1–Q2 2026 actuals and guidance is pointing to FY26 landing at 33–37% growth, not 30%. Management's conservatism in maintaining ">30%" as the formal guide rather than raising to "mid-30s%" is consistent with their historical guidance style. We expect the guide to formally migrate to "mid-30s%" language at Q2 earnings.

8. Capex: Toward the High End of $52–56B

Management confirmed the capex trajectory is "toward the high end" of the $52–56B 2026 range, implying approximately $55–56B. Wei's explanation was unusually direct: demand is "very robust," TSMC is "trying very hard to speed it up and pull in all the equipment," but supply of EUV equipment and cleanroom capacity remains tight. The capex is constrained by supply chain, not by demand uncertainty.

This is the most bullish possible framing for the capex commitment. Prior periods where TSMC committed large capex amid demand uncertainty typically involved hedged language. Here, Wei is saying: we want to spend even more, and the limiting factor is equipment delivery schedules, not customer order books. This context removes the speculative-capex-bear narrative entirely.

Assessment: The capex being supply-constrained rather than demand-constrained validates the multi-year thesis at its deepest level. The capex commitment is not a forecast; it is an allocation of what TSMC can actually order and deploy given ASML's EUV production limitations. TSMC is capacity-constrained on both the supply side (equipment deliveries) and demand side (more demand than available capacity), which is the exact definition of pricing power and sustainable margin expansion.

9. Arizona: Third Fab Under Construction, Phase 2 in Tool Installation

The Arizona expansion timeline progressed as expected. Phase 2 (N3P process technology) has entered the tool installation phase in 2026, with production volume scheduled for H2 2027. A third fab is now under construction; a fourth fab and advanced packaging facility have applications submitted. The $165B commitment across up to 12 Arizona fabs provides a multi-decade framework for TSMC's US footprint.

The overseas fab dilution framework remains at 2–3% of gross margin initially, widening to 3–4% as more fabs ramp. With the Phase 1 Arizona fab (N4 node) now in HVM since Q4 2024 and demonstrating yield parity with Taiwan — a milestone not previously achieved by a US semiconductor facility at this node — the fundamental execution risk of Arizona has been retired. The debate has shifted from "will it work?" to "how fast can it scale?"

"There are no shortcuts. That's a fundamental of the foundry industry." — C.C. Wei, Chairman & CEO (on the Terafab/Tesla-Intel manufacturing JV timeline)

Wei's Terafab comment was in response to analyst questions about Tesla and Intel's joint venture announcement for AI compute manufacturing in Austin. Wei's response — "no shortcuts" — is a polite but direct statement that chip manufacturing at advanced nodes requires 2–3 decades of process development and supply chain ecosystem accumulation. TSMC views Terafab as a long-horizon project that, if it materializes, could become a TSMC customer rather than a competitive threat. The foundry's current customer base for N2 and beyond is locked in through contractual capacity commitments that predate Terafab by years.

Assessment: Arizona is now an operating asset, not a speculative commitment. Phase 1 yield parity, Phase 2 in tool installation, and Phase 3 under construction represent a $165B physical commitment that is de-risked on execution. The geopolitical hedge value is real: under any tariff escalation scenario, US-fab TSMC chips are tariff-insulated by statute.

10. US-Taiwan Tariff Relief: Equipment Duty-Free, Finished Goods at 15%

The US-Taiwan trade accord reached in January 2026 reduced tariffs on Taiwanese goods from 20% to 15%, and crucially granted duty-free treatment for semiconductor manufacturing equipment imported during construction phases. For TSMC's Arizona buildout, which involves importing billions of dollars of ASML EUV scanners, Applied Materials deposition tools, KLA metrology systems, and specialized cleanroom infrastructure, this equipment exemption directly reduces the effective cost premium of US-based manufacturing.

TSMC's business model involves selling wafers to chip designers who manufacture final products; TSMC does not directly export finished chips. The 15% tariff on finished goods is therefore primarily relevant to TSMC's downstream customers (Nvidia, AMD, Apple, Qualcomm) rather than to TSMC's own P&L. The equipment duty exemption, by contrast, is directly margin-accretive for the Arizona buildout over the next 3–5 years of construction.

