Stabilization Turns to Seasonality: A Clean Beat, AI Sized at $250M, Margins to 38% — Maintaining Hold as the Restocking Cycle Stays on the Sidelines

Results vs. Consensus

Q3 2025 Scorecard

Year-Over-Year Comparison

Quarter-Over-Quarter Comparison

Quality of Beat

Revenue: The 5.6% sequential growth is the cleanest the cycle has produced in onsemi's results, and the breadth matters more than the rate: automotive ($787M, +7%), industrial ($426M, +5%), and "Other" (+2%) all grew, and so did all three business units (PSG, AMG, ISG). Regionally, the Americas grew 22% sequentially and Japan 38% (partly an order-routing shift from Europe), while China declined 7% on a solar-deployment pause management characterized as temporary. The year-over-year decline narrowed to 11% from 15% in Q2, the clearest sign yet of a flattening base.

Margins: The 38.0% non-GAAP gross margin (above the 37.5% midpoint) is genuine but mix-driven progress, not yet a structural step-up. Utilization rose to 74% from 68%, but that lift came partly from building die-bank inventory for the mass market, which management expects to complete in Q4, taking utilization back down. The margin remains hostage to volume: management reaffirmed the 25–30bps-per-utilization-point math and was explicit that the climb toward 40%+ is utilization-driven. The 53% long-term target is intact but unscheduled.

EPS: Non-GAAP EPS of $0.63 (up 19% sequentially) is high quality. It came on real operating leverage (operating margin expanded 190bp to 19.2%), disciplined operating expense ($291M), a ~16% tax rate, and a share count reduced to 408M after a $325M Q3 buyback. The convergence of GAAP and non-GAAP EPS at $0.63 is itself a quality signal: the Q1 restructuring charges that depressed GAAP earnings have rolled off, so reported and adjusted earnings now agree.

Segment Performance

End-Market Revenue — Q3 2025

Business-Unit Revenue — Q3 2025

Automotive — Off the Trough, Stabilizing

Automotive grew 7% sequentially to $787M, confirming management's Q2 call that the segment troughed in the second quarter. Strength was broad across the Americas, China, and Japan. The SiC content story continued to build: onsemi expanded its position at NIO (SiC traction inverter for a new brand, plus an 8-megapixel image sensor for ADAS) and won high-voltage traction inverters with a leading Chinese Tier 1 serving multiple local OEMs, extending reach beyond the top-10 OEMs. Management was careful to frame quarter-to-quarter moves as project-ramp lumpiness rather than a clean seasonal read.

"We've always said the second half of the year is going to outgrow the first half… Auto, we said the bottom was going to be in Q2. So that has been the case, and we're going to grow from there. Grow meaning closer to demand, but no restocking yet." — Hassane El-Khoury, President & CEO

Assessment: Auto is doing exactly what management said it would: troughing in Q2 and growing in H2 toward end-demand, without a restocking tailwind. The customer-diversification strategy across China's 100-plus EV brands is de-risking the lumpiness and compounding share. The key unlock from here is the restocking cycle, which has not started. Auto is healthier, but not yet in the kind of broad recovery that would lift utilization decisively.

Industrial — Stabilizing, With an AI-Power Overlay

Industrial grew 5% sequentially to $426M, led by aerospace/defense and helped by AI-adjacent power demand that onsemi reports within Industrial: energy storage systems, solar inverters, UPS, and microgrid deployments tied to hyperscale AI buildouts. Management secured wins with two leading Chinese utility solar-inverter suppliers and a large U.S. stationary-storage OEM, and expects its latest Field Stop 7 IGBT revenue to grow in 2025 and at double digits in 2026. The industrial image-sensor design funnel is up 55% year-over-year.

Assessment: The AI-power overlay is the most interesting development in Industrial. "Everything prior to the data center wall is in Industrial" (energy storage, infrastructure), so a portion of the AI buildout shows up here rather than in "Other." That makes Industrial a quieter beneficiary of the AI power theme than the headline AI number suggests, and supports a more durable industrial recovery than the cyclical core alone.

