Guide Raised to 30%, HPC at 60%: The AI Foundry Thesis Just Got Louder — Maintain Outperform

Results vs. Consensus

Quality of Beat

• Revenue: The top-line beat understates the underlying strength because TSMC discloses monthly revenue — the April-May-June cadence was visible in real-time, so the aggregate print was largely locked in. The more important signal is the USD magnitude: revenue grew +17.8% sequentially in USD terms versus +11.3% in NTD, meaning favorable FX translation during the quarter added meaningfully to the reported USD number. But the operational growth is still the story. The +44.4% USD YoY growth versus Q1's +35.3% USD YoY growth is an acceleration, not a deceleration, from already elevated levels.
• Margins: The 58.6% gross margin came in at the top of the 57-59% guide range and 90bps above consensus. Even more striking is the 49.6% operating margin — a +110bps sequential improvement despite overseas fab dilution and the first meaningful Taiwan dollar appreciation of the year. The implied operating leverage is approximately 3-4x: every dollar of incremental revenue is converting to roughly $0.60-$0.70 of incremental operating income. This is the signature of a business where fixed-cost absorption is dominating the cost structure.
• EPS: The +60.7% YoY EPS growth materially outpaced revenue growth of +44.4% — yielding the same 1.4x operational leverage ratio seen in Q1. Net profit margin of 42.7% is a level you would expect from a pure-play IP company, not a capital-intensive foundry. The beat was entirely operational: no tax rate games, no below-the-line items, no share count benefits. Clean.
• FX optics: The USD beat was wider than the NTD beat because the Taiwan dollar weakened during most of Q2 but reversed sharply near quarter-end. Management disclosed an average NT$/US$ rate for the quarter of ~32.1 versus a June 30 close of ~29.3 — a reminder that while the quarter benefited from favorable FX translation, the exit rate is a significant headwind for Q3.

Segment Performance

The AI supply chain continues to pull TSMC's mix in one direction: HPC gets bigger, everything else gets smaller as a share. The 60% HPC share is a new all-time high, and the 8pp YoY expansion is accelerating rather than moderating as we approach a larger base.

HPC: Now Two-Thirds of the Business and Still Growing Faster Than the Rest

On a $30.07B revenue base, HPC revenue was approximately $18.0B in Q2 alone — up roughly ~67% YoY if we assume HPC share of 52% in Q2 2024 on a ~$20.8B base. This is a third consecutive quarter of 60%+ YoY HPC growth, and management's full-year 2025 target of "doubling AI accelerator revenue" remains on track based on H1 momentum. CoWoS remains capacity-constrained through year-end despite having already doubled from 2024 levels.

"HPC platform accounted for 60% of our second-quarter revenue, up from 59% in the first quarter. This reflects the continued robust demand for data center AI, cloud computing, and networking applications. We remain confident in the multi-year growth trajectory of AI." — C.C. Wei, Chairman and CEO

One subtle but important data point from the Q&A: management characterized CoWoS supply-demand as "tight" — a slight upgrade from Q1's "improved slightly." The implication is that the demand side is continuing to expand even as TSMC doubles supply. The binding constraint remains capacity, not demand.

Assessment: HPC is no longer a single segment within TSMC — it is the business. At 60% of revenue and growing much faster than the rest, the operating leverage into 2026 is extraordinary. The risk is no longer "will AI demand be strong enough"; it is "can TSMC add capacity fast enough without sacrificing margin structure." The current gross margin of 58.6% suggests the answer is yes.

Smartphone: Expected Drag, Not a Concern

Smartphone revenue fell to 27% of the mix in Q2, down one point sequentially and six points YoY. Management's commentary framed this as seasonal (pre-iPhone Q4 ramp) rather than structural, but the multi-quarter trend is clearly secular share loss as HPC grows faster. On an absolute dollar basis, smartphone revenue was approximately $8.1B in Q2, up modestly YoY in USD terms even as the share declines.

Assessment: Smartphone provides ballast. Watch Q3 for the pre-iPhone ramp — any softness in Q3's smartphone pickup would matter, but the Q3 revenue guide of $31.8-33.0B implies the smartphone ramp is progressing normally alongside the continued HPC expansion.

