ON — onsemi (ON Semiconductor)

Executive verdict

If you want exposure to the SiC/GaN high-voltage thesis Irrational Analysis is pounding the table on, onsemi is the cleanest vehicle among the four comps. It is a $6B-revenue, $1.4B-FCF, $46B-mcap operating business with real EBITDA, a vertically integrated SiC franchise built on the 2021 GT Advanced Technologies acquisition, and a coherent strategic narrative that ties EV traction inverters, industrial power, and AI datacenter power into a single product portfolio (Power Solutions Group, ~50% of revenue). It does not require you to underwrite a 10× revenue ramp like NVTS does. It does not have an insider-selling cliff like NVTS did. And — uniquely among the four — management bought back $1.38B of stock in FY2025, the trough year, when the price was ~$45–80. That is the kind of capital allocation that earns the benefit of the doubt.

The catch is timing. Onsemi is at $118.37, up +191% from the $40.62 low set in mid-2025, at the 52-week high, above the 29-analyst mean target of $103.45 (−13% implied downside), at 28× forward earnings on still-recovering EPS. The automotive cyclical bottom is real and Q1 2026 delivered the first YoY revenue growth (+4.7%) in two years, but the stock has already priced two quarters of acceleration. Gross margins remain compressed (33% in FY2025 vs 47% in FY2023) and will take 2–3 years to fully recover. The bull case from here is mid-teens revenue growth + 8pp gross margin recovery + multiple-holding, which gets you to perhaps $135–145 by end-2026. Not a bad return, but not a fat margin of safety either.

Action: WATCH, with a buy zone of $90–100 for a starter (5% portfolio weight) and $75–85 for a full position. If Q2 2026 print (early August) extends the YoY revenue acceleration to +8–10%, the bull case firms and the entry window may close at $105–110.

What the company actually does

Onsemi sells three things, roughly in order of revenue weight: (1) power semiconductors — discrete IGBTs, MOSFETs, EliteSiC silicon carbide MOSFETs and modules, GaN HEMT devices, integrated power modules — into automotive electrification, industrial drives, AI data center power supplies, and consumer fast-charging; (2) analog and mixed-signal — power management, sensor interface, connectivity ICs for the same end markets; (3) CMOS image sensors for automotive vision (the legacy Aptina franchise from the 2014 acquisition), industrial automation, and security. The three segments are run as Power Solutions Group (PSG), Analog and Mixed-Signal Group (AMG), and Intelligent Sensing Group (ISG).

Power Solutions is the franchise — it's where the SiC story lives, where the AI datacenter power story is starting to live, and where most of the recent margin pressure showed up. EliteSiC, the branded MOSFET line, is the product that matters for this thesis. Onsemi runs SiC vertically: they grow their own 200mm SiC boules at the former GT Advanced facility in New Hampshire, slice their own wafers, and fab their own devices. That vertical model is rare — most SiC players (Wolfspeed had to, Infineon partially, II-VI/Coherent) had to scramble to assemble wafer supply during the 2022–2023 SiC capacity crunch. Onsemi paid $415M for GTAT in 2021 and has spent another ~$2B+ on SiC capacity since. That bet is the moat. It is not a faith bet like Navitas's vertical GaN claim; it ships volume today.

Geographic mix is roughly one-third North America, one-third EMEA, one-third Asia-Pacific. Automotive end market is the single largest exposure at ~55–60% of revenue, which is also why the FY2024–FY2025 revenue decline tracked the global EV/auto inventory destocking so closely.

First principles — why the SiC vertical model matters

Silicon carbide power devices need three things to ship at volume: a wide-bandgap crystal (SiC boule), wafers sliced from that boule, and the device-fab capability to build MOSFETs on those wafers. The boule-growth step is the bottleneck. Growing a single 200mm SiC boule takes a week or more in a physical vapor transport reactor at 2,500°C+, vs hours for a comparable silicon boule. Defect rates (micropipes, basal plane dislocations) are orders of magnitude higher than silicon. Yield on the device side is sensitive to wafer defect density, so the supply chain has historically been controlled by whoever owns the boule-growth capacity: Wolfspeed (pioneer, now restructured), II-VI/Coherent, SK Siltron CSS, ROHM, and onsemi via GTAT.

The 2022–2023 SiC crunch happened because EV demand inflected before boule capacity could ramp. Wolfspeed went into financial stress trying to fund the buildout. Onsemi's vertical position let them lock in supply to Tier 1 automakers (VW, Hyundai, BMW, Stellantis all have named EliteSiC design wins) and price aggressively. As the 2024–2025 EV slowdown hit, that pricing power compressed — the auto OEMs renegotiated long-term supply agreements (LTSAs), and onsemi took the hit on margin (47% GM FY23 → 33% FY25) but kept the volume commitments. The volume comes back in the next cycle up; the margin compression may be partly structural (LTSAs at lower prices) and partly cyclical (capacity underutilization absorbing fixed costs).

GaN is a smaller piece of onsemi's portfolio. They announced a vertical GaN-on-GaN technology in 2024 targeting 1200V applications, but as Irrational Analysis correctly flags, there are no public datasheets and no shipping products yet. The vertical-GaN claim is best treated as optionality, not as a current revenue line. It is the part of onsemi's story that requires faith.

