Stabilization, Not Recovery: onsemi Beats and Meets but Sheds 16% on the Long Road to 53% Margins — Initiating at Hold

Results vs. Consensus

Q2 2025 Scorecard

Year-Over-Year Comparison

Quarter-Over-Quarter Comparison

Quality of Beat

Revenue: The 1.2% top-line beat is real but thin, and the more important number is the 15.4% year-over-year decline. onsemi is two-plus years into a semiconductor down-cycle concentrated in its two largest end-markets (automotive and industrial), and the sequential growth of 1.6% is the first sign of a flattening rather than a bounce. The quality wrinkle is the composition: the Power Solutions Group rose 8% sequentially while Analog/Mixed-Signal and Intelligent Sensing both declined, so the sequential gain leaned on a single business unit rather than a broad-based lift.

Margins: A 37.6% non-GAAP gross margin in line with the guide midpoint is, in isolation, fine. In context it is the crux of the bear case. The figure embeds roughly 900bps of under-absorption charges; in other words, idle-capacity cost is suppressing the reported margin by nearly a thousand basis points, and that cost only unwinds with volume. The Q1 capacity impairment lowered the depreciation base (a ~$5M quarterly reduction begins in Q4) and reset "fully utilized" higher, but the unmistakable read is that the margin cannot meaningfully expand until demand returns. The 53% long-term target is an aspiration resting on a recovery that is not yet in the run-rate.

EPS: Non-GAAP EPS of $0.53 met consensus and benefited from disciplined operating-expense control (OpEx fell $17M sequentially as Q1 restructuring actions began to bite) and a steady $300M-per-quarter buyback pace that is shrinking the share count toward ~410M. The gap between GAAP ($0.41) and non-GAAP ($0.53) is narrowing as restructuring charges roll off, which improves earnings quality. But the year-over-year halving of EPS (from $0.96) is the dominant fact: this is a trough earnings stream, and the print did nothing to establish how steep the climb out will be.

Segment Performance

End-Market Revenue — Q2 2025

Business-Unit Revenue — Q2 2025

Automotive — The Trough, With China Doing the Heavy Lifting

Automotive at $733M (down 4% sequentially) is the single most important number for the cyclical thesis, and management framed it as the bottom: weakness in the Americas and Europe was offset by a 23% sequential surge in China revenue driven by silicon-carbide content on new EV platforms. The Xiaomi YU7 electric SUV integrates onsemi's 1,200-volt EliteSiC M3e devices, and the company extended its Schaeffler collaboration to a global OEM's plug-in-hybrid traction inverter using its newest EliteSiC M4T trench technology. China is now a structural share-gain story for onsemi, with management citing roughly 50% share in its target SiC segments.

"We talked about Automotive hitting the low end in the second quarter… we expect Automotive to be up in the third quarter. So you're starting to see that stabilization. I'm not there calling a recovery." — Hassane El-Khoury, President & CEO

Assessment: The China SiC ramp is the most encouraging operational data point in the quarter and validates the silicon-carbide investment. But automotive overall remains roughly 30% below its prior peak, materially worse than analog peers who are within 4–5% of their auto peaks. Part of that gap is onsemi's own portfolio rationalization (which will not repeat), and part is genuine end-demand softness in Western EV programs. The trough call is credible; the slope of the recovery out of it is the open question, and it is the reason the cyclical thesis is not yet ratable as Outperform.

Industrial — Bouncing Along the Bottom

Industrial rose 2% sequentially to $406M, but the internal mix matters: medical and aerospace/defense grew while "traditional industrial" declined slightly and came in below management's own expectation. CFO Thad Trent characterized traditional industrial as "bouncing across the bottom," with stabilization but continued choppiness quarter to quarter. The segment is not yet in recovery; it is in a low-amplitude oscillation off the bottom.

Assessment: Industrial is the clearest read on the broader cycle, and its message is "stabilized, not recovering." Until traditional industrial inflects, blended utilization (and therefore gross margin) stays pinned. We treat industrial as a coincident indicator for the recovery timing, and it is not flashing green.