Automotive: Bottoming, Not Breaking

Automotive held at 5% of revenue, the same level as Q1, suggesting the prolonged inventory digestion is stabilizing rather than deepening. No explicit timing guidance on recovery, but the lack of further deterioration is itself a modest positive.

Assessment: Automotive is a non-material swing factor at this point. A return to growth would be a small tailwind; continued stagnation is essentially priced in. Watch for commentary in Q3 on whether auto customers are re-engaging for H1 2026 orders.

Technology Node Mix

N3 at 24% of wafer revenue is the leading-edge workhorse now, combined with a still-dominant N5. Together, these two nodes represent 60% of TSMC's wafer revenue, and their ASP uplift over mature nodes drives the ongoing gross margin expansion. With N2 volume production starting in H2 2025 and A16 in H2 2026, the roadmap ahead is dense with accretive node transitions.

Key Topics & Management Commentary

Overall Management Tone: The defining tonal shift of this print was the move from Q1's measured defense to Q2's assertive confidence. In Q1, C.C. Wei's messaging was anchored around maintaining the full-year outlook through tariff uncertainty — a calibrated posture of reassurance. In Q2, management went further: not just maintaining but raising the guide, projecting confidence in 2026, and explicitly characterizing AI demand as a multi-year, structural trajectory. Wendell Huang remained surgical on the numbers, but his willingness to quantify the FX impact on Q3 margins (180-200bps) signaled a more transparent engagement with sensitive topics. Tariffs were barely mentioned — consistent with a lower-profile policy cycle heading into summer.

1. The FY25 Guide Raise: From Maintain to Beat-and-Raise

The single most important data point in this release was the upward revision of full-year 2025 USD revenue guidance from "close to mid-20s percent" at Q1 to "approximately 30%" at Q2. On a base year (FY24) of roughly $90B, the guide raise moves the implied FY25 revenue from ~$112-115B to ~$117-118B — a ~$3-5B incremental revenue upgrade embedded in the range change. This is not a maintained guide being delivered with upside; it is an explicit upward revision with another half-year of visibility still to be reported.

"Based on our strong first-half performance and continued robust demand for our leading-edge technologies, particularly for AI and HPC applications, we now expect our full-year 2025 revenue to grow approximately 30% in US dollar terms, up from our previous outlook of close to mid-20s percent growth." — Wendell Huang, CFO

The last time TSMC raised a full-year revenue guide mid-year was Q2 2024 (when it raised the 2024 outlook). The historical pattern has been to preserve optionality at the high end and let the run-rate speak for itself. Making the explicit upward revision to 30% is therefore a deliberate signal — management wanted the Street to have the updated framework rather than having it derived from monthly revenue disclosures.

Assessment: The guide raise is the defining event. Everything else in this release — the beat, the margin, the HPC share — can be explained by the quarter's run-rate. But the forward guide raise is a statement about H2 visibility, and specifically about the confidence management has in the demand signals they are seeing from Tier 1 HPC customers through Q4 and into H1 2026. This is the kind of update that typically leads to multi-quarter EPS upgrades.

2. Q3 Gross Margin Guide: Cosmetic FX Headwind, Not Structural

The Q3 gross margin guide of 55.5-57.5% implies a 100-250bps step-down from Q2's 58.6%. On the surface, this looks concerning for a stock that has rallied on margin expansion. But the CFO's explicit attribution was striking in its transparency: ~180-200bps of the decline is attributable to the appreciation of the Taiwan dollar against the US dollar, and the balance is split between electricity cost increases in Taiwan and the continued dilution from overseas fab ramps.

"The sequential decline in our gross margin guidance is primarily driven by three factors: first, the significant appreciation of the New Taiwan dollar against the US dollar, which is estimated to impact gross margin by approximately 180-200 basis points; second, higher electricity costs following recent rate hikes in Taiwan; and third, the continued dilution from our overseas fab ramps." — Wendell Huang, CFO

The math is straightforward: TSMC's cost base is predominantly Taiwan dollar (wages, electricity, local supplier costs), while its revenue is heavily dollar-denominated through export customers. When the NT$ strengthens, the USD-denominated revenue translates into fewer NT$, while the NT$ cost base does not move — margin compresses. Every 1% NT$ appreciation is worth approximately 40bps of gross margin pressure. The NT$ has appreciated roughly 4-5% from Q2 average levels through early July 2025, explaining the 180-200bps impact.