Wei flagged Middle East instability as a potential risk to chemical and gas prices used in semiconductor manufacturing. This is a modest second-order risk — semiconductor-grade gases (nitrogen, argon, helium, specialty etching gases) have diversified global supply chains and are not single-sourced from the Middle East — but it is worth noting as a new risk acknowledgment from management that was not present in Q4.

Assessment: The tariff accord is a net positive. Equipment duty-free provision reduces Arizona buildout costs; the 15% finished goods rate vs. a hypothetical 25%+ worst-case reduces the tail risk for TSMC's customers. Middle East gas/chemical risk is real but contained to 50–100bp of potential gross margin impact at most, and is not structural.

11. Capital Position: NT$3.4T Cash, NT$699B Operating Cash Flow

TSMC's balance sheet ended Q1 2026 with cash and marketable securities of NT$3.4 trillion (~$106B USD), up from NT$3.1T at year-end 2025. Operating cash flow for Q1 was NT$699B (~$21.8B), demonstrating the extraordinary cash generation of the business even as it absorbs the early quarters of its record capex cycle. Management maintained the dividend growth posture without announcing a buyback, consistent with Q4 commentary.

Assessment: NT$3.4T of cash at NT$3.4T is arguably the most underappreciated data point in TSMC's financial profile. Even with $55–56B of FY26 capex commitments, the company is generating cash faster than it is deploying it for capex in any given quarter. The structural FCF machine running underneath the capex cycle provides a resilience floor that many capital-intensive businesses lack. The dividend growth posture and the absence of a buyback reflect rational capital allocation given the capex commitment horizon — not a balance sheet weakness.

Guidance & Outlook

The Q2 2026 revenue guide of $39.0–40.2B is the headline forward data point. A $39.6B midpoint represents +10.3% sequential growth from an already-elevated Q1 base of $35.9B. For a company generating $40B per quarter — $160B per year — a 10% sequential growth rate is exceptional at any scale. Q2 is not a seasonally strong quarter for TSMC; it typically benefits from modest smartphone recovery but is not driven by the Apple A-series ramp (that begins Q3). The 10% sequential guide therefore reflects AI/HPC demand acceleration, not seasonal recovery.

The Q2 gross margin guide of 65.5–67.5% (midpoint 66.5%) is essentially flat with Q1's 66.2% actuals. This is a bullish signal: management is absorbing initial N2 ramp dilution (which begins contributing in Q1–Q2 2026) without any sequential compression in guided margins. The implication is that N3 maturation tailwinds are offsetting N2 ramp headwinds on a roughly one-for-one basis. When N3 fully crosses the corporate average in H2 2026, the offset dynamic improves and margins should expand further.

Implied FY26 Revenue Trajectory: Q1 actual $35.9B + Q2 guided midpoint $39.6B = H1 total ~$75.5B. Management's ">30% USD growth" requires $159.1B minimum. H2 requires $83.6B or ~$41.8B/quarter. With Q2 already at $39.6B and AI demand described as supply-constrained rather than demand-constrained, H2 averaging $42–44B per quarter (supported by Apple A20 N2 ramp in Q3 and continued AI infrastructure demand) is the base case. Our FY26 revenue estimate of $163–168B (+33–37% USD) is maintained.

Implied FY26 Gross Margin: If Q1 (66.2%) and Q2 (66.5% midpoint) establish H1 at ~66.3%, and H2 benefits from N3 crossing the corporate average while incurring moderate N2 ramp expansion, a full-year FY26 gross margin of 65–67% is plausible. This compares to our prior model of 60–62% and represents a structural upward revision. The through-cycle margin floor has been repriced by this print.

Capex Guidance: At $55–56B (toward the high end of $52–56B), FY26 capex represents approximately +35–37% over FY25's $40.9B and the largest single-year semiconductor capital commitment in history. The supply-constrained framing (Wei explicitly said TSMC is limited by EUV equipment delivery, not demand uncertainty) removes the speculative-capex bear narrative. FY27 capex is not yet guided but Huang's prior "significantly higher than FY26" framework implies $60B+ annually — sizing for 2028 output on top of the 2027 output from this year's equipment orders.