Other / AI Data Center — Small Base, Right Direction

The "Other" bucket rose 16% sequentially, with AI data center the standout contributor, nearly doubling year-over-year for a second consecutive quarter. This is where the secular bull case lives. onsemi positions itself as the only broad-based U.S. power-semiconductor supplier addressing the data-center power-delivery bottleneck, working with NVIDIA on the 800-volt DC architecture and shipping single smart power stages (SPS) in an industry-standard 5x5 package while sampling a dual SPS in the same footprint.

Assessment: AI data center is the highest-quality growth in the portfolio, but it is still a small share of revenue and management declined to size it. The franchise is real and the design-in cadence is encouraging, yet at current scale it cannot offset auto/industrial cyclicality. This is a 2026–2027 contributor that earns the company a structural call option, not a near-term earnings lever.

Key Topics & Management Commentary

Overall Management Tone: Measured and deliberately un-triumphant. Management leaned into the word "stabilization" while explicitly refusing to "call a recovery," and paired every constructive data point (China SiC, AI data center, Treo traction) with a caution about end-demand uncertainty and customer order-timing behavior. The posture is that of a team running the business for a trough it does not control while investing in differentiation it does, and choosing balance-sheet and margin discipline over chasing low-quality revenue. It is a credible tone, but a conservative one that gave the Street no reason to extrapolate a sharp upturn.

1. Stabilization vs. Recovery: The Word Management Would Not Say

"We're seeing stabilization relative to where we have been over… the last 3, 4, 5 quarters, that is a positive development… I'm not there calling a recovery. There's still a lot of uncertainty and customers are being cautious." — Hassane El-Khoury, President & CEO

The single most important framing of the call. Management sees the bottom (auto troughed, all three end-markets guided up in Q3) but refuses to underwrite an upturn. Customers are ordering late and cautiously, with the tariff backdrop and end-demand uncertainty cited as reasons. The company is running lean on inventory and channel to be positioned for a turn it can react to rather than forecast.

Assessment: This is the right way to manage a trough and the wrong setup for a re-rate. A management team that will not call recovery gives the market no anchor for second-half acceleration, which is precisely why a revenue beat sold off. For our rating, the absence of a recovery call is the deciding factor: we cannot pay for an upturn that the people closest to the order book will not yet endorse.

2. The Margin Math: 900bps of Under-Absorption and the Road to 53%

"The key to margin expansion for us is all about utilization… every point of utilization is 25 to 30 basis points of gross margin improvement. That math still holds." — Thad Trent, CFO

The Q3 guide carries ~900bps of under-absorption charges, consistent with Q2. Management's bridge to the 53% long-term gross-margin target decomposes as: roughly 700bps from utilization recovery (now that "fully utilized" has reset to ~92% post-impairment, up from the mid-80s), ~200bps from Fab Right initiatives plus the in-sourcing/monetization of divested fabs, and the remainder from favorable mix and new-product margins. Today's blended margin is ~37.6%; the target is 53%. That is more than 1,500bps of aggregate uplift, and the largest single block (utilization) is entirely demand-dependent.

Assessment: The math is coherent and the levers are real, but the timeline is long and front-loaded on a recovery onsemi will not forecast. Until utilization inflects, gross margin is range-bound in the high-30s, capping near-term EPS power. This is the structural reason the stock is a Hold and not an Outperform at the trough: the upside is genuine but back-end-loaded and contingent.

3. Silicon Carbide: China Wins and a Competitor in Bankruptcy

Silicon carbide was the growth engine within Power Solutions, with China up 23% sequentially on new EV ramps. Management emphasized that wins are earned on device performance (range, efficiency) rather than price, citing the new EliteSiC trench generation and dual die/module flexibility. On the bankruptcy of a SiC competitor (Wolfspeed), management was pointed: it is not the trigger for share shifts, which were already underway based on roadmap and sourcing decisions made well before any filing.

