The AI Foundry Is Earning Through the Tariff Storm: Initiating Coverage of TSMC at Outperform

Results vs. Consensus

Quality of Beat / Miss

• Revenue: The +0.5% top-line beat understates what actually happened. TSMC was always going to land in its guided range thanks to transparent monthly revenue disclosures, but the mix — HPC exploding to 59% of revenue from 46% a year ago, 3nm holding 22% share despite the smartphone seasonal drag — is the real story. Growth was organic, with no M&A, no deferred revenue gimmicks, and no FX tailwind (USD revenue grew +35.3% while NTD grew +41.6%, meaning FX was actually a headwind). This is clean, volume-driven growth in the highest-margin nodes.
• Margins: The 58.8% gross margin beat is the number to watch. Management had guided 57-59%, so the high end delivery means operating leverage continued through smartphone seasonality rather than compressing with it. Operating margin of 48.5% came in at the top of the 47-49% range. This level of margin on a sequentially lower revenue quarter tells you the fixed-cost absorption on advanced-node capacity is dominating the mix. The overseas fab dilution overhang (flagged at 2-3% for 2025) is real but not yet visible.
• EPS: The +60.3% YoY earnings growth versus +41.6% revenue growth is the clearest possible evidence of operating leverage in a foundry business, which historically prices for modest scale benefits. Net profit margin of 43.1% is a level we'd expect to see in a pure-play IP company, not a capital-intensive foundry. The beat was overwhelmingly operational, not tax or share count.

Segment Performance

TSMC reports revenue by technology node, platform, and geography. The platform cut is the most important lens — it shows where the dollars are coming from and whether the AI narrative is actually showing up in the numbers (it is, emphatically).

HPC: The New Center of Gravity

A 13-percentage-point re-mix in 12 months is an extraordinary data point in a business this size. On a $25.5B revenue base, HPC revenue was roughly $15.1B in Q1 alone — up approximately ~75% YoY if we assume HPC shares of ~46% in Q1 2024 on ~$19B base. This is the single clearest indicator that the AI buildout is hitting the foundry layer at scale, not just in Nvidia's downstream pricing. CoWoS advanced packaging capacity remains fully loaded and management reiterated plans to double capacity in 2025.

"Previously, CoWoS demand was much higher than supply — the situation has improved slightly." — C.C. Wei, CEO/Chairman

The subtle "improved slightly" qualifier matters: it does not mean demand is weakening, but rather that the supply side is finally catching up after a year of crisis-level constraint. AI accelerator revenue (GPUs, TPUs, ASICs, HBM controllers) is expected to roughly double YoY in 2025.

Assessment: HPC is structurally the highest-margin platform in the portfolio thanks to its advanced-node mix (most HPC is on 5nm and 3nm). As it displaces smartphone in the mix, the natural drift in corporate gross margin is upward. The question is not whether HPC is growing but whether its growth slows to 40% or 50% or stays at 60%+ in 2026 — a rich-man's problem.

Smartphone: The Expected Drag

Smartphone revenue fell back to 28% of the mix in Q1 as the iPhone and flagship Android seasonal ramp completed in Q4. Management flagged this upfront: "Our business in the first quarter was impacted by smartphone seasonality, partially offset by continued growth in AI-related demand." The read is straightforward — flagship smartphone demand is not collapsing, but the secular share loss to HPC is real. On a unit basis, smartphone volumes are growing modestly; on a revenue basis, they are being crowded out by the hyperscale AI build.

Assessment: Smartphone is a slow grower that provides ballast and seasonal counter-cyclicality to HPC. The Q1 seasonal dip is fine. Watch Q3 for the pre-iPhone ramp — any softness there would matter more than this Q1 print.

Automotive: Digesting, Not Dying

Automotive grew sequentially from 4% to 5% of the mix. The industry-wide auto semiconductor inventory digestion that dogged 2024 is slowly clearing, but the recovery remains tepid. No mention of a step-function in this area.

Technology Node Mix

The 3nm ramp from 9% to 22% of wafer revenue in 12 months is the operational story of the quarter. N3 is now a genuine scale node, not a "tip-of-the-spear" niche. Combined with a still-dominant N5, this means 58% of TSMC's wafer revenue is now on nodes where it has no real competition. Samsung's 3nm yields remain commercially uncompetitive; Intel 18A is targeting H2 2025 for limited production but has yet to demonstrate customer wins at scale.