Analyst Q&A Highlights

What Drives the Decision to Expand N3 Capacity After 4+ Years in Production

The opening question pressed management to identify which applications justify breaking TSMC's historical policy against adding capacity to a node that has reached its target. The question framed a node that began HVM in late 2022 as one that conventionally would be entering a managed-decline phase, not a multi-fab global expansion phase. Management's answer was a single application set — HPC AI — with no hedging across smartphone, IoT, or automotive end-markets that the press release also listed as N3 demand sources.

Q: "I would like to start with your 3-nanometer gross margin outlook. You just mentioned the node is going to cross the corporate average gross margin in second half this year… And we understand the technology is in severe undersupply backed by strong AI demand, and you already forecasted the capacity expansion through conversion and greenfield through 2028. Would you be able to discuss more in detail on what kind of applications are driving such strong business for you and convince you to expand more?"
— Haas Liu, Bank of America

A: "Let me answer that. I think the application is simple. It's still the HPC AI applications. Does that answer your question?"
— C.C. Wei, CEO

Assessment: The terseness of Wei's answer is itself the signal. When a CEO collapses a multi-application capacity decision to a single demand vector, it confirms that the marginal capacity decision is HPC AI alone — smartphone, IoT, and automotive N3 demand is being served from existing capacity, and every incremental fab is being built for AI. This is a cleaner read of N3 capacity intent than any blended-mix framing would have provided.

What Pulled 2026 CapEx to the High End of the $52–56B Range

A follow-up question probed what changed in the customer-demand picture between January (when the $52–56B range was set with no high-end commitment) and April (when management formally guided to the high end). The question explicitly cited the conversations with customers and customers' customers that drive the CapEx framework, asking what specifically tilted those conversations toward the upper bound.

Q: "You revised up to the high end of your guidance for USD 52 billion to USD 56 billion for this year. Compared to 3 months ago, what gives you the incremental confidence when you discuss with your customers and also customers' customers regarding the demand outlook to support your stronger or the upper half of your guidance for the CapEx this year?"
— Haas Liu, Bank of America

A: "A very simple answer is, the demand are very robust, especially from the HPC and AI applications. And also, we try very hard to speed it up and pull in all the equipment as we can. Still, our supply is very tight. Demand is continuing to increase. And so we continue to work with our suppliers to speed it up. And that's why we are towards our high end of CapEx forecast."
— C.C. Wei, CEO

Assessment: "Still our supply is very tight" three months after the most aggressive capacity plan in TSMC's history is the most important phrase here. The high-end CapEx isn't a comfort move — it is supply-side acceleration against demand that the original $52–56B range was designed to meet. The implication for 2027 CapEx is that the upward bias is structural; this is the start of a multi-year elevated-capex regime, not a one-year peak.

Duration of the Leading-Edge Supply Tightness

The question asked whether the expanded 3-nanometer plan plus accelerated CapEx is sufficient to bring supply and demand into balance over a 2–3 year window, given the consistent customer message that wafers remain the binding constraint. Management's answer extended the supply-tightness window beyond 2027 and pointed to the 2027 capex announcement as the next disclosure that will inform investors on the trajectory.

Q: "Any read, C.C., on when we can kind of meet this demand? Or do you think the next couple of years is still going to be very challenging to meet — that supply is still going to be running below demand, let's say, into '27 also?"
— Gokul Hariharan, JPMorgan

A: "Gokul, you know we are — it takes 2 to 3 years to build a new fab. And with the current schedule, we believe that '27, we will announce it anyway when we enter '27, but let me say that, it takes time to build a new fab, it takes time to ramp it up. And so we expect this to continue to be very tight. So that's why we just announced that we try to build 3 new N3 fab to meet the demand."
— C.C. Wei, CEO

Assessment: Wei extending the "very tight" framing through 2027 is the cleanest possible confirmation that the 2026–2027 capacity expansion is demand-pulled, not precautionary. For modeling, this supports continued pricing power and high utilization through at least the H1 2027 capacity additions, with the H2 2027 Arizona Fab 2 and 2028 Japan Fab 2 ramps providing the next relief points — both of which are 18–30 months away.