"I've always maintained my focus on we're going to win… because of our own products and because of our own investment, not because one of our peers are struggling." — Hassane El-Khoury, President & CEO

Assessment: The SiC franchise is onsemi's clearest competitive edge, and the China share gains are tangible proof. The competitor distress is a modest tailwind to supply-chain confidence among customers, even if management downplays it. SiC gross margin is currently below the corporate average on under-utilization, so the franchise is a volume-and-mix story that pays off as the cycle turns rather than today.

4. AI Data Center Power: The 800-Volt Bet

onsemi is building a power-tree position for AI data centers, from high-voltage JFETs to smart power stages adjacent to the GPU/XPU. The company frames AI growth as power-limited rather than compute-limited and positions itself as the only broad-based U.S. power-semiconductor supplier addressing the 800-volt DC architecture transition (with NVIDIA explicitly named). It is shipping single SPS products and sampling a dual SPS in the same footprint, on a 12–18-month qualification-to-production cadence tied to customer platform roadmaps.

Assessment: Strategically this is the most attractive part of the story, with a credible multi-year demand vector and genuine U.S.-supplier differentiation. The catch is timing and scale: meaningful revenue tracks the XPU vendors' 800-volt rack deployments, which management would not date precisely. We value this as embedded optionality, not as a 2025–2026 numbers driver.

5. Treo: The Analog Platform Doubling Its Funnel

The Treo analog platform, built at the East Fishkill 300mm fab, is the company's flagship new-product engine in Analog/Mixed-Signal. The design funnel more than doubled sequentially against a $1B revenue target; the company recognized first Treo revenue in Q1 (a few quarters ahead of plan) and is on track to double the number of sampling products year-over-year. In automotive specifically, onsemi has sampled 10BASE-T1S to more than ten customers building zonal architectures.

Assessment: Treo is the credibility test for onsemi's claim that it can move up the value chain. A doubling funnel and pulled-forward first revenue are encouraging leading indicators, but a $1B target against a ~$6B revenue base is a multi-year build. It supports the long-term margin-mix thesis without changing the near-term P&L.

6. Portfolio Rationalization: A Smaller, Higher-Quality Base

"In 2026, we expect that approximately 5% of our 2025 revenue will not repeat. This includes the end of life of certain legacy products, ongoing noncore exits and the repositioning of ISG." — Thad Trent, CFO

Management is deliberately shrinking the revenue base to raise its quality: ~$200M of exits in 2025 (about $100M realized through Q2) and ~5% of 2025 revenue (~$300M) not repeating in 2026, inclusive of a $50–100M ISG repositioning toward ADAS/machine vision. Fab Right capacity actions are being matched to these exits so the footprint shrinks with the revenue.

Assessment: Strategically sound, optically painful. Exiting low-margin revenue is the correct long-term move and supports the eventual margin target, but it means the 2026 revenue line starts ~5% in the hole before any cyclical growth, which structurally caps the near-term top-line recovery and was a clear contributor to the sell-off. The Street has to model a shrinking base recovering, not a growing base accelerating.

7. Capital Allocation: 100%+ of Free Cash Flow Returned

onsemi raised its 2025 buyback target to 100% of free cash flow and is running ahead of it, repurchasing $300M of stock in Q2 and returning 107% of year-to-date free cash flow. The balance sheet holds $2.8B of cash and short-term investments with $4B of total liquidity (including $1.1B undrawn on the revolver). Free cash flow of $106M in Q2 was soft on working-capital timing, with management reaffirming a 25% full-year free-cash-flow margin target.

Assessment: The capital-return posture is shareholder-friendly and signals management confidence in normalized cash generation, while the buyback at depressed prices is accretive. But returning more than 100% of a trough free-cash-flow stream is a statement about valuation conviction, not a substitute for the earnings recovery the equity needs to re-rate. It supports the floor; it does not move the ceiling.