Other / AI Data Center — Now Sized, Now Material

For the first time, management sized the AI data center business: roughly $250M expected in 2025, about 4% of sales, doubling year-over-year for a third consecutive quarter. The "Other" bucket grew only 2% sequentially because the non-core exits land here and partly offset AI growth, so the underlying AI momentum is stronger than the segment line shows. onsemi positions itself as able to address the full power tree from grid to processor, with content per AI rack management says can rise from a few thousand dollars today toward ~$50,000 by 2027.

"If you take that 'crowded space' of 13 or however many companies and you look at who can go from wall to core, there's only 2. And we are the share gainer of the two." — Hassane El-Khoury, President & CEO

Assessment: Sizing AI at ~$250M resolves the biggest disclosure gap from our initiation and converts a vague option into a measurable, fast-growing line. The "wall to core" differentiation is specific and defensible: full-power-tree coverage is a genuine moat in a fragmented field. At ~4% of sales it is not yet a needle-mover for the consolidated P&L, but the trajectory (doubling annually) and the 2027 content-per-rack opportunity make it the single most important driver of the long-term re-rating case.

Key Topics & Management Commentary

Overall Management Tone: Cautiously, deliberately constructive. Management upgraded its language from "stabilization" to "seasonality is the first step to recovery," and for the first time sounded "cautiously excited" about end-markets, while still refusing to call a restocking cycle that has not begun. The posture pairs genuine confidence in the secular portfolio (AI, vGaN, Vcore, Treo, SiC) with disciplined realism about a cyclical recovery that is improving but not yet self-sustaining. It is the most optimistic onsemi has sounded in this down-cycle, calibrated carefully so as not to over-promise.

1. From Stabilization to Seasonality: The First Step to Recovery

"It's very positive that we've gone from the stabilization to now seeing seasonal patterns. I think that's the first step to recovery." — Thad Trent, CFO

Management's framing evolved meaningfully from Q2. The business is now exhibiting normal seasonal patterns (Q4 guided flat to down ~2%, in line with history; Q1 typically down 2–3%), which management reads as a healthier signal than the abnormal trough behavior of prior quarters. Order patterns are improving, lead times pushed out slightly to ~20 weeks (from the mid-teens), and visibility has improved, if not to the level management would like.

Assessment: The return of seasonality is a real and underappreciated positive: it means the violent destocking is over and the business is behaving normally again. But "normal seasonality" still means a sequentially down Q4 and Q1, so it does not yet translate into the kind of growth that lifts utilization and margin. This is progress toward, not arrival at, recovery.

2. The Restocking Cycle That Hasn't Started

"What they need to see, one is a credible demand signal… consumer level confidence… And the biggest thing… is stabilization in the geopolitical aspect… consumer confidence and geopolitical stabilization will start adding more and more confidence for OEMs to restock." — Hassane El-Khoury, President & CEO

The single most important governor on the recovery. onsemi is shipping closer to true end-demand with no restocking tailwind, and management argues inventory has been drained below critical levels across the supply chain for two years. The Nexperia disruption in the channel is, in management's view, proof that any supply-chain trip causes a chain reaction when buffers are this thin, and that the only durable fix is OEMs and Tier 1s placing backlog with visibility.

Assessment: This is the pivotal swing factor for the rating. When restocking begins, onsemi has lean inventory, a smaller footprint, and ~700bps of utilization-driven margin upside to capture quickly. But it requires consumer-demand confidence and geopolitical stability that management cannot manufacture. The absence of restocking is precisely why we maintain Hold rather than upgrade: the recovery is set up but not yet triggered.