The Terafab Initiative and the Question of Customer Defection

A pointed line of questioning surfaced the Tesla/SpaceX/xAI "Terafab" announcement and the broader observation that one of TSMC's largest design customers had recently signed wafer commitments with Samsung. The question asked directly how TSMC plans to win that business back. Wei's answer leveraged structural foundry barriers — technology leadership, manufacturing excellence, customer trust, service — framed against the same "2 to 3 years to build a fab" lead-time argument.

Q: "You have the traditional competitors, Samsung, Intel. But one of your customers, Elon Musk, also announced Terafab Initiative recently. What is TSMC's perspective on this initiative? They have also been a customer of yours, and they recently signed a deal with Samsung a few months back. So what is TSMC's response here now that they are also trying to kind of build chips on their own? How are you trying to win back this customer?"
— Gokul Hariharan, JPMorgan

A: "Both Intel and Tesla, they are TSMC's customers. But again, they are our competitors, and we view Intel as our formidable competitor and do not underestimate them. But having said that, there are no shortcuts. The fundamental rules of the foundry game never change. They need the technology leadership, manufacturing excellence and customer trust, and most of all, the service… Again, let me say that it takes 2 to 3 years to build a new fab, no shortcuts. And it takes another 1 to 2 years to ramp it up."
— C.C. Wei, CEO

Assessment: The "no shortcuts" framing is the correct analytical answer. The Terafab project, even if fully executed, is a 5–10 year horizon with first production likely serving Tesla/xAI internal silicon only. The Samsung wafer commitment for the customer's LPU product is a real near-term share concession, but the structural answer is unchanged: leading-edge foundry capability has a 5-year build cycle, and TSMC's customer relationships are 2–3 year pre-production contracted commitments. This is not a near-term thesis-break risk.

Whether a Multi-Year CapEx Guide Is Coming Back

A question revisited the 2021 management practice of providing a three-year aggregate CapEx commitment (the $100B framework for 2022–2024), asking whether the strength of the current AI demand pipeline would justify reviving that level of forward disclosure. The CFO declined to provide a multi-year dollar number but framed the next three years' CapEx as "significantly higher" than the past three years' $101B aggregate — a useful directional signal even without a number.

Q: "I remember back in 2021, management also provided 3-year CapEx guidance as USD 100 billion given very strong demand. I'm not sure if TSMC can provide a little bit longer-term CapEx guidance? Because as you said, right, the equipment supply is also pretty tight… would the management provide kind of a long-term CapEx guidance to investors?"
— Charlie Chan, Morgan Stanley

A: "We don't have a number to share with you. But look at it this way. In the past 3 years, our total CapEx was USD 101 billion. This year, we're already saying CapEx is towards the high end, which is USD 56 billion, which is already over 50% of the past 3 years in total. So we have a strong conviction in the AI megatrend. So we expect the CapEx in the next few years, in the next 3 years, will be significantly higher than the past 3 years."
— Wendell Huang, CFO

Assessment: Huang's reframing is the most consequential capital intensity disclosure on the call. The 2024–2026 CapEx baseline of $101B over three years, where 2026 alone is $56B (55% of that triennial baseline), implies the next-three-year aggregate could easily exceed $150–180B if 2027 and 2028 sustain $50B-plus. This is a multi-year elevated-capex commitment without the explicit three-year dollar headline, and it sets up a longer-than-expected high-investment regime that funds the AI capacity expansion.

Whether the 56%-Through-Cycle Margin Target Is the New Floor

A question asked whether the strength of current margin performance — Q1 actual 66.2% against a long-term guide of "56% and higher through the cycle" — has prompted management to revisit the through-cycle margin framework upward. The CFO held the 56%-and-higher framing as the current target and characterized the long-term planning process as ongoing rather than committing to a near-term upward revision.