8. Tax and Depreciation Tailwinds

Two modeling tailwinds emerged. First, the recently enacted U.S. tax legislation ("One Big Beautiful Bill") is expected to reduce onsemi's non-GAAP tax rate to ~15% in 2026 and beyond, down from the prior ~19% expectation, with no material 2025 change. Second, the Q1 capacity impairment lowers the depreciation base, producing a ~$5M quarterly reduction on the income statement starting in Q4 2025.

Assessment: Both are modest but real EPS tailwinds into 2026 that are independent of the demand cycle. The ~400bps tax-rate reduction alone is worth roughly $0.10–0.12 of annualized non-GAAP EPS at normalized pre-tax income. These help the out-year earnings bridge but do not change the 2025 trough story.

Guidance & Outlook

The Q3 revenue guide of $1,465–1,565M (midpoint $1,515M) implies 3.1% sequential growth, with management guiding all three end-markets up: automotive and industrial up low-single-digits and "Other" up mid-to-high-single-digits on AI data center and other ramps. The EPS midpoint of $0.59 represents an 11% sequential step, driven less by gross margin (guided flat at a 37.5% midpoint) and more by the full realization of Q1 restructuring savings flowing through operating expenses, where the guide drops to a $287.5M midpoint.

Implied gross-margin signal: Flat sequential gross margin at the 37.5% midpoint despite ~3% revenue growth is the tell. With ~900bps of under-absorption persisting and utilization guided only "flat to up slightly," the near-term margin ceiling is firmly in the high-30s. Margin expansion is deferred to the point at which management gains enough Q4/2026 visibility to take utilization up, which it explicitly declined to commit to this quarter.

Guidance style: Consistent with management's conservative posture through the down-cycle. The revenue range is wide ($100M, ~6.6% of midpoint), reflecting genuine uncertainty about order timing, and the EPS bridge leans on cost actions the company controls rather than demand it does not. This is credible guidance, not sandbagged optimism, which is part of why it could not arrest the sell-off.

Analyst Q&A Highlights

Where the Cycle Sits and When Secular Drivers Offset the Exits

The opening exchange pressed management to reconcile a more constructive cyclical tone with the drag from businesses being exited, and to identify when the secular growth engines (AI data center, Treo) would offset the deliberate revenue attrition. Management framed the answer around stabilization and the maturation of multi-year investments rather than a near-term inflection.

Q: "You sound better cyclically than you have in a while… we still have the offsets of businesses you're exiting… probably a 5% headwind. Can you talk a little bit about the traction in those secular drivers and when the good ones are going to offset the exits?"
— Ross Seymore, Deutsche Bank

A: "We're seeing stabilization relative to where we have been… I'm not there calling a recovery… All of these investments that we've been making over the last few years are what's driving our longer-term view… now they're starting to come together with traction in the market that we are able to use as a foundation for where we go next."
— Hassane El-Khoury, President & CEO

Assessment: Management confirmed the secular engines are turning on but would not put a date on the crossover where they offset the exits. The honest answer is that 2025 is a year of deliberate base-shrinking, with the secular drivers becoming material in 2026–2027. That is the correct strategy and the wrong news for anyone underwriting near-term top-line growth.

Why Gross Margin Is Flat-to-Down Near Term and the Bridge to 53%

A direct challenge on the central bear point: with revenue rising, why is gross margin not, and what specifically gets the company from the mid-40s structural level to the 53% target. The response decomposed the bridge into utilization, Fab Right, fab-divestiture monetization, and new-product mix, while reaffirming that utilization is the dominant near-term lever.

Q: "In the near term, why is it flat to slightly down if your revenues are up? And then longer term… what are the key levers you think you can pull to get to the 53% long-term target, if indeed, that is still the target?"
— Ross Seymore, Deutsche Bank

A: "The key to margin expansion for us is all about utilization… you've got about 900 basis points of underutilization charges in our Q3 guide… every point of utilization is 25 to 30 basis points… We also have the monetization of the divested fabs… another 200 basis points kind of in that neighborhood. And then… as we ramp these new products, they are at favorable margins."
— Thad Trent, CFO

Assessment: The bridge is internally consistent, but it confirms that the bulk of the margin recovery is gated by demand-driven utilization that management will not forecast. This exchange is the analytical core of the sell-off: the Street wanted a self-help margin path independent of the cycle, and the honest answer is that there isn't one of meaningful size.