Key Topics & Management Commentary

Overall Management Tone: Measured and confidence-signaling without triumphalism. C.C. Wei repeatedly acknowledged tariff uncertainty in the same breath as maintaining the full-year outlook — a calibrated tone designed to inoculate against criticism while still reassuring customers. Wendell Huang stayed surgical on the numbers. The call had none of the defensiveness that typically accompanies a company under geopolitical pressure; management projected the demeanor of an entity that believes the secular trend is strong enough to absorb the political noise.

1. The AI Demand Signal Held

The most important question heading into this print was whether the explosive HPC growth of 2024 would decelerate in 2025 as hyperscale capex guidance moderated and as Nvidia's H20 China export ban (announced April 15, two days before the call) threatened a direct demand hit. The answer was unambiguous: HPC share of revenue expanded a further 6 percentage points sequentially (53% to 59%), AI accelerator revenue is on track to double in 2025, and CoWoS remains capacity-constrained despite a planned capacity doubling.

"AI reasoning models including DeepSeek will drive greater efficiency and lower the barrier to future AI development." — C.C. Wei, CEO/Chairman

This DeepSeek-adjacent framing is important. After the January DeepSeek shock, a thesis emerged that more efficient models would reduce aggregate compute demand. Wei is explicitly pushing back: lower-barrier AI means more models, more inference, more customers, and ultimately more silicon.

Assessment: The single most important data point in this release is that TSMC's Q2 guide of US$28.4-29.2B implies +13% QoQ acceleration from a Q1 base that was already +35% YoY. That is not a company seeing demand weakness anywhere in the AI stack. If anything, the supply-side is the binding constraint, not demand.

2. Tariffs: The "Private Company" Defense

C.C. Wei's positioning on tariffs was notable for its structural elegance. He separated the geopolitics ("discussions between countries") from TSMC's operational reality ("no change in customer behavior"), and then anchored that position to the numeric outlook ("continue to expect full-year 2025 revenue to increase by close to mid-20s percent in US dollar terms"). This maintained commitment is particularly striking given:

• April 2 Trump "reciprocal tariffs" announcement that initially roiled the Taiex down ~10%
• Ongoing Section 232 semiconductor investigation that could produce a specific chip tariff regime
• The April 15 disclosure by Nvidia that its H20 chips — which run on TSMC wafers — now require export licenses for China, with Nvidia taking a $5.5B charge

"We understand there are uncertainties and risk from the potential impact of tariff policies. However, we have not seen any change in our customers behavior so far, therefore, we continue to expect our full year 2025 revenue to increase by close to mid-20s percent in US dollar terms." — C.C. Wei, CEO/Chairman

Wei further characterized tariff discussions as being "between countries" and noted that TSMC, as a private company, is "not getting involved." This is a textbook neutral-actor posture, the kind of messaging that plays equally well to customers worried about supply disruption, Taiwanese regulators, and a Trump administration that has cited TSMC's Arizona expansion as a personal win.

Assessment: Management's tariff commentary is carefully calibrated but contains a real signal: they are not yet seeing customer pre-buying or order cancellations. Either this is true (in which case the tariff risk is overstated in the stock price), or customers are frozen in a wait-and-see mode that will eventually rupture in one direction. We lean toward the former because the underlying demand is structural AI buildout, which operates on multi-year contracts rather than quarter-to-quarter order books.

3. Arizona: $165B Becomes a Hedge, Not an Overhang

In March 2025, TSMC announced an incremental $100B of Arizona investment on top of the original $65B commitment, bringing the total to $165B. The plan now includes three additional wafer fabs, two packaging/testing plants, and an R&D center. Management reiterated on this call that construction of Fab 2 and Fab 3 is accelerating, though labor shortages in Arizona continue to complicate timelines. TSMC is targeting roughly 30% of its 2nm capacity in Arizona.

"Geographic manufacturing flexibility is an important part of our value proposition to the customers." — Wendell Huang, CFO

The pivot here is subtle but important. Twelve months ago, Arizona was an expensive necessity — a lower-yield, higher-cost hedge against geopolitical tail risk, with the CFO reluctantly owning a 2-3% gross margin dilution. Today, Arizona is being framed as a strategic value proposition for US customers and as a political asset under the Trump administration. The margin dilution math hasn't changed, but the optionality has improved materially.