Assessment: This is a cosmetic rather than structural issue. If the Taiwan dollar stabilizes or reverses direction, the margin headwind reverses. Underlying unit economics — ASP, utilization, mix — are all tracking higher, not lower. The bears will point to Q3 margin as evidence of peaking earnings; we see it as a transient FX optic that will unwind as the NT$ normalizes. The more important structural headwind is the overseas fab dilution, which management maintained at 2-3pp for FY25 and 3-4pp in following years — unchanged from Q1.

3. Tariffs: The Absence Was the Story

The tariff discussion in Q1 dominated the call; in Q2, it was conspicuously absent. Wei's commentary was brief and declarative: no change in customer behavior, strong results reflect underlying demand strength, TSMC continues to monitor policy but is focused on execution. No analyst pressed hard on Section 232 outcomes, and management did not volunteer scenario analysis. This tonal shift reflects the broader policy cycle — after the April 2 reciprocal tariffs shock and the April 9 pause, the policy environment through May-July stabilized into a quieter, more bilateral negotiation phase. No specific chip tariffs materialized during Q2.

"We have not seen any change in our customers' behavior due to tariff discussions. Our strong second-quarter performance and raised full-year outlook reflect the underlying strength of AI and HPC demand." — C.C. Wei, Chairman and CEO

Assessment: The Q1 thesis that customer behavior would be unchanged absent specific chip tariffs has been vindicated. TSMC's customers did not pre-buy, did not defer, and did not reroute orders — they bought more advanced silicon. The tariff overhang is still technically unresolved (Section 232 investigation continues), but the probability of a chip-specific punitive regime has likely declined from its April peak. The stock's ~60% rally from April lows reflects this recalibration.

4. CoWoS: Still Capacity-Constrained Despite Doubling

CoWoS has doubled in 2025 versus 2024 and remains fully loaded. Management explicitly committed to further expansion into 2026 and said demand is expected to continue outstripping supply. This is a critical data point for the 2026 AI accelerator outlook: if CoWoS supply will expand again in 2026 and will still be absorbed, the implication is that AI demand is not just holding at 2025 levels but growing in 2026.

"Our CoWoS advanced packaging capacity remains tight throughout 2025, despite our capacity roughly doubling from 2024 levels. We are working diligently with our customers to expand capacity further into 2026 to meet continued strong demand. Demand for CoWoS and related advanced packaging services is expected to continue outstripping our supply into 2026." — C.C. Wei, Chairman and CEO

Assessment: This is the most important forward indicator in the release. CoWoS is the physical bottleneck of the AI supply chain, and management is describing a second year of capacity-constrained demand. The Q1 2026 setup therefore remains favorable: AI accelerator revenue is on track to double in 2025 and is already positioned for meaningful growth in 2026. The "AI demand peaks with Blackwell" bear case looks increasingly difficult to defend against this signal.

5. N2 and A16 Roadmap Remains On Track

N2 remains on schedule for volume production in H2 2025 at both Hsinchu and Kaohsiung fabs. Management highlighted that customer engagement is strong across both smartphone and HPC applications, with N2 tape-out pace expected to exceed both N3 and N5 at their respective early ramps. A16 (backside power delivery) remains on track for H2 2026, targeting HPC applications specifically.

"Our N2 technology is progressing very well. We remain on track for volume production in the second half of 2025 at both our Hsinchu and Kaohsiung sites. Customer engagement is strong across smartphone and HPC applications. We expect N2 adoption to exceed the pace of both N3 and N5 at their respective ramps." — C.C. Wei, Chairman and CEO

Assessment: With Samsung's 3nm yields continuing to underperform commercially and Intel 18A ramping but with limited customer commitments, TSMC's roadmap moat is arguably widening. The N2 tape-out pace running ahead of N3 and N5 is a leading indicator of 2026-2027 revenue composition. The competitive landscape at N2 and below is a one-horse race.