Why onsemi's Automotive Recovery Lags Its Analog Peers

A pointed comparison noted that several analog peers are within 4–5% of their prior automotive peaks while onsemi's auto business sits roughly 30% below peak, and asked why the recovery is so much slower and when auto could grow year-over-year again. Management attributed the gap to portfolio rationalization (which will not repeat), Western EV-program softness, and mix/exposure differences rather than share loss.

Q: "Some of them are within… 4%, 5% of their prior automotive peaks, whereas ON's auto business is still… 30% of your prior peak. So why is the automotive recovery so slow for ON? And when do you think that your Auto business could start to regrow year-on-year?"
— Vivek Arya, Bank of America

A: "Both Europe and North America are weak… where we are in the EV ramps continue to happen, of course, not at the same rate that we all expected… we hit the bottom in the second quarter. We're starting to post growth. Our expectation in Q3 will be growth."
— Hassane El-Khoury, President & CEO

Assessment: Management did not fully close the peer-gap question. Part of the 30%-below-peak gap is self-inflicted (deliberate exits) and part is genuine Western-EV weakness, but the answer left open how much is structural share/mix versus cyclical. This is the most important unresolved bear point on the name and a reason to wait for year-over-year auto growth to actually print before upgrading.

The Post-Impairment Definition of "Fully Utilized"

A modeling-precision question challenged the math: if 900bps of under-absorption maps to 25–30bps per point of utilization, the implied "fully utilized" level looked implausibly high against management's historical mid-80s framing. The answer recalibrated the full-utilization level upward post-impairment.

Q: "You said there's kind of [900] basis points of underutilization charges… it kind of implies you need to get to 98% utilization… I thought you guys in the past had said full utilization was more low to mid-80s. So can you just clarify… where do you see full utilization?"
— Quinn Bolton, Needham & Company

A: "Previously, we had said fully utilized… was kind of in that mid-80% range. Post the impairment… that is now kind of in the low 90s, call it, somewhere around 92%… you get to somewhere around 700 points of improvement. There's also another 200 basis points of Fab Right initiatives."
— Thad Trent, CFO

Assessment: A clean and useful clarification. The Q1 impairment shrank the denominator, so the same volume now maps to a higher utilization percentage and full utilization resets to ~92%. The practical implication is that the ~700bps of utilization-driven margin recovery requires demand to fill a meaningfully reduced footprint, which is achievable but, again, demand-gated.

Smart Power Stage Qualification Timing for AI Data Center

A question probed the ramp timing for the single and dual smart power stages and the broader 800-volt rack opportunity tied to the XPU vendors' roadmaps. Management gave a standard 12–18-month qualification-to-production window and tied the ramp explicitly to customer platform deployment schedules.

Q: "You'd mentioned the smart power stages, both single and dual phase that… you're sampling now… is that sort of about a year-long qualification timing? Do you think you could ramp faster?… the 800-volt rack opportunity. Would you expect that to sort of ramp in the 2027 time frame[?]"
— Quinn Bolton, Needham & Company

A: "The power stages is really a standard design cycle… you can think about it as the 12 to 18 months qual cycle in production… we're tied up… on a road map specific with the XPU suppliers. So as they ramp and deploy their new platforms, we will be ramping with them."
— Hassane El-Khoury, President & CEO

Assessment: The cadence confirms AI data center power is a 2026-and-beyond revenue event, gated by the XPU vendors' 800-volt deployment timelines rather than by onsemi's own readiness. The franchise is real and the roadmap engagement is deep, but the revenue is not a 2025 story, which keeps it in the optionality bucket for valuation purposes.