Assessment: Arizona is now a dual-purpose asset: it protects TSMC against tariff escalation and provides political cover in Washington. The $165B is large but manageable against a 2025 capex budget of $38-42B and a cash-generation profile of roughly $50B+ in operating cash flow at current earnings power. The margin dilution is the real cost, but it is being absorbed by mix shift into higher-margin HPC.

4. CoWoS Capacity Doubling

CoWoS (Chip-on-Wafer-on-Substrate) advanced packaging is the bottleneck in the AI accelerator supply chain. Every Nvidia H100, H200, B100/B200, and GB200 requires CoWoS, as do competing accelerators from AMD, Broadcom (for hyperscaler ASICs), and Marvell. CoWoS has been TSMC's most constrained resource for the past 18 months, with demand multiples of supply in 2024.

Management confirmed that CoWoS capacity is still fully loaded and is on track to double in 2025. This implies a CoWoS unit exit rate entering 2026 materially higher than the current annualized run rate, with all of that incremental output already spoken for by customers.

Assessment: CoWoS capacity is the single most important leading indicator for AI accelerator revenue. Doubling it is the highest-confidence growth lever for 2025-2026. The fact that the new capacity is already contracted suggests Nvidia's Blackwell ramp and the hyperscaler ASIC wave are operating at full throttle through end-2025 at least.

5. Intel Foundry JV Denial

Speculation had built up through March and early April that the Trump administration might broker a TSMC-Intel arrangement to rescue Intel's foundry ambitions — with TSMC as either a minority investor, technology licensor, or operational partner. Wei's denial was unambiguous.

"TSMC is not engaged in any discussion with other companies regarding any joint venture, technology licensing, or technology transfer." — C.C. Wei, CEO/Chairman

Assessment: This is the right answer. A partnership with Intel would be dilutive to TSMC's process-technology moat and would cannibalize revenue from existing fabless customers who depend on TSMC's neutrality. Wei's direct denial removes an overhang and confirms TSMC will continue to compete rather than cooperate with Intel Foundry.

6. N2 Ramp and A16 on Schedule

N2 (2nm) remains on track for H2 2025 volume production at Hsinchu and Kaohsiung sites. A16 (TSMC's backside power delivery node) is targeted for 2026. These timelines have not moved. At a time when Samsung's 3nm is still commercially underwhelming and Intel 18A remains an unproven commercial proposition, TSMC's node roadmap is arguably extending its lead rather than narrowing it.

Assessment: The combination of a proven N3 ramp in 2024, on-schedule N2 in H2 2025, and A16 in 2026 means TSMC's process-technology leadership is secure for at least the next 24 months. The Arizona expansion aligned with N2 also means US-manufactured leading-edge silicon is finally becoming a reality.

Guidance & Outlook

The guide is the star of the release. Q2 revenue at the US$28.4-29.2B range implies +13% QoQ acceleration from Q1's base, which was already running +35% YoY. At the midpoint of US$28.8B, Q2 YoY growth would be ~38-40%, actually accelerating from Q1's 35.3% YoY. The only way this math works is if HPC is growing at 60%+ YoY and outpacing the decelerating smartphone comp. Management is effectively telling the Street: "We see enough demand to confidently call for 40%-range growth in Q2 despite the tariff headlines."

The full-year guide of "close to mid-20s percent" USD revenue growth is being maintained — which was not a foregone conclusion. Given early April tariff shock, a less-confident management team would have trimmed the full-year guide to a "low-20s" range to create room. They didn't. This maintained guide is the core of the bullish call: management is signaling confidence that the full-year demand picture is intact.

Implied H2 trajectory: For FY25 revenue to grow "close to mid-24s" percent on a full-year basis off a FY24 base of ~US$90B (estimated), total FY25 revenue should land around US$112-115B. H1 revenue at the Q2 guide midpoint puts TSMC at ~US$54.3B through the first half, implying H2 revenue of US$58-61B, or ~8-12% H2-over-H1 growth. This is fully achievable if AI demand remains on pace and smartphone normalizes in Q3/Q4.

Gross margin setup: The Q2 gross margin guide of 57-59% is flat with Q1's 58.8% at the high end. The full-year gross margin framework is complicated by overseas fab dilution (Arizona Fab 21 Phase 1 now in volume production, with the associated ~2-3% dilution landing through 2025). This is a headwind to watch but not yet materially visible in the consolidated numbers.