6. Arizona and Overseas Expansion Update

Fab 21 Phase 1 (N4) is in smooth high-volume production with yields comparable to Taiwan. Construction on Fab 21 Phase 2 and Fab 22 is accelerating, and management reaffirmed the plan to produce N2 in Arizona Fab 21 Phase 2. Japan's Kumamoto specialty fab is in mass production; the second Japan fab construction depends on local infrastructure. Dresden specialty fab remains on track for late 2027.

On margins: the FY25 dilution guide of 2-3pp was maintained. The FY26+ dilution of 3-4pp was reiterated. No surprises.

Assessment: The Arizona story has shifted permanently from "high-risk strategic necessity" to "manageable margin headwind absorbed by mix." The dilution is real, but mix shift into HPC and the pricing structure of N2 are more than offsetting it. The political value of Arizona as a tariff hedge has become an unspoken tailwind in the Trump administration's trade framework.

Guidance & Outlook

The guide is a classic beat-and-raise structure: Q2 blew past the high end of the Q1 revenue guide ($30.07B vs. $28.4-29.2B range), and the forward framework has been upgraded meaningfully. The Q3 revenue guide midpoint of $32.4B represents ~+8% QoQ growth from Q2's $30.07B, and implies YoY growth of approximately +38% — a slight deceleration from Q2's +44% but still one of the strongest growth rates TSMC has ever delivered in a fiscal Q3.

The FY25 guide raise to "approximately 30%" from "close to mid-20s" is the more consequential change. If we assume FY24 USD revenue of ~$90B, the 30% growth target implies FY25 revenue of approximately $117-118B — up from the ~$112-115B implied by the Q1 outlook. Through H1, TSMC has delivered ~$55.6B in USD revenue; the implied H2 requirement is ~$62-63B, or an average of $31-32B per quarter. The Q3 guide midpoint of $32.4B is already above this run-rate, and Q4's seasonal ramp (pre-iPhone launch cycle) typically delivers above-trend revenue. The math on the raised guide is conservative, not aggressive.

Implied Q4 math: To hit a 30% FY25 growth target from a ~$90B FY24 base, Q4 needs to deliver approximately $29.5-30.5B — actually a slight sequential step-down from the Q3 guide midpoint. This is consistent with smartphone seasonality normalizing post-iPhone ramp and FX normalization.

Gross margin setup: The Q3 gross margin guide of 55.5-57.5% is a ~180-200bps step-down from Q2's 58.6%. Management's explicit FX attribution (180-200bps) means that, on an FX-neutral basis, the Q3 gross margin would be approximately flat with Q2's level. The 2-3pp full-year dilution from overseas fabs continues to be maintained. Underlying unit economics — mix, utilization, yield — are unchanged or improving.

Guidance style: Historically conservative, as noted in our Q1 report. TSMC has now beaten its own EPS guide in six consecutive quarters, so the FY25 "approximately 30%" framing likely contains another 100-300bps of upside if AI demand remains on track through H2. The bears will argue the guide now fully reflects the strong year and limits forward upside; we think that reads the conservative history backwards. Upside optionality is still in the setup.

Analyst Q&A Highlights

Quality of the FY25 Guide Raise

• Morgan Stanley: Probed whether the ~30% guide reflects organic AI demand vs. customer pull-forward. Management characterized customer orders as steady through the quarter, with demand visibility extending into Q4 2025 and H1 2026; no indication of inventory building.
  Assessment: This is the right answer. Customer pull-forward would show up as elevated bookings earlier in the quarter followed by deceleration; TSMC's monthly data showed the opposite pattern (April strongest; June somewhat softer on base effects). The guide raise is supported by forward order visibility, not inventory stocking.