3. AI Data Center: Sized, Differentiated, "Wall to Core"

Management sized AI at ~$250M for 2025 and articulated the differentiation clearly: onsemi addresses the entire power tree (grid, solar/storage, UPS, rack-level PSUs, 48V bus converters, hot swaps, and compute-board smart power stages), and of the crowded field, only two suppliers can go "wall to core." The Vcore/Aura acquisition closes gaps in multiphase controllers and monolithic smart power stages, with sampling this quarter and production in early 2026. The NVIDIA collaboration on 800-volt DC architecture is a named anchor.

Assessment: The breadth claim is the heart of the bull case and it is credible. Full-power-tree coverage, validated by the SiC JFET and AMG revenue doubling annually, is a structural advantage that a point-product competitor cannot match. The risk is that revenue scale still lags the narrative and is gated by the XPU vendors' deployment timelines. We treat AI as a confirmed and quantified growth leg, no longer pure optionality, but not yet large enough to drive consolidated numbers.

4. vGaN: A Breakthrough From the Syracuse Fab

onsemi launched vertical GaN (vGaN), built on a proprietary GaN-on-GaN architecture at its Syracuse, New York fab acquired with the NexGen Power Systems technology in late 2024. Vertical conduction enables higher voltage without enlarging die size (the structural limit on lateral GaN), with reduced energy loss and faster switching, targeting AI data centers, EVs, renewables, and aerospace/defense. Sampling is underway with lead customers in automotive and AI; production revenue is expected around 2027.

"Anytime you need high voltage and high switching frequency, vertical GaN is the solution… I don't believe anybody is able to sample such technology outside. So it gives our customers… a broad portfolio of high voltage, high-efficiency products." — Hassane El-Khoury, President & CEO

Assessment: vGaN extends onsemi's wide-bandgap leadership beyond silicon carbide and is explicitly being run through the same playbook that built the SiC franchise from the GTAT acquisition. It is a credible long-dated differentiator, but with first revenue around 2027 it is a 2027–2028 story that supports the structural thesis without affecting the near-term model.

5. The Vcore / Aura Acquisition: Buying Time-to-Market for AI Power

onsemi acquired Vcore Power Technology and IP assets from Aura Semiconductor, adding advanced multiphase controllers and monolithic smart power stages that support both x86 and ARM compute platforms. Management framed it as both a near-term time-to-market lever (taking Aura products into the market for 2026) and a long-term architectural advantage once the IP is integrated into the Treo platform. Sampling begins this quarter, with production in early 2026.

Assessment: A sensible, synergistic bolt-on. It fills a real gap in onsemi's point-of-load AI offering and accelerates the roadmap by a year or more, which matters in a market where design windows are tied to XPU launch cadences. The dual approach (sell the acquired products now, re-engineer on Treo later) is the right way to convert IP into both near-term revenue and durable advantage.

6. Silicon Carbide: New Applications, Hybrid Adoption, 8-Inch in Production

SiC is "coming in exactly where we expected," with share gains continuing. Two trends are expanding the addressable market: SiC is now designing into AI data center PSUs (a market that "a few years ago… was not even a conversation point"), and SiC is penetrating plug-in hybrids and range-extender EVs that were historically assumed to stay on IGBT. onsemi's 8-inch SiC is in production at a best-in-class 350-micron thickness and ships in production volume in 2026; 300mm is "too far from now" to model.

Assessment: The SiC franchise is broadening on both ends: up into AI infrastructure and down into hybrids. That diversification reduces the dependence on a pure BEV ramp, which has been slower than the industry expected. SiC margins remain below the corporate average on under-utilization, so the franchise is still a volume-and-mix story that pays off as the cycle turns, with 8-inch providing a structural cost lever in 2026.

7. Capital Allocation and Cash Generation

onsemi repurchased $325M of stock in Q3 ($925M year-to-date, ~100% of free cash flow) and has bought back $2.1B since launching the program in February 2023, with ~$861M remaining on the authorization. Free cash flow of $372.4M (24% of revenue, +22% YoY) was the cash highlight, reversing Q2's working-capital drag, and inventory fell $39M to 194 days (with bridge inventory down to 82 days). The balance sheet holds ~$2.9B of cash and short-term investments and $4B of total liquidity.