Guidance style: Historically conservative. TSMC has beaten its own revenue guidance in each of the past four quarters and the top end of the gross margin range in most of them. The maintained full-year guide therefore likely has real upside, particularly if CoWoS capacity ramps ahead of schedule and if the tariff scenario resolves into a narrower outcome than the worst-case Section 232 result.

Analyst Q&A Highlights

Tariff Impact and Customer Behavior

• Multiple analysts pressed on whether customers were pre-buying inventory ahead of potential tariff implementation, an obvious concern given the April 2 reciprocal tariffs announcement. Wei answered directly that TSMC has not observed changes in customer behavior or advance purchasing. He repeated the maintained full-year guide as evidence.
  Assessment: The absence of visible pre-buying can be read two ways. Either demand is genuinely structural (our read), or customers are frozen and waiting for clarity — a posture that could rupture in either direction. The fact that Wei is willing to maintain the full-year guide suggests his channel checks do not indicate a sudden demand cliff.

Gross Margin Trajectory and Overseas Dilution

• Analysts probed CFO Wendell Huang on the shape of gross margin through 2025 given Arizona Fab 21 Phase 1 now producing commercial wafers. Huang reiterated the 2-3% dilution expectation for 2025 and 3-4% in later years as more global capacity ramps.
  Assessment: The dilution is real but should be more than offset by the mix shift into HPC and 3nm. Watch Q3 gross margin carefully — if it holds at 58%+, the dilution concern is more than offset; if it drops toward 55%, the Arizona math becomes a legitimate issue.

AI Demand Durability

• Analysts asked about sustainability of AI accelerator demand into 2026, given hyperscaler capex comments and the DeepSeek efficiency debate. Wei was categorical that AI-related demand remains the strongest growth driver and framed lower-cost reasoning models as demand-expanding rather than demand-cannibalizing.
  Assessment: Wei's framing is the right one. The history of compute is that efficiency gains have been absorbed by exponentially expanding use cases. Jevons paradox applies to AI as cleanly as it applied to steam engines and semiconductors historically.

Intel Joint Venture Speculation

• Several analysts questioned Wei on whether TSMC was in discussions with Intel about a foundry partnership or JV. Wei's denial was immediate and unambiguous: no discussions of any kind on JV, technology licensing, or transfer.
  Assessment: The market needed a clear answer; Wei provided one. This removes an overhang but does not preclude future pressure from the Trump administration. A forced partnership remains a low-probability, high-impact tail risk.

N2 Ramp and 2nm Economics

• Analysts asked about N2 pricing, yield progress, and whether early N2 wafers would be margin-accretive or dilutive. Management was less specific on pricing but maintained the H2 2025 volume production timeline and indicated yields are tracking in-line with expectations.
  Assessment: TSMC has consistently executed on node ramps; N2 is unlikely to be the first to disappoint. The more interesting question is whether N2 ASPs will be a step-function above N3, reflecting the capex intensity of 2nm. Early indications are yes.

What They're NOT Saying

1. China revenue trajectory: Geography share to China was 7% in Q1, and the mix has been quietly declining for multiple quarters under advanced-node export controls. Management did not break out China-specific trends or quantify the H20-related revenue at risk. Given Nvidia's $5.5B charge on H20, the TSMC wafer-level exposure is real but unquantified.
2. 2025 capex flexibility under tariff escalation: The $38-42B capex range was maintained, but management did not address whether a more severe tariff outcome could trigger capex cuts. In prior downturns (2019, 2022), TSMC trimmed capex at the high end — the current environment could warrant similar flexibility that is not being signaled.
3. Automotive recovery timing: Auto was flagged as still digesting inventory, but no explicit call on when the segment returns to YoY growth. Given the platform was 6% of Q1 2024 and is 5% of Q1 2025, the implied decline is ~17% YoY — notable but buried in the HPC narrative.
4. Pricing actions: Historically TSMC has commentary around annual pricing cycles, especially after cost inflation. The Q1 call was light on specific ASP commentary, particularly around N3 and N2 pricing tiers. This may reflect management sensitivity to being seen as "gouging" in a geopolitical moment, but it leaves an information gap.
5. Specific Section 232 scenario planning: Wei deflected tariff questions with a consistent "watch and see" posture, but did not share any internal scenario analysis. Given the stock has been down 23% YTD largely on tariff fears, more transparency on scenario ranges would be investor-friendly. Its absence is defensible politically but unhelpful for modeling.
6. Samsung competitive dynamics on N2: With Samsung's 3nm struggles well-known, TSMC had an opportunity to comment on its N2 competitive position. Wei did not, preserving the neutral-supplier posture that has served TSMC well.