Q3 Gross Margin and FX Sensitivity

• Bernstein: Asked about FX sensitivity and whether the ~180-200bps gross margin headwind is a durable feature. CFO Huang confirmed the 40bps per 1% NT$ appreciation ratio, and noted that if NT$ reverses, the margin outlook improves.
  Assessment: This answer is the single most important bear-case defense in the release. If the Q3 gross margin compression is primarily FX-driven rather than unit-economics-driven, then the "margin peaking" narrative is wrong. Watch the NT$/US$ rate through Q3 for real-time read on margin trajectory.

2026 AI Accelerator Trajectory

• Bernstein: Asked about the 2026 AI accelerator growth rate given the high 2025 base. CEO Wei declined to give specific 2026 guidance but pointed to sovereign AI, enterprise AI, and continued hyperscaler capex through 2026, and explicitly stated that 2026 would be "another strong growth year for HPC."
  Assessment: "Another strong growth year" is a meaningful forward statement; it rules out deceleration scenarios and implies mid-teens to high-teens growth at minimum. This is consistent with CoWoS capacity expanding again in 2026 and remaining fully loaded.

CoWoS 2026 Supply-Demand

• Goldman Sachs: Asked about 2026 CoWoS capacity targets and the probability of oversupply. CEO Wei said CoWoS will expand further through 2026 and he did not expect "meaningful oversupply."
  Assessment: The fact that Wei is willing to rule out 2026 oversupply while declining to give specific capacity numbers suggests the demand visibility is strong enough to sustain another year of capacity-constrained growth. This is bullish for AI accelerator revenue through 2026.

Arizona Dilution Trajectory

• JPMorgan: Pressed on gross margin dilution from Arizona Fab 21 Phase 2 and Fab 22 ramps. CFO Huang reaffirmed 2-3pp FY25 dilution, 3-4pp in subsequent years.
  Assessment: Unchanged from Q1 — no new information. The dilution is absorbed by mix shift; the concern is whether mix continues to shift favorably if HPC growth decelerates. So far, no sign of deceleration.

Tariff Scenario Planning

• Multiple analysts: Asked repeatedly about Section 232, Commerce Department investigation, and customer behavior. CEO Wei's consistent response: no change in customer behavior, maintained guide, declined scenario analysis.
  Assessment: The tariff discussion was notably shorter than Q1. Wei's deflection is consistent; the lower frequency of tariff questions reflects a quieter policy environment. The absence of scenario planning is defensible but leaves an information gap.

Competitive Position at N2

• UBS: Asked about Samsung's 3nm struggles and Intel 18A progress. CEO Wei declined to comment on competitors, preserving neutral-supplier posture; he affirmed TSMC's own N2 competitive position.
  Assessment: The answer TSMC has given on competitors for a decade. The relevant signal is the tape-out pace on N2, which management said exceeds both N3 and N5 at the equivalent stage — a concrete bullish data point that substitutes for direct competitor commentary.

2026 Capex

• Citi: Asked whether 2026 capex will exceed 2025's $38-42B. CFO Huang declined to guide 2026 specifically but implied continued high investment to support demand, with formal guidance coming in January 2026.
  Assessment: "Continued high investment" is the right signal given the demand picture. We'd be concerned if management suggested capex would flatten; a flat capex trajectory would imply demand concerns. The openness to higher 2026 capex is bullish.