Assessment: The cash story is the most unambiguously positive part of the quarter. Generating 24%-of-revenue free cash flow at a cyclical trough, while inventory declines, demonstrates a structurally cash-generative model. Returning ~100% of it via buyback at a depressed valuation is accretive and signals management conviction. This firmly supports the floor under the stock, even as it does not by itself force a re-rating.

8. Mass-Market Strategy and the Die-Bank Build

onsemi is deliberately building die-bank inventory to seed the mass market (small and emerging customers served through distribution), where customer count rose ~20% year-over-year. Roughly 58% of revenue flows through distribution, split about evenly between fulfillment and demand creation, with the mass market a subset (~25% of distribution revenue). The die-bank build lifted Q3 utilization but completes in Q4, after which utilization normalizes lower.

Assessment: The mass-market push is a sound margin-mix strategy (these are richer-margin, stickier revenues), and the die-bank build is a rational bet on the +20% customer-count growth. The wrinkle for modeling is that the Q3 utilization uptick is partly this build rather than pure demand, so the Q4 utilization step-down is not a demand warning. It does, however, mean Q3's margin lift is not fully a clean demand signal.

Guidance & Outlook

The Q4 revenue guide of $1.48–1.58B (midpoint $1.53B) is a 1.3% sequential decline, which management explicitly characterized as in line with normal Q4 seasonality (flat to down 2%). Within the guide, automotive and industrial are expected down low-single-digits and "Other" up mid-to-high single digits on AI data center plus seasonal non-core dynamics. The gross-margin guide is flat at a 38.0% midpoint as the mass-market die-bank build completes and utilization steps down.

The seasonality signal: The most important interpretive point is that a normal seasonal Q4 is, paradoxically, good news. After multiple quarters of abnormal trough behavior, the return of textbook seasonality is management's evidence that the destocking is over and the business has normalized. But it also means the recovery does not show up as sequential growth in Q4 or, on normal seasonality, Q1, deferring the utilization-and-margin inflection into 2026.

2026 setup: Management reiterated the ~5% / ~$300M 2025-revenue non-repeat from non-core exits and ISG repositioning, with no change. It declined to guide 2026 by market but pointed to (a) the ~15% non-GAAP tax rate (from ~19%), (b) the ~$5M/quarter depreciation reduction beginning in Q4, (c) double-digit growth in Field Stop 7 IGBT for energy storage, and (d) utilization-driven margin upside as the swing factor. The 2026 algebra is a smaller base growing, with margin leverage gated on a recovery management still will not date.

Analyst Q&A Highlights

Automotive Upside and the Sustainability of the Sequential Growth

The opening exchange pressed management on what drove automotive's beat versus the low-single-digit guide and whether the growth is sustainable into Q4 and 2026. Management deflected the quarter-on-quarter read in favor of a first-half/second-half framing and flagged the still-absent restocking cycle.

Q: "Upside in the quarter nicely versus the low single-digit guide… what caused the upside, and perhaps what's the sustainability of that sort of growth as you look into the fourth quarter and then 2026 as well[?]"
— Ross Seymore, Deutsche Bank

A: "What we're seeing in automotive is really stabilization… The quarter-on-quarter, I don't think I read anything into it. It's purely between seasonality and ramps… demand is stabilizing, we're starting to see a seasonal trend. But one thing I would highlight is we haven't seen a restocking cycle yet."
— Hassane El-Khoury, President & CEO

Assessment: The "no restocking yet" caveat is the analytical core of the recovery debate. Auto is growing toward end-demand without a channel-replenishment tailwind, which means there is a coiled spring of utilization-and-margin upside waiting for the restocking trigger, but also no near-term forcing function for it. This exchange is why the rating stays Hold: the upside is identified and undelivered.