Silicon-Carbide Durability: Die vs. Modules and the Hybrid-Inverter Debate

A two-part line of questioning tested the durability of the SiC franchise: a customer shift toward buying die rather than modules, and a large customer's stated intent to cut silicon-carbide content ~75% in favor of a hybrid SiC/IGBT traction inverter. Management argued neither changes its market view, citing die/module flexibility and a long-held expectation that the high-performance tier stays SiC while mainstream migrates to hybrid and IGBT.

Q: "One of the big consumers in silicon carbide has discussed a migration to hybrid inverter from silicon carbide to silicon carbide combined with IGBT… reduce SiC by 75% in that traction inverter… I wonder if you're seeing that influence your go-forward view of this market[?]"
— William Stein, Truist Securities

A: "No change… higher end, higher range, higher performance will be silicon carbide. And as you go to more mainstream, you'll end up with a hybrid all the way down to… IGBT… We will continue to win based on the performance of the products specifically."
— Hassane El-Khoury, President & CEO

Assessment: Management's answer is reasonable: onsemi's SiC-plus-IGBT breadth lets it serve whichever architecture an OEM chooses, and die/module flexibility is a feature, not a threat. Still, the hybrid-inverter trend is a genuine medium-term governor on SiC content-per-vehicle that bears watching, even if it does not break the franchise thesis.

Section 232 Tariffs and the Diversified-Footprint Defense

A question on how onsemi is preparing for potential Section 232 semiconductor tariffs drew a deliberately non-committal answer centered on controllables: the company's geographically diversified manufacturing footprint as a customer-valued sourcing hedge, and a refusal to plan around an outcome that is unknowable.

Q: "Section 232, 301. Obviously, a lot of noise around that. How are you guys preparing for that? What are you expecting in terms of when this happens?"
— Vijay Rakesh, Mizuho

A: "I don't know… if anybody told you I have a crystal ball better than my peers… we remain focused on what we can control… our manufacturing footprint as a competitive advantage… we need to maintain flexibility… the focus on what we can control."
— Hassane El-Khoury, President & CEO

Assessment: The footprint-diversity argument is genuine and is one of the few tariff hedges a power-semiconductor supplier can credibly claim. Management's refusal to forecast the tariff outcome is appropriate, but it leaves Section 232 as an unquantified overhang that the multiple will carry until the policy resolves.

What They're NOT Saying

1. No date for the recovery: Management repeatedly refused to "call a recovery" or to commit to taking utilization up, deferring to "better visibility into Q4 and early next year." The absence of any second-half acceleration framework is the loudest silence on the call.
2. No size for AI data center: The fastest-growing, highest-quality part of the portfolio is buried in "Other" and management explicitly declined to break it out ("we may talk about it in more detail" once it is "sizable"). Investors cannot yet model the one business that would justify a multiple premium.
3. No clean answer on the auto peer-gap: The question of why onsemi's auto business is ~30% below peak while peers are within 4–5% was answered with "mix and exposure" plus portfolio rationalization, without quantifying how much of the gap is structural share/mix versus cyclical.
4. SiC content-per-vehicle risk soft-pedaled: A major customer's intent to cut SiC content ~75% via hybrid inverters was waved off as "no change," with no discussion of the content-per-vehicle implications if the hybrid architecture proliferates faster than expected.
5. Pricing characterized as "stable," with little detail: After flagging more aggressive share-defense pricing in the prior quarter, management said pricing is now "stable" and "within expectations," but offered no specifics on blended ASP trends in a still-soft demand environment.

Market Reaction

• Pre-print setup: ON closed at $56.82 on Friday Aug 1, having recovered substantially off its 52-week closing low of $31.95. The stock was down 9.9% year-to-date and 16.4% over the trailing twelve months, but up modestly (+0.4%) over the trailing 30 days, reflecting a market that had begun to anticipate a cyclical turn.
• Reaction-day move (Aug 4, before-open report): The stock gapped down ~5% at the open and closed at $47.97, down 15.6% ($8.85), on volume of 30.5M shares versus an 8.5M-share 30-day average (3.6x normal).
• Intraday: Traded a $47.32–$54.84 range, finishing near the low.
• Market context: The S&P 500 rose 1.5% on the session, so onsemi's decline was entirely idiosyncratic, not a risk-off tape.