Q: "Has management's thinking changed about the sustainable margin structure and what appropriate longer-term returns might be for the business? […] That thinking is not changing against the backdrop where other parts of the AI supply chain are clearly starting to print super normal returns?"
— Jim Fontanelli, Arete

A: "As we said in the last earnings calls, we've revised up our long-term margin target and ROE target. From 2024 to 2029, we're now saying the gross margins will be 56% and higher through the cycle, and we're looking at ROE of high 20% through the cycle. That's what we're currently looking at, and that's already higher than before… the long-term planning is an ongoing and continuous process. So we do that all the time, and we will update you when there is a change."
— Wendell Huang, CFO

Assessment: Huang's posture is appropriately disciplined — raising the long-term target requires demonstrated structural sustainability, and N2 ramp dilution plus overseas fab widening to 3–4% in the later years are the two binding factors that argue against a near-term upward revision. Our read: the through-cycle target will be revised again in 2026–2027 once H2 2026 N3 corporate-average crossover is confirmed and overseas dilution is stable in the 2–3% band. The current 66%+ print is above the through-cycle expectation, not the new through-cycle baseline.

Whether the "Higher 50s" AI Accelerator CAGR Is Holding Through 2029

The question asked whether the rapid token-consumption growth observed in Q1 has prompted any revision to the mid-to-high-50s AI accelerator revenue CAGR previously communicated for 2024–2029. Wei held the framework at the upper end of that range without formally raising it, while reinforcing the multi-year visibility from customers and customers' customers.

Q: "Management has been guiding that AI accelerator revenue to grow about like mid- to high 50s CAGR in between 2024 and '29. So how does TSMC plan and forecast AI-related demand? I mean, does TSMC incorporate metrics such as total consumption growth in your assumption? Because the recent consumption in the first quarter is definitely accelerated and faster than earlier expectation. Do we see the changes for the AI accelerated revenue growth in the coming years?"
— Bruce Lu, Goldman Sachs

A: "I think I say now that it's a very strong demand, and we continue to receive a very positive signal from our customers and customers' customers. And so what you say is whether we change our CAGR on AI accelerator? Actually, we continue to see strong demand, but again, let me say that it is toward higher 50s of CAGR that we observe."
— C.C. Wei, CEO

Assessment: "Toward higher 50s" is a subtle upgrade from "mid- to high 50s" — Wei is explicitly anchoring the framework at the upper end of the previously-given range rather than at the midpoint. For 2026–2029 modeling, this supports an effective ~55–58% AI accelerator CAGR base case. The "very positive signal from customers and customers' customers" framing matters because it implies hyperscaler order books are still expanding rather than plateauing.

Whether CPUs Will Be Folded Into the AI Accelerator Definition

A specific definitional question raised the increasing importance of data-center CPUs in agentic AI inference workflows and asked whether TSMC's AI accelerator revenue definition — currently scoped to GPU, ASIC, and HBM controllers — will be expanded to include CPUs. Wei acknowledged the structural shift but cited a practical attribution problem: TSMC cannot identify ex ante which CPUs are going to data centers versus PCs and desktops, which is why CPU revenue stays out of the AI accelerator framework today.

Q: "TSMC's definition of AI revenue includes data center GPU, AI accelerator, HBM-based stack… but it specifically excludes data center CPU… with the CPU, there's more and more conversation about CPU now becoming part of the AI infrastructure, especially for agentic workloads. Any chance for TSMC to maybe provide us revised numbers for AI revenue and maybe AI revenue growth, CAGR projection going into 2029, 2030[?]"
— Charles Shi, Needham

A: "Certainly, CPUs become more and more important in today's AI data center. But actually, let me share with you — this is a good question, by the way. Let me share with you that we are not able to identify which CPU goes where, right? It's PC or desktop or it's AI data center. So today, we still not include the CPUs in our AI HPCs calculation. Someday later, we might consider."
— C.C. Wei, CEO

Assessment: The "someday later, we might consider" language opens the door for a CPU inclusion expansion that would mechanically lift the AI accelerator revenue baseline and the forward CAGR framework. If/when this redefinition lands — most plausibly at an investor day in 2027 once Vera Rubin and other AI CPU products are at scale — the AI accelerator revenue line would step-function up, with the CAGR framework retained on the new base. This is a positive-skew optionality the Street is largely not pricing today.