What They're NOT Saying

1. Specific 2026 revenue framework: Management confirmed 2026 will be "another strong growth year" but declined to quantify. For a company trading at a re-rated multiple post-April bottom, offering no 2026 framework leaves investors to interpolate from the 2025 run rate. Historical precedent is that TSMC gives preliminary 2026 views in October 2025 and formal guide in January 2026 — so the absence is procedural, not strategic.
2. China revenue decline: China share dropped to 9% in Q2 from 11% in Q1 and 16% a year ago. Management did not discuss the trajectory; given ongoing advanced-node export controls and the Nvidia H20 situation, the China decline is self-explanatory. But the structural point is that TSMC's "China exposure" has halved in 12 months while the business has grown 44% YoY in USD — the narrative of TSMC as a China-risk stock is increasingly stale.
3. Specific Section 232 scenario planning: Unchanged from Q1 — management deflects. The investigation is still formally ongoing. Given the policy environment has calmed, this is a fair posture, but it remains the largest binary tail risk for the stock.
4. Pricing actions on N2: Management said customer engagement is strong and tape-out pace exceeds N3/N5, but did not disclose N2 pricing or whether wafer ASPs represent a step-function uplift. History suggests N2 ASPs will be materially higher than N3 to reflect capex intensity, but management declined to quantify.
5. Buyback authorization: No discussion of expanded share repurchases despite the strong cash generation. Given capex intensity through 2026-2027, a material buyback is unlikely, but the absence of even a modest program continues to frustrate investors comparing TSMC's capital return to peer averages.
6. Specific dividend growth framework: Wendell Huang reiterated commitment to "steady dividend growth" but offered no forward framework. Given cash generation of ~$60B+ at current earnings power, a more aggressive dividend policy would be well-received but is not being signaled.
7. Samsung or Intel disruption risk: Wei would not speak to competitors. Given Samsung's 3nm yield issues and Intel 18A's limited commercial traction, TSMC's competitive position at leading edge is arguably strongest it has been in years — but management will not say so directly, preserving the neutral-supplier posture.

Market Reaction

• Pre-earnings ADR (July 16, 2025): ~$236-239, up ~60% from April lows
• Pre-market July 17: +4-5% on the beat and guide raise
• Open July 17 (NYSE): ~+4%
• Intraday high July 17: +5-6% on initial reaction to FY25 guide raise
• Close July 17, 2025: +3.4% at approximately $245-247
• Volume: ~1.8-2.0x average daily volume
• Pre-market July 18: Flat to slightly down on Q3 margin guide concerns
• Analyst actions within 24 hours:
  • Broad reiteration of Buy/Overweight ratings
  • Price target revisions clustered in the $260-$285 range, up from prior $225-$265
  • No downgrades reported within 24 hours
  • Morgan Stanley, Goldman Sachs, Barclays, Bernstein, Needham, Citi, UBS, JPMorgan all raised price targets
  • FY25 EPS estimates moving +5-8% on the guide raise

The +3-4% reaction is subdued relative to the operational magnitude of the beat-and-raise. This reflects three dynamics: (1) the stock had already rallied ~60% from April lows heading into the print, absorbing a significant portion of the "AI demand continues" thesis; (2) TSMC's monthly revenue disclosures meant the top-line was largely pre-positioned, so only the FY25 guide raise and operating margin expansion were net-new information; (3) the Q3 gross margin guide of 55.5-57.5% provided near-term bears with a convenient narrative to take profits, even though the FX-driven nature of the compression is clearly explained.

The more informative signal is the analyst community's reaction: broad price target increases, no downgrades, and FY25/26 estimate revisions moving higher. When price targets advance while the stock grinds sideways, the gap tends to resolve in favor of fundamentals over the subsequent 6-12 months — this is the setup we think patient investors should lean into.

Street Perspective

Debate: Quality of the FY25 Guide Raise

Bull view: The raise from "mid-20s" to "approximately 30%" is the first explicit upward revision of the year, and it reflects management confidence in H2 demand durability. At a ~$90B FY24 base, the implied ~$3-5B incremental revenue upgrade will flow through to FY25 and FY26 EPS via sustained operating leverage. This is not FX-driven; it is genuine demand acceleration.

Bear view: The guide raise partially reflects favorable FX translation (the Q2 average NT$ was ~32, weaker than start-of-year). If the NT$ appreciates further into H2, some of the guide raise reverses on translation. The bears argue the organic component of the raise is narrower than it appears.

Our take: The bull view has the stronger footing. The raise is primarily AI-driven: HPC share expanded to 60%, CoWoS remains capacity-constrained, and management's commentary about 2026 being another strong growth year implies the order book is building through H2 2025 and into H1 2026. FX adds noise to quarterly cadence but does not change the structural trajectory. The raise is real.

Debate: Q3 Gross Margin — Peak or Pause?

Bull view: Q3's guide of 55.5-57.5% reflects 180-200bps of FX headwind that will reverse as NT$ normalizes. Underlying unit economics — mix, yield, utilization, ASPs — continue to improve. The margin profile will return to the 58%+ range as FX stabilizes, and the FY25 guide for 2-3pp overseas fab dilution remains intact.