A 15.6% single-day decline on a revenue beat and an in-line EPS print demands explanation, and three dynamics combined.

Recovery not confirmed: The stock had rallied off its lows in anticipation of a cyclical turn. When management explicitly declined to call a recovery and guided gross margin flat, the buyers who had positioned for confirmation had their thesis denied and sold. This is the dominant driver: a positioning unwind on a "stabilization, not recovery" message.

Margin disappointment: The 37.6% gross margin and the flat Q3 guide, paired with ~900bps of under-absorption and a 53% target gated on demand, told the market the earnings power is capped until volume returns. For a stock that had re-rated on recovery hopes, a deferred margin path is a de-rating catalyst.

The shrinking base: The disclosure that ~5% of 2025 revenue (~$300M) will not repeat in 2026 reframed the forward model. Investors now have to underwrite a smaller 2026 base recovering, rather than a stable base accelerating, which lowers the trajectory of every forward estimate. We read the 15.6% move as an overshoot driven by positioning, but a justified de-rating in direction given the deferred-recovery message.

Street Perspective

Debate: Is This the Cyclical Bottom or a False Dawn?

Bull view: Automotive troughed in Q2 and is guided up in Q3, all three end-markets are guided higher, China SiC is surging, inventory is lean across onsemi and the channel (disti at 10.8 weeks, within the 9–11 target), and a lean setup means utilization and margin snap back quickly when demand returns. The bottom is in; the only question is slope.

Bear view: "Stabilization" is not recovery, traditional industrial is still declining, Western EV demand is genuinely weak, and management itself refuses to call the turn. The 2026 base shrinks ~5% before any growth, so even a cyclical recovery may not produce year-over-year revenue growth until well into 2026. A bottom that does not grow is a value trap at 20x.

Our take: The trough is credible but the recovery is unscheduled. We side with neither camp outright: the floor looks real, but a flat-margin guide and a shrinking base mean the recovery does not pay off in near-term numbers. That is the textbook setup for Hold.

Debate: Can the 53% Gross-Margin Target Be Reached on a Realistic Horizon?

Bull view: The bridge is concrete: ~700bps from utilization (with full utilization reset to ~92%), ~200bps from Fab Right and divested-fab monetization, plus mix and new-product accretion (Treo, SiC, AI power). As volume returns to a smaller, higher-quality footprint, operating leverage is powerful and 50%+ is achievable.

Bear view: More than 1,500bps of margin uplift from today's 37.6%, with the largest block entirely demand-dependent and a management team that will not forecast the demand. The 53% target has been a multi-year "someday" number, and under-utilization could persist through 2026 if the cycle stays soft. Mid-40s, not 53%, is the realistic medium-term ceiling.

Our take: Lean bear on timing, neutral on the destination. The levers are real and the math holds, but the timeline is long and gated on a recovery onsemi will not call. We model gross margin in the high-30s near-term, rising toward the mid-40s as utilization recovers in 2026, with 50%+ a 2027-plus aspiration contingent on a full cyclical upturn.

Debate: Does the AI Data Center Power Franchise Justify a Premium?

Bull view: onsemi is the only broad-based U.S. power-semiconductor supplier addressing the 800-volt DC data-center transition, partnered with NVIDIA, with AI data center revenue doubling year-over-year for two straight quarters. As the XPU vendors deploy 800-volt racks in 2026–2027, this becomes a high-margin secular growth leg that re-rates the whole company.

Bear view: It is too small to matter near-term, management will not even size it, and the ramp is gated by other companies' deployment timelines. Power semiconductors for AI is a competitive field, and design-in is not the same as sole-source dollar revenue. The premium is speculative until the numbers show up.