Market Reaction

• Pre-earnings ADR (April 15 close): ~$377, reflecting strong pre-print AI semiconductor sentiment and TSMC's ~$190B YTD ADR appreciation from the January 2026 close of ~$310.
• Post-release ADR reaction (April 16–17): Declined approximately 2.5–3.0% to the ~$365 range. The decline occurred despite a comprehensive beat on all financial metrics and exceptional forward guidance.
• Volume: Elevated but not extreme — consistent with a "sell the news" dynamic rather than fundamental repositioning.
• Peer reactions: NVDA, AMD, AVGO, ASML, AMAT, KLA moved approximately +1–2% on the results, consistent with the AI infrastructure demand confirmation.

The 2.5–3% decline on a comprehensive beat warrants explanation. Two dynamics appear to be driving the move:

"Sell the News" Positioning: TSM appreciated significantly in the weeks heading into the print on elevated AI semiconductor sentiment. Investors who bought in anticipation of a strong report took profits on confirmation. This is a technical supply/demand dynamic in the ADR, not a fundamental read.

Macro Overhang: April 2026 macro conditions include ongoing US tariff policy uncertainty, Middle East instability affecting energy and logistics prices, and concerns about the consumer electronics end-market. TSM as an ADR is subject to risk-off flows that affect global equities broadly. The 2.5–3% decline on this print is consistent with multiple prior TSMC prints where strong results have been met with modest sells rather than rips — Q3 2025 (−3% on a beat-and-raise) being the clearest precedent.

The precedent from Q3 2025 is instructive: TSMC printed −3% on that quarter's beat-and-raise, and the stock then re-rated to +25% over the following three months as the fundamental results continued to compound. We expect a similar pattern here. The business is operating at maximum clarity; the decline is noise.

Street Perspective

Debate: Is the 10% Sequential Q2 Guide Sustainable Through H2?

Bull view: Q2's $39.6B midpoint is supply-constrained; demand exceeds available capacity. H2 benefits from Apple A20 N2 ramp (beginning Q3), continued AI infrastructure buildout, and sovereign AI programs. H2 averaging $42–44B per quarter implies FY26 at $165B+ — a 35%+ year on top of 36% in FY25. Two consecutive 35%+ years at $100B+ revenue is unprecedented in semiconductor history but is supported by contracted customer demand.

Bear view: Q2's 10% sequential guide is partly catch-up from Q1 seasonality. H2 will face the cyclical smartphone normalcy (iPhone 17 cycle base comparisons) and potential HPC demand pause if hyperscalers absorb current capacity and moderate incremental orders. A 5–7% H2 sequential growth rate implies FY26 at $155–160B — still above 30% but below the bull case. Macroeconomic deterioration (tariff escalation, consumer weakness) would reduce the upside case further.

Our take: Bull view for H1, moderate view for H2. The supply-constrained framing supports Q2; H2 cadence depends on whether AI demand pull-through from agentic AI deployments begins scaling before fiscal year-end. We model H2 at $41–43B per quarter and FY26 at $163–168B.

Debate: Have Gross Margins Structurally Repriced, or Is This the Cycle Peak?

Bull view: 66.2% in Q1 with N2 barely contributing and overseas dilution still tracking. When N3 fully crosses the corporate average in H2 2026, the baseline rises further. N2 ASP premium (50% above N3) means each percentage point of N2 share adds 50bp+ to blended ASP. The "56% long-term target" is a regulatory guidance floor, not an operational expectation. Through-cycle gross margin is now 63–66%.

Bear view: 66.2% is a cycle-peak driven by peak utilization, FX stability, and N3 yield maturation occurring simultaneously. When utilization normalizes to 85% (from near-100%), N2 ramp dilution peaks, and overseas fabs widen to 3–4% dilution in 2027–2028, gross margins compress toward 61–63%. The 66% print is the high-water mark of the current cycle.