Sizing AI at $250M and the "Wall to Core" Differentiation

A request to size the AI business and explain onsemi's differentiation, given that NVIDIA's 800-volt list includes many suppliers, drew the most important strategic answer of the call: full-power-tree coverage as the moat.

Q: "You, for the first time, I believe, sized the AI business at about $250 million… roughly 4% of sales. Can you just talk about how ON differentiates in that?… there's 13 other folks on the list as well… how do you believe that $250 million will grow?"
— Ross Seymore, Deutsche Bank

A: "If you take that 'crowded space' of 13 or however many companies and you look at who can go from wall to core, there's only 2. And we are the share gainer of the two… we're one of the only companies that are able to support the power delivery from the high voltage all the way to the core."
— Hassane El-Khoury, President & CEO

Assessment: The "wall to core, only 2" framing is the most investable sentence on the call. Full-power-tree breadth is a structural moat that point-product rivals cannot replicate, and being one of two such suppliers in a doubling market is a genuine share-gain setup. The constraint is scale and timing, not differentiation. This is the disclosure that most strengthens the long-term bull case.

Q4 Utilization Step-Down and Q1 Seasonality

A modeling question probed why utilization is guided down in Q4 and how Q1 seasonality would flow into gross margin. The answer separated the demand signal from the die-bank build and reaffirmed the multi-quarter lag between utilization and margin.

Q: "You said something about utilization perhaps flat to slightly down. Anything more to read into it? And if you could just remind us what is your seasonal pattern in Q1?… how does that kind of reflect in gross margins in Q1[?]"
— Vivek Arya, Bank of America

A: "Our utilization increased to 74% in Q3… we've been building die bank inventory for the mass market… I expect… utilization to be down in Q4 because we expect those builds to be completed… If you take the midpoint of our guidance for Q4, it's directly in line with normal seasonality, which is typically flat to down 2%… Q1, it's typically down 2% to 3%."
— Thad Trent, CFO

Assessment: A useful clarification that the Q4 utilization step-down is a completed-build artifact, not a demand warning. But it also confirms that, on normal seasonality, neither Q4 nor Q1 produces the sequential growth needed to lift utilization durably. The margin inflection is a 2026 event contingent on the recovery, which is the crux of the patient stance.

The Vcore / Aura Acquisition Scope

A question on what the Vcore acquisition adds, specifically whether it helps on the XPU-side voltage-regulator content, drew confirmation that it is both a near-term product set and a long-term Treo-integrated architecture play.

Q: "Exactly what comes into the business with that acquisition?… there's a huge number of voltage regulators on the XPU side. Does Vcore help you on that? Or does that come from the existing ON product portfolio?"
— Quinn Bolton, Needham & Company

A: "A combination of both… it complements the product offering that we're already offering with Treo. It provides products also in the short term… revenue generation coming here in 2026… In long term, it gives us an architectural advantage… once we leverage the performance of Treo."
— Hassane El-Khoury, President & CEO

Assessment: The dual rationale (immediate revenue plus eventual Treo-integrated performance) is the right way to think about a capability bolt-on in a fast-moving market. It accelerates onsemi's point-of-load AI roadmap by buying products it would otherwise have to develop, while preserving the long-term differentiation of the Treo platform. A well-structured tuck-in.

vGaN Maturity and Time to Revenue

A two-part line of questioning tested vGaN's reliability versus lateral GaN and asked when production revenue arrives, plus how onsemi commercialized a technology the acquired startup could not. Management argued vertical conduction solves both the reliability and the die-size economics, and tied commercialization to the same muscle that built the SiC franchise.