Market Reaction

• Premarket move (April 17): +2% to +3.5% in US premarket trading after the beat and maintained guide.
• Pre-earnings YTD performance: Stock down approximately 23% YTD entering the print, significantly worse than S&P 500's ~-10% YTD at that point.
• Pre-earnings setup: Stock had slumped ~10% the week of April 7 on Trump reciprocal tariff announcement; partial recovery after 90-day tariff pause announcement; traded around the low-$150s entering the print.
• Analyst actions within 48 hours:
  • Overweight/Buy ratings broadly maintained, reflecting Street consensus that Q1 was a win.
  • Price targets moved in mixed directions: some firms trimmed Taiwan-listed PTs on tariff-related valuation re-rating, while others maintained or reiterated higher ADR targets, citing the maintained full-year guide.
  • Bernstein maintained Outperform with a $251 ADR price target — a call reflecting the "tariff truce + AI tailwind" framework.

The muted pre-market reaction (despite a clear beat) reflects two things: first, the expectation bar was high after a strong Q4 and transparent monthly revenue disclosures — the beat was incremental, not a shock. Second, the market is trading TSM as a political asset right now more than an earnings asset. The question on investors' minds is not "did Q1 numbers beat" but "will Section 232 tariffs crush the multiple further." Until the tariff investigation concludes, multiple expansion on fundamentals will be capped.

Street Perspective

Debate: Is the tariff risk largely priced in, or is there another 20% of downside?

Bull view: Stock is already down 23% YTD, trading at a low-teens forward P/E on what could be a 60%+ earnings growth year. The April 17 print shows demand is durable, management is maintaining guidance, and customer behavior has not changed. Even a relatively punitive Section 232 outcome (say, 15-25% semiconductor-specific tariffs) is mostly absorbed by customers (Nvidia, Apple, AMD) rather than by TSMC, and pricing pass-through would be normal industry practice. The de-rating has overshot.

Bear view: Tariffs are an accelerant, not an endpoint. Section 232 remains a live risk. More fundamentally, the geopolitical clock is running on Taiwan — a US-China trade war increases tail risk of cross-strait escalation. The $165B Arizona commitment is dilutive to margins in the near term and fails to cover the worst-case scenario of a direct Taiwan disruption. Multiple compression is a feature, not a bug, of what TSMC actually is.

Our take: The bull view has the better empirical footing at this moment. Q1 earnings delivery, maintained guidance, and customer behavior data are hard information; tariff tail risk is speculation. We think the stock is pricing a 25-35% probability of severe tariff escalation; we'd put that probability at 15-20%. The risk/reward skews favorable at current levels, but the bear case deserves respect: if Section 232 results in chip-specific 25%+ tariffs and customer pass-through slips, gross margin could compress meaningfully.

Debate: AI demand — structural or cyclical?

Bull view: HPC at 59% of revenue, CoWoS fully loaded with capacity doubling, AI accelerator revenue on track to double in 2025. These are not cyclical indicators — they are the signature of a secular buildout measured in years, not quarters. The Jevons-paradox framing around DeepSeek-style efficiency gains supports demand-elastic interpretation.

Bear view: Hyperscaler capex guides (Microsoft, Google, Meta, Amazon) decelerated late in Q1. The combined messages from Nvidia's H20 export hit, a pending Section 232 outcome, and moderating hyperscaler capex suggest 2026 AI orders could be lower than the 2025 run rate implies. TSMC's order book is effectively a forward projection of Nvidia's; that's a concentrated bet.

Our take: Structural demand wins this debate in 2025; 2026 depends on whether inference workloads scale faster than training infrastructure normalizes. We are comfortable with the bull framing for the next 12 months. Beyond that, visibility is lower and earnings risk is asymmetric to the downside. The "double AI accelerator revenue" target for 2025 looks directionally correct; we would fade the idea that this doubles again in 2026.