Our take: Partial bull. Q2–Q3 gross margin staying at 65.5–67% (per the Q2 guide) while N2 ramps is the definitive test. If H2 2026 sustains 65%+ gross margin despite N3 being at corporate average and N2 at 3–5% share, the structural repricing narrative wins decisively. We model FY26 at 65–67% and through-cycle at 62–65% — materially above the 56% historical framework.

Debate: Does Agentic AI Represent a Genuine Incremental Demand Wave?

Bull view: Wei's agentic AI framing is grounded in customer order book observations, not speculation. Agentic AI's token consumption profile is structurally higher than generative AI — multi-step workflows, persistent state, real-world action execution. If enterprise agentic AI deployments scale through 2026–2027 (signs are already visible in Salesforce, ServiceNow, and enterprise software AI agents), the incremental compute demand is an order of magnitude beyond current generative inference scaling. The AI accelerator CAGR upgrade to 54–56% is the quantitative expression of this view.

Bear view: Agentic AI is the latest rebranding of AI capex justification. Enterprise adoption has been slower than hyperscaler announcements imply. The incremental token consumption per enterprise agentic workflow is real but the addressable market of deployed enterprise agents is still small. The CAGR upgrade could be rationalized by training demand alone, without agentic AI contributing meaningfully by 2026.

Our take: Bull wins for 2026; the verdict on 2027–2028 agentic contribution depends on enterprise deployment velocity. What matters for TSMC near-term is that management is observing this trend in customer commitments — which are made 2–3 years pre-production. If agentic AI is in customer order books today, it is producing silicon demand in 2026–2028.

Debate: Valuation — Does the 2.5–3% Decline Create an Entry Point?

Bull view: TSM at ~$365 ADR with FY26 EPS likely at $14.50–16.00 (revised up from Q4's $13–14.50 on the margin trajectory) implies a forward P/E of 23–25x. For a business growing revenue 33–37% and earnings 30–40% with 65%+ gross margins, technology leadership with no near-term competitive threat, and capex commitment de-risking through 2028, this multiple is below the market's treatment of comparable AI beneficiaries (NVDA 30–35x, AVGO 30–35x). The 2.5–3% sell-off is a gift.

Bear view: The stock ran from $310 at Q4 to $377 pre-Q1 print (+21% in three months). The business is priced for continuation of the AI super-cycle; any moderation in H2 guidance or a macro event creates disproportionate downside. At 23–25x, there is limited multiple expansion from here unless earnings grow faster than expected. The risk/reward is 1:1 at best; not asymmetrically attractive.

Our take: The post-earnings price at ~$365 is a better entry than $377 pre-print, but the asymmetry has compressed materially versus the April 2025 initiation or the July 2025 Q2 confirmation. We model a 12-month PT range of $390–430 (base: 26x our $15.25 FY26 EPS midpoint = $396; bull: 28x $16 = $448; bear: 20x $12 = $240). The base case implies ~8–18% upside, supporting the Outperform rating vs. the S&P 500. The bear case downside (~34%) requires a fundamental thesis break, which has not occurred.

Model Update & Valuation Framework

Valuation framework: At ~$365 ADR price post-print and FY26 ADR EPS of $14.50–16.00, TSM trades at approximately 23–25x forward earnings. For a business growing revenue 33–37% in FY26 with 65%+ gross margins, multi-year AI demand contracted visibility, process-technology leadership with no near-term competitive threat, and a capex commitment de-risking 2027–2028 capacity, this multiple is below fair value on a growth-adjusted basis. NVDA and AVGO command 30–35x forward P/E with similar or lower revenue growth profiles and narrower process-technology moats. TSM's geopolitical discount and Taiwan concentration risk are the primary multiple caps — both of which the Arizona buildout progressively addresses.

Revised risk-reward: At $365 ADR, base case ($396–430) implies +9–18% upside; bull case ($445–480) implies +22–31%; bear case ($230–250) implies −32–37% downside. The up-to-down ratio is approximately 1.0–1.5x in base-to-bear scenarios — less asymmetric than April 2025 (3:1) or July 2025 (1.5:1) but still favorable for Outperform relative to the S&P 500 expected return. The AI super-cycle thesis is structurally stronger than at any prior point in our coverage.