Q: "GaN to date hasn't been used that much in the high-power segments… because of reliability issues. Can you just address how do you feel the vertical GaN technology compares with lateral GaN on reliability? And… when… that might start to go into production?"
— Quinn Bolton, Needham & Company

A: "Vertical GaN is better on the reliability side… the current goes vertically, which means that we can go higher and higher voltage without increasing the die size… We have lead customers in… AI and automotive… you can think about it in the '27 time frame [for revenue]."
— Hassane El-Khoury, President & CEO

Assessment: The technical answer is coherent (vertical conduction decouples voltage from die size, the structural limit on lateral GaN), and the GTAT/SiC analogy is the right credibility frame given management's track record scaling a hard wide-bandgap technology. But 2027 first revenue makes this a long-dated option that does not affect the investment case for the next several quarters.

Why Build Die Bank Ahead of Seasonally Down Quarters

A question challenged the logic of taking utilization up to build die-bank inventory just before two seasonally soft quarters. Management defended it as a disciplined, demand-justified investment in the fast-growing mass market rather than building blind.

Q: "Why… it's the right time to start building up die bank inventory and taking utilization rates up, especially ahead of a couple of down seasonal quarters[?]"
— Joshua Buchalter, TD Cowen

A: "Our customer count increased almost 20% year-on-year just in the mass market. Therefore, the demand is there for that, and we will make sure that we have it in die bank internally so we can respond to changes in demand… that doesn't mean that we're going to be building blind… even with that die bank increase, our inventory actually declined quarter-on-quarter, $39 million."
— Hassane El-Khoury / Thad Trent

Assessment: The defense is solid: a ~20% rise in mass-market customer count justifies seeding die bank for quick-turn fulfillment, and total inventory still fell $39M, so this is a mix shift, not a build-up. It is a reasonable, margin-accretive bet, though it does mean Q3's utilization uptick overstates the pure demand signal.

Lead Times and Order Visibility

A question on short-lead-time orders and visibility drew confirmation that lead times pushed out and order patterns are improving, supporting the stabilization read without yet signaling a sharp acceleration.

Q: "I was wondering if you could talk about the rate of short lead time orders that you're seeing… And if you're seeing any increased visibility?"
— Joe Quatrochi, Wells Fargo

A: "Our lead times actually pushed out slightly. We're… up around 20 weeks or so now… Customers are layering in backlog as they have visibility. We probably have seen order patterns that continue to improve, which gives us that confidence in the stabilization right now."
— Thad Trent, CFO

Assessment: Lead times extending to ~20 weeks and improving order patterns are constructive leading indicators that corroborate the stabilization narrative. They fall short of the kind of bookings surge that would precede a restocking cycle, which keeps the read at "improving visibility," not "recovery confirmed."

What They're NOT Saying

1. Still no 2026 guidance, and no restocking date: Management explicitly declined to guide 2026 by market and would not put a timeline on the restocking cycle, deferring to consumer confidence and geopolitical stability it cannot control. The biggest swing factor for the stock remains undated.
2. No quantified margin path to 40% by revenue level: Asked for a revenue number that gets gross margin to 40%, management would only say it is "utilization driven," declining to give a revenue or utilization threshold. The margin recovery is real but deliberately unbounded in time.
3. AI growth rate not guided: AI was sized at ~$250M for 2025 but management gave no 2026 growth target for it, leaving the single most important secular driver unquantified beyond "continue to grow."
4. SiC revenue still not disclosed: Management again declined to size silicon carbide ("we didn't provide any guidance on silicon carbide"), saying only that it is "coming in exactly where we expected." A franchise this central to the thesis remains a black box on absolute dollars.
5. Nexperia impact downplayed: Despite Tier-1 customers' visible anxiety about the Nexperia disruption, management characterized any benefit as "too soon to call," declining to claim share or a restocking catalyst it could plausibly capture.

Market Reaction

• Pre-print setup: ON closed at $50.08 on Friday Oct 31, down 20.6% year-to-date and 29.5% over the trailing twelve months, but up modestly (+1.6%) over the trailing 30 days. The stock had recovered off the Q2 trough but remained deeply out of favor (52-week closing range $31.95–$74.70).
• Reaction-day move (Nov 3, before-open report): The stock dipped ~1.9% pre-market on the beat, then recovered through the session to close at $50.46, up 0.8% (+$0.38), on volume of 18.1M shares versus an 8.9M-share 30-day average (2.0x normal).
• Intraday: Traded a $47.99–$52.26 range, a ~4% swing in both directions before settling marginally green.
• Market context: The S&P 500 was roughly flat (+0.2%), so the muted move was idiosyncratic.