Bear view: The Q3 guide marks the end of the margin expansion story. Between NT$ appreciation, electricity rate hikes, and accelerating overseas fab dilution, gross margins are peaking at ~58%. The structural ceiling on gross margin is lower than the bulls think, and the multiple needs to compress as the margin expansion story matures.

Our take: The bear view confuses a cyclical FX optic with a structural change. TSMC's underlying margin trajectory is determined by mix (HPC share), yield (advanced nodes at maturity), and utilization (consistently high). None of these are deteriorating; they are all improving. If NT$ stabilizes, Q4 gross margins should recover. The Q3 guide is a pause, not a peak.

Debate: 2026 AI Accelerator Growth

Bull view: CoWoS capacity will expand again in 2026 and management expects demand to continue outstripping supply. AI accelerator revenue doubling in 2025 is on track, and "another strong growth year" in 2026 implies at least mid-teens growth in AI accelerator revenue. The demand signal is structural and multi-year.

Bear view: 2025's doubling is unrepeatable. Hyperscaler capex is at or near peak, sovereign AI is overhyped, and the inference workload expansion has not yet materialized at enterprise scale. The 2026 growth rate will be dramatically lower and will trigger multiple compression.

Our take: Bulls win this debate in 2025 and likely 2026; bears have the better odds in 2027. The CoWoS expansion into 2026 combined with Blackwell, the hyperscaler ASIC wave, and Apple's M-series scaling all imply a strong 2026. The more interesting question is 2027, when the base gets higher and inference-driven demand must pick up the slack from training. That is a 2H 2026 concern, not a 2025-2026 concern.

Debate: Valuation — Still Attractive After the Rally?

Bull view: At ~$245 and FY25 EPS estimates moving toward $9.40-$9.60, TSM trades at ~25x forward P/E — reasonable for a business growing revenue 30% and earnings 45%+ with widening process-technology leadership. The multiple is below historical average and well below peer averages for companies with similar growth profiles.

Bear view: The stock has doubled from April's lows. Much of the good news is priced in. At ~25x forward P/E, there is limited room for multiple expansion. Any growth deceleration or margin compression triggers downside; the asymmetry has flipped from April's setup.

Our take: Valuation has become fair but not expensive. The April entry was asymmetric; today's entry is still attractive but requires conviction that 2026 holds up. We think it does — the CoWoS capacity signal and the continued mix shift to HPC should support another year of earnings growth in the 20-30% range. A 25-27x forward multiple on FY26 EPS of $11-12 implies $275-325, or 12-33% upside. The bear case (margin peaking, growth decelerating) would imply 18-20x on FY26 EPS of $10 = $180-200, or 18-26% downside. The risk/reward has compressed versus April but remains attractive at ~1.5:1 upside to downside.

Debate: Tariff Tail Risk — Contained or Dormant?

Bull view: The Q2 results prove that customer behavior is unchanged; the Section 232 investigation is dragging, and the policy environment has quieted since April. The probability of a punitive chip-specific tariff regime has declined, and the Arizona capacity expansion provides additional political insulation.

Bear view: Tail risk is inherently hard to quantify, and the 2026 election cycle could re-activate policy brinksmanship. Section 232 remains unresolved, and a sudden chip tariff announcement could still compress the multiple. The April 2025 sell-off showed how quickly sentiment can turn.

Our take: Contained, but not eliminated. The probability of severe tariff escalation has declined meaningfully from April peaks but is not zero. The stock should carry a modest geopolitical risk discount, which it does. We don't see this as a primary driver of rating decisions absent new policy catalysts.

Model Update & Valuation Framework

Valuation framework: At a ~$245 ADR price, TSM trades at approximately 25-26x our FY25 EPS estimate of $9.40-9.80 and 20-22x our FY26 EPS estimate of $11-12. For a business growing revenue 30% and earnings 45%+ in 2025, and that will likely grow 18-22% and 15-25% in 2026, this multiple is fair but not stretched. A 24-26x multiple on FY26 EPS of $11-12 implies a 12-month price target range of $264-$312 — upside of 8-27% from current levels. A more bullish re-rating to 28x on $12 implies $336 — upside of 37%.