Our take: Real option, not yet a line item. We assign the franchise meaningful long-term value but no material 2025–2026 revenue weight. If management begins to size AI data center revenue and it scales as the 800-volt cycle arrives, that is a concrete upgrade catalyst we would act on.

Model Implications & Valuation Framework

Valuation framework: At the post-print price of ~$47.97 and our FY25 non-GAAP EPS estimate of ~$2.20–2.35, ON trades at roughly 20–22x trough earnings, and ~17–19x our FY26 estimate of ~$2.50–2.90. For a high-quality cyclical with a depressed-but-recovering margin, a genuine SiC/AI-power growth option, and disciplined capital return, that multiple is fair rather than cheap: it neither prices in a sharp recovery nor implies one is being ignored. The stock's >50% peak-to-trough decline (52-week range $31.95–$78.25) reflects the cycle the company is living through.

Price-target framework (12-month):

• Base case ~$50–55: ~19–21x our ~$2.65 FY26 EPS midpoint. Recovery proceeds gradually, utilization begins to lift in H1 2026, margins drift toward the low-40s. Implies roughly flat-to-modestly-higher from the post-print price.
• Bull case ~$65–72: ~22–24x a ~$3.00+ FY26 EPS if the cycle inflects faster, utilization recovers sharply, and AI data center plus SiC drive mix accretion. A clean re-rate to a recovery multiple.
• Bear case ~$36–40: ~17–18x a ~$2.20 FY26 EPS if the recovery stalls, traditional industrial and Western auto stay soft, and the ~5% exit headwind produces another year of declining revenue with margins pinned in the high-30s.

The base case implies modest upside from $47.97; the bull/bear spread is roughly symmetric. That balanced risk/reward, with the swing factor being a cyclical recovery management itself will not yet call, is the quantitative basis for the Hold.

Thesis Scorecard: Establishing the Framework

As this is our initiation of coverage, the scorecard below establishes the bull and bear pillars we will track each quarter. There is no prior thesis to grade; future recaps will mark each pillar Confirmed / Challenged / Neutral against this baseline.

Overall: The bull pillars are structurally attractive but back-end-loaded; the bear pillars are all currently active and explain the trough. The framework is balanced, with the swing variable being the timing and slope of the cyclical recovery.

Action: Initiate at Hold. Wait for utilization to inflect, for auto to print year-over-year growth, and for AI data center to be sized before paying for the recovery.

Bottom Line: A Good Business at the Wrong Point in Its Cycle

Rating decision: We initiate coverage of onsemi at Hold. This is a structurally improving company (intelligent power, silicon carbide, AI data center, the Treo analog platform) running through a deep cyclical trough in its two largest end-markets, with a credible but demand-gated path to materially higher margins and a disciplined capital-return program supporting the floor. The Q2 print was a beat-and-meet that the market sold 15.6% because the message (stabilization without recovery, flat margins, a shrinking 2026 base) denied the cyclical-turn thesis that the stock had rallied to price.

At ~$47.97 the stock trades at roughly 20x trough earnings, which is fair for a cyclical at the bottom but not the kind of mispricing that compels an Outperform. The transformation is real and the secular engines are turning on, but they pay off in 2026–2028, and the near-term P&L is capped by under-absorption. We would rather wait for evidence (utilization inflecting, auto growing year-over-year, AI data center sized and scaling) than underwrite a recovery management itself will not yet call.

What would move us to Outperform: Utilization inflecting upward with gross margin breaking decisively above 40%; automotive returning to year-over-year growth; AI data center revenue being disclosed and scaling toward a material share; or a further de-rating that prices in a permanent trough the fundamentals don't support.

What would move us to Underperform: The recovery stalling into 2026 with traditional industrial and Western auto still declining; gross margin stuck in the high-30s for multiple additional quarters; or the ~5% revenue exit headwind compounding into a second consecutive year of declining revenue without a margin offset.

Signposts for Q3 2025 earnings (early November 2025):