A clean beat that closed +0.8% is, in its own way, as informative as Q2's 15.6% drop. The market reaction tells us three things.

The beat was expected: After the Q2 sell-off and a stock down ~20% YTD, low expectations were already in the price. A 2% revenue beat and a 38% gross margin met a bar that had been reset downward, so there was little fuel for a rip.

The seasonally flat guide capped the upside: A −1.3% sequential Q4 guide, however "normal," gave momentum buyers nothing to chase. The market wanted a guide that signaled acceleration; it got one that signaled normalization.

The restocking absence kept buyers cautious: With management explicitly declining to call restocking and utilization guided down in Q4, the recovery-positioning trade had no catalyst. The +0.8% close reflects a market that believes the stabilization story but is unwilling to pay up until the inflection is visible. That is the same posture we are taking with the rating.

Street Perspective

Debate: Is the Recovery Real or Just Normalized Seasonality?

Bull view: All three end-markets and all three business units grew, auto is off the trough, lead times are extending, order patterns are improving, and the return of normal seasonality is itself the signal that destocking is over. With lean inventory and a smaller footprint, onsemi has powerful operating leverage waiting for the restocking trigger. This is the early innings of a multi-quarter recovery.

Bear view: "Normal seasonality" means sequentially down Q4 and Q1, so there is no actual growth on the horizon. Management itself won't call restocking, won't guide 2026, and won't quantify the margin path. Revenue is still down 11% year-over-year and the 2026 base shrinks ~5% on exits. A recovery that produces flat-to-down quarters is not a recovery the equity gets paid for yet.

Our take: Lean bull on direction, neutral on timing. The destocking is clearly over and the setup is constructive, but the absence of restocking and the seasonally soft next two quarters mean the payoff is a 2026 event. We need to see the inflection start before upgrading, which is exactly the Hold thesis.

Debate: Does AI Data Center Justify a Re-Rating?

Bull view: AI is now sized at ~$250M, doubling annually, with a credible "wall to core" moat that only one other supplier shares. Content per rack can rise toward ~$50,000 by 2027, the Vcore acquisition fills the point-of-load gap, and vGaN extends leadership into 2027–2028. As AI scales from 4% toward a double-digit revenue share, it re-rates the entire company to a growth multiple.

Bear view: At ~4% of sales, AI is too small to move consolidated numbers for years, the ramp is gated by other companies' deployment timelines, and onsemi is a share-gaining latecomer in a field of established power-IC incumbents. Design-in lists are not dollar revenue. The AI premium is a story the income statement won't support until 2027 at the earliest.

Our take: Real and quantified, not yet a numbers driver. We give onsemi credit for genuine differentiation and a doubling trajectory, but at 4% of sales it earns the company an option value, not a re-rate. The catalyst we would act on is AI scaling toward a high-single-digit revenue share with a disclosed growth rate.

Debate: Can Gross Margin Reach 40%+ on a Realistic Horizon?

Bull view: Gross margin already lifted to 38% on mix with utilization at just 74%. With ~700bps of utilization-driven upside to a reset ~92% full-utilization level, plus Fab Right, divested-fab monetization, 8-inch SiC cost gains, and richer AI/Treo mix, 40%+ is achievable as the recovery fills the smaller footprint, with 53% the long-term destination.

Bear view: Management won't name a revenue level that gets to 40%, the Q4 margin guide is flat as utilization steps down, and the biggest lever is a demand recovery that won't be dated. SiC remains below the corporate average. Mid-to-high-30s, not 40%+, is the realistic 2026 range unless restocking arrives.