Thesis Scorecard: Q4 2025 Signposts Revisited

The Q4 2025 Recap laid out twelve specific signposts for Q1 2026. The results are uniformly bullish, with several exceeding the bull thresholds by material margins.

Scorecard summary: 10 of 13 signposts tripped bullish or better, 3 neutral, 0 bearish — including the N-1 capacity allocation signpost, which tripped exceptionally bullish on the rare N3 expansion. The six highest-conviction signposts (revenue, gross margin, operating margin, N2 ramp, HPC mix, FY26 guide) all exceeded the bullish threshold, three of the six by material margins. This is the cleanest quarterly scorecard in our five-quarter TSM coverage arc.

Full Bull / Bear Matrix: Updated Through Q1 2026

Overall matrix assessment: Nine primary bull points — all confirmed or strengthened, with two new bull points introduced this quarter (agentic AI as incremental demand wave, and rare capacity decisions at N-1 nodes). One primary bear point materially resolved (Section 232 / tariffs), one resolved (FX), one active-moderate (valuation), three active-mild (customer concentration stable, depreciation, memory shortage), and one new mild bear (Middle East). The net bull-bear balance has improved from Q4 to Q1, with three positive developments not anticipated at the Q4 recap: the agentic AI framing, the tariff partial resolution, and the rare N-1 capacity expansion.

Bottom Line: The Fifth Consecutive Outperform

Rating decision: We maintain Outperform on TSM. This is the fifth consecutive Outperform rating in our coverage of TSMC, across seven published reports (Q1 2025 initiation, Q2–Q4 2025 recaps, Q1 2026 flashcap, and this recap). Unlike situations where rating maintenance reflects inertia, the Q1 2026 result requires active confirmation of the Outperform thesis — and it passes every criterion in the framework:

• Revenue above guide: $35.9B vs. $34.6–35.8B guide range. ✓
• Gross margin above guide: 66.2% vs. 63–65% guide. 120bp above ceiling. ✓
• FY26 guide maintained or raised: Raised from "close to 30%" to ">30%." ✓
• No N2 ramp issue: "Good yields," "very fast ramp," more revenue faster than N3 effort. ✓
• No customer concentration reveal: HPC mix at 61% with broad-based AI demand. ✓
• No geopolitical escalation: US-Taiwan accord improved the tariff picture; no new Taiwan-China tension. ✓

The downgrade criteria stated in the Q4 recap were: material guide cut, customer concentration reveal, N2 ramp issues, or new geopolitical shock with concrete trade restrictions. None were triggered. The FY26 guide was raised, not cut. N2 ramp is ahead of N3 pace. HPC is at a record 61% with diversified customers. The US-Taiwan accord reduced tariff exposure. The Outperform rating is confirmed on all four criteria.

The agentic AI observation is the most important new development. Prior quarters confirmed and extended a thesis rooted in generative AI training infrastructure. Q1 2026 introduces an additive demand vector — agentic AI — that structurally increases token consumption per user interaction. If agentic AI deployments scale through 2026–2027 as enterprise AI software increasingly delegates multi-step tasks to AI agents, the demand for silicon is not growing linearly with user count — it is growing super-linearly with each user's agentic engagement depth. TSMC is the only foundry positioned to manufacture the chips for this demand at advanced nodes at scale.

What could change our mind:

• Downgrade to Hold: Q2 2026 gross margin delivery below 63% on sustained N2 dilution exceeding management's 2–3% framework; FY26 guide trimmed below 28% on a Q2 earnings call; hyperscaler capex guides for 2027 in the sub-10% growth range; or a formal Section 232 semiconductor tariff exceeding 15% without pass-through ability.
• Downgrade to Underperform: A direct Taiwan-Strait escalation with operational impact to TSMC fabs; gross margin compression sustained below 57% for two consecutive quarters; or evidence of material N2 yield problems delaying volume production to 2027.

Signposts for Q2 2026 earnings (July 2026):