Bull-case scenario: AI accelerator revenue doubles in 2025 and grows another 30% in 2026; gross margin returns to 58-60% as FX stabilizes; Q4 2025 delivers above the implied $29-30B run rate. FY25 EPS lands at $9.80-$10.00, FY26 EPS at $12.50-$13.00. Target multiple of 25x implies $312-$325 — upside of ~30%.

Bear-case scenario: Taiwan dollar appreciates further, pushing gross margins to 54-56% for H2 2025; electricity costs and Arizona dilution combine to push FY26 margins below 55%; AI accelerator growth moderates to ~20% in 2026. FY26 EPS lands at $9.50-10.00; a 20x multiple implies $190-200 — downside of ~20%.

Risk-reward calculus: Upside of ~25-30% versus downside of ~18-22% implies a favorable asymmetry of roughly 1.4-1.5x, down from Q1's 2-3x but still attractive. The entry point is less extraordinary than Q1, but the fundamental thesis has strengthened.

Thesis Scorecard: Q1 Watch Items Revisited

Our Q1 2025 initiation report laid out six specific signposts for Q2. Every single one tripped favorably.

Thesis Scorecard: Original Bull/Bear Points

Overall: Thesis meaningfully strengthened coming out of Q2. Every bull point either confirmed or strengthened; two bears weakened (tariff, Arizona), two remain static (concentration, Taiwan), one remains open (Section 232), and one new bear emerged (FX). Net-net, the operational thesis is stronger than at Q1 initiation. The new FX bear is cosmetic and reversible. The stock's re-rating from April lows reflects some of this improvement but not all of it.

Action: Maintaining Outperform. The thesis has strengthened materially even as the entry point has become less extraordinary. We see roughly 25-30% upside to a $310-320 target over 12 months versus 18-22% downside, implying a favorable risk/reward. For investors already long, this is a quarter to hold through; for new money, the entry is still reasonable.

Bottom Line

TSMC just delivered the kind of earnings print that answers, rather than raises, questions. Every operational question we flagged at Q1 — can HPC mix keep expanding? can margins hold through overseas fab dilution? can the FY25 guide be maintained through tariff uncertainty? can CoWoS capacity scale? — was answered affirmatively, and the FY25 guide was raised for good measure. The operational thesis is now stronger than at initiation: HPC at 60%, AI accelerator revenue on track to double, operating margin expanding, CoWoS capacity committed for 2026 expansion, and N2 ramping on schedule with industry-leading tape-out pace.

The only new wrinkle is the Q3 gross margin guide, which compresses 100-250bps from Q2's 58.6% level. Management's explicit FX attribution (180-200bps of the compression) makes clear that this is a Taiwan dollar appreciation story, not a structural cost story. If NT$ stabilizes, Q4 margins recover; if it appreciates further, the drag extends. Either way, the underlying unit economics are strengthening.

The stock entered the print up ~60% from April lows and absorbed a modest +3-4% reaction. That muted response is consistent with the monthly revenue disclosures having pre-priced most of the top-line beat — the incremental information is the FY25 guide raise and the margin expansion, both of which feed FY26 EPS upgrades that will play out over the next several quarters. Sell-side price targets are advancing into the $260-285 range, and the gap between price and targets tends to resolve over 6-12 months. We maintain our Outperform rating.

What could change our mind:

• Downgrade to Hold: Evidence that the 2026 AI accelerator growth rate moderates meaningfully (sub-10% growth) on hyperscaler capex cuts; or Q3/Q4 gross margin falls below 55% on FX-plus-structural factors; or a Section 232 outcome that triggers 15%+ chip-specific tariffs with limited pass-through.
• Downgrade to Underperform: A forced Intel JV or foundry partnership mandated by the Trump administration; direct Taiwan-strait escalation with operational impact; or gross margin compression below 52% on sustained NT$ strength plus Arizona ramp acceleration.

Signposts for Q3 2025 earnings (October 2025):