Our take: Neutral on timing. The levers are real and the 38% print is encouraging, but the margin inflection is a 2026 event gated on the same recovery the whole thesis hinges on. We model gross margin flat-to-modestly-up near term, reaching toward 40% as utilization recovers through 2026.

Model Implications & Valuation Framework

Valuation framework: At the post-print price of ~$50.46 and our FY25 non-GAAP EPS estimate of ~$2.34, ON trades at ~21.5x trough earnings and ~18–19x our FY26 estimate of ~$2.55–2.95. The valuation is essentially unchanged from our initiation: fair for a high-quality cyclical at the bottom, with a now-quantified AI growth leg and disciplined capital return, but not the kind of mispricing that compels an upgrade. The clean beat narrows the bear tail (the trough looks more defensible) without yet widening the bull case enough to re-rate.

Price-target framework (12-month):

• Base case ~$52–58: ~19–21x our ~$2.75 FY26 EPS midpoint. Recovery proceeds gradually through 2026, utilization climbs as restocking begins, margins drift toward 40%. Modest upside from the post-print price.
• Bull case ~$68–76: ~23–25x a ~$3.00+ FY26 EPS if restocking starts in H1 2026, utilization recovers sharply, and AI plus SiC drive mix accretion. A clean re-rate to a recovery multiple.
• Bear case ~$38–42: ~16–18x a ~$2.30 FY26 EPS if restocking is deferred, Western auto and traditional industrial stay soft, and the ~5% exit headwind produces another year of declining revenue with margins stuck in the high-30s.

The base case implies modest upside from $50.46; the risk/reward is balanced, with the swing factor being the timing of the restocking cycle. That is the quantitative basis for maintaining Hold.

Thesis Scorecard Post-Earnings

Grading the standing thesis established at our Q2 2025 initiation against this quarter's print and call.

Overall: Thesis strengthened modestly versus initiation. Management delivered on every Q2 commitment we were watching: all three end-markets grew, AI was sized as promised, the Treo funnel crossed $1B, and free cash flow inflected. Bull #1 strengthened materially; the bear points are softening but not yet resolved.

Action: Maintain Hold. The path to an upgrade is now clearly visible (restocking + utilization inflection + AI scaling); the trigger has not been pulled.

Bottom Line: The Best Print of the Cycle, but the Turn Is Still on the Sidelines

Rating decision: We maintain Hold on onsemi. Q3 was the strongest quarter of our coverage: a clean beat, broad-based sequential growth, gross margin up to 38%, free cash flow inflecting to $372M, and the AI data center business sized and material at ~$250M with a credible "wall to core" moat. Management delivered on every commitment we set at our initiation. This is a company executing well through a trough and broadening its secular portfolio with genuine differentiation (vGaN, Vcore, SiC's expansion into AI and hybrids).

But the recovery that would justify an Outperform is set up, not started. Management itself will not call restocking, the Q4 guide is seasonally flat, utilization steps down in Q4, and 2026's base shrinks ~5% on deliberate exits. At ~21x trough earnings the stock is fairly valued, and the +0.8% reaction to a clean beat tells us the market shares our view: the stabilization story is believed, but nobody is paying for the inflection until it is visible. We would rather wait for the restocking cycle to begin and utilization to climb than upgrade into a seasonally soft two quarters.

What would move us to Outperform: Evidence the restocking cycle has begun (bookings acceleration, lead times extending further, channel replenishment); utilization climbing back above 74% on demand rather than die-bank builds with gross margin breaking above 40%; or AI data center scaling toward a high-single-digit revenue share with a disclosed growth rate.

What would move us to Underperform: Q4 or Q1 coming in below normal seasonality (a relapse rather than a soft patch); gross margin slipping back toward the mid-30s; or the ~5% exit headwind compounding into a 2026 that fails to grow despite a stabilizing market.

Signposts for Q4 2025 earnings (early February 2026):