Debate: Can gross margin hold through the Arizona ramp?

Bull view: Mix shift into 3nm and HPC is more than offsetting the Arizona dilution. Gross margin at 58.8% in Q1 versus the 57-59% guide range is proof that mix benefits are winning. As N2 ramps into HPC applications, the margin profile gets better, not worse.

Bear view: The 2-3% dilution flagged for 2025 is not yet visible because Arizona volumes are still small. As Arizona Fab 2 and Fab 3 ramp into 2026-2027, the dilution approaches 3-4%, and those are premium-node fabs where the dilution bites the highest-margin business. The structural ceiling on gross margin is coming down, and the multiple should reflect that.

Our take: Both are partially right. The mix shift will continue to offset dilution in 2025. In 2026-2027 the dilution becomes more material, but N2 pricing should capture some of that via higher ASPs. We model gross margin settling at 56-58% through the Arizona ramp, with downside to 54% if tariff-driven pricing pass-through is constrained.

Debate: Is the "political asset" framing a bug or a feature?

Bull view: The $165B Arizona commitment and Trump's explicit citation of TSMC as a tariff-policy win make TSMC a protected entity within US semiconductor policy. In a worst-case Taiwan scenario, Arizona becomes a strategic moat. TSMC's neutrality, partnering with all major US customers, makes it essential to the AI buildout regardless of geopolitics.

Bear view: Being a political asset means headline risk in both directions. A Trump admin forced-partnership with Intel is a non-zero tail risk. Taiwan election cycles, Chinese aggression, and US-China trade negotiations all create exogenous variance that no operational excellence can offset.

Our take: Bug at the multiple level, feature at the business level. The multiple is compressed because the geopolitical tail is real; the business is protected because customers have no alternative at the leading edge. Net-net, this is a stock that trades at a structural discount to its earnings power. The discount is well-known and partially justified; our call is that the current level overdoes the discount.

Model Update & Valuation Framework

Valuation framework: At a ~$155 ADR price entering the print, TSM trades at roughly 17-18x FY25 estimated ADR EPS of $8.90-9.20. For a business growing revenue 25%+ and earnings 40%+ with unique process-technology leadership, this multiple is below historical average and below peer-group leaders. A re-rating back to ~22-25x on our FY25 EPS of $9.00-9.20 implies $200-230 per ADR (upside of 30-50%). A more conservative 20x multiple implies $178-185 (upside of 15-20%). Either case offers a meaningfully positive risk/reward over a 12-month horizon.

Bear-case scenario: If Section 232 results in a 20-25% semiconductor-specific tariff, and if customer pass-through is incomplete (say 70% pass-through), TSMC's effective gross margin might compress 2-3% below baseline. FY25 EPS in that scenario moves toward $7.50-8.00, and at a compressed 15x multiple, implies $115-125 per ADR (downside of 20-25% from current levels). This is a credible bear case and establishes the downside guardrail.

Thesis Scorecard Post-Earnings (Initiating Coverage)

Overall: Thesis net-positive coming out of Q1. Every operational bull point was confirmed; every bear point remains unresolved rather than confirmed. The geopolitical tail is genuine and deserves respect, but it is already reflected in the multiple.

Action: Initiating at Outperform. The risk/reward from current levels is attractive, particularly for investors willing to hold through potential tariff-related volatility over the next 2-3 months.

Bottom Line

TSMC just posted one of the cleanest earnings quarters in its history — revenue up 42% in local currency, net income up 60%, margins expanding, HPC overtaking smartphone as the dominant platform, and full-year guidance maintained through the largest geopolitical shock the chip industry has faced in years. If you believe AI is a real multi-year secular buildout, this is the company whose earnings prove it — and whose stock is currently trading 23% below its start-of-year price because of tariff fears that are real but overstated. We initiate TSM at Outperform.

What could change our mind:

• Downgrade to Hold: A Section 232 outcome resulting in 25%+ chip-specific tariffs with limited customer pass-through; or evidence of real AI demand deceleration (CoWoS utilization below 90% or HPC revenue sequential decline).
• Downgrade to Underperform: A forced Intel JV or foundry partnership mandated by the Trump administration; a direct Taiwan-strait escalation; or gross margin compression below 54% on the Arizona ramp.

Signposts for Q2 2025 earnings (July 2025):