Where onsemi sits in the value chain

[SiC boules ★] → [SiC wafers ★] → [SiC devices ★] → [SiC modules ★] → [Tier 1 auto / industrial / DC] → [EV / drive / rack]
  Onsemi internal    Onsemi internal    Onsemi fabs       Onsemi pkg
  (GTAT NH)          (NH/Czech)         (Korea, US)       (multi)

Vertical integration from boule to module is the structural difference between onsemi and the SiC pure-plays (Wolfspeed) or the multi-sourced incumbents (Infineon, STMicro, Renesas). The benefit shows up in two places: gross margin durability when wafer supply is tight (onsemi captured the 2022–2023 SiC supply premium internally), and customer LTSA wins (Tier 1 automakers preferred a fully-integrated supplier for multi-year EV programs). The cost is capital intensity — onsemi has spent ~$3B+ over five years building this position, and the depreciation tail is the source of half of the FY2025 margin compression.

Upstream of onsemi the only meaningful bottleneck is the SiC epitaxial-equipment supplier — primarily Aixtron (AIXA, already in Pink's vault) for MOCVD reactors and TEL for wafer-handling. These are bottleneck candidates for SiC volume regardless of which device-maker wins, which is part of why Aixtron remains a cleaner sector long.

Customer concentration

Onsemi does not name its automotive customers in public filings beyond a generic "five-or-six Tier 1 automakers" disclosure, but the trade press has confirmed multi-year EliteSiC LTSAs with VW Group, Hyundai/Kia, BMW, Stellantis, and most recently Honda. Top-10 customers represent roughly 35–40% of revenue per the FY2024 10-K. No single customer above 10%. Concentration risk is moderate but not idiosyncratic — onsemi is exposed to the auto cycle in aggregate rather than to any one OEM's program. The AI datacenter business is a much smaller piece (probably <5% of revenue today) but the customer set there is different: Astera Labs, Vicor, and the hyperscaler power-supply OEMs who feed Nvidia/AMD reference designs.

The cleanest concentration play is the EV traction inverter, where onsemi has won at minimum five named Tier 1 platforms with $4B+ of cumulative LTSA value disclosed as of late 2025. That LTSA backlog is the floor under the bull case.

Sector inflection — where onsemi sits in it

The high-V SiC/GaN thesis Irrational Analysis frames is real and onsemi participates in three legs of it:

1. EV traction inverters (the SiC story): Tier 1 LTSAs locked, volume cyclical-bottom in 2025, 2026–2028 should see auto SiC volume +50–80% vs 2025 trough. This is onsemi's biggest leg by revenue weight.
2. AI datacenter rack power (the GaN + SiC story): 800V DC architecture transition starting 2026, onsemi has both SiC (1200V for rack-level conversion) and emerging vertical GaN (650V for PSU). Small revenue today, asymmetric upside if hyperscaler adoption ramps.
3. Industrial power (the cycle story): drives, robotics, factory automation — flat through the auto downcycle, expected to pick up with the industrial capex cycle in 2026–2027.

The Why now is specifically that the auto-SiC inventory destocking ended in late 2025 and Q1 2026 was the first YoY revenue positive quarter (+4.7%). EPS recovered before revenue did (Q1 2026 EPS $0.64 vs Q1 2025 $0.53, +21% on +4.7% revenue = operating leverage starting to work in reverse-of-reverse). The cycle position is right. The valuation reflects part of this already.

What the sell-side is currently missing is the margin trajectory durability question. Consensus models a snapback to 40%+ gross margin by FY2027. Onsemi's LTSA pricing locked in during 2024 means the snapback will be slower than that — call it 36–38% by FY2027, then 40%+ by FY2028. If consensus has to lower FY2027 GM estimates by 2pp, that's a ~$0.30 EPS revision and a ~$8 stock impact at 28× — manageable, not catastrophic.

Numbers, with the cycle position front and center

The Core Four read:

1. Organic revenue growth has been negative for two years and just turned positive. FY2023 $8.25B → FY2024 $7.08B (−14%) → FY2025 $6.00B (−15%). Quarterly: $1.45B → $1.47B → $1.55B → $1.53B → $1.51B Q1 2026, +4.7% YoY — first YoY positive quarter in 8 quarters. Q2 2026 consensus is +8–10% YoY = ~$1.58–1.62B. The acceleration is real but priced.

2. Margins are recovering more slowly than the bulls want. Gross margin FY2023 47.0% → FY2024 45.5% → FY2025 33.0%. Trough was Q3 2025 (~32%). Recovery path requires both utilization (capacity absorbing fixed costs) and LTSA renegotiation cycles. Operating margin: FY2023 31.6% → FY2024 26.8% → FY2025 12.5%. The op-margin trough is mostly behind us but the snapback is 12–18 months, not 2–3 quarters.

3. Capital intensity is high — onsemi has spent ~$3B+ in cumulative SiC capex over five years, plus the GTAT acquisition. Capex peaked at $1.54B in FY2023 (during the buildout) and has since dropped to $340M in FY2025 as the major SiC facilities completed. That capex unwind is what's driving the FCF inflection ($440M FCF FY23 → $1.21B FY24 → $1.42B FY25 even on lower revenue). Capital intensity is now structurally lower and will support buybacks.

4. Capital deployment is the surprise positive. Onsemi spent $1.38B on buybacks in FY2025, vs $650M FY2024 and $560M FY2023. The FY2025 buyback was executed when the stock was trading $45–80 — i.e. at the trough. That's a genuinely well-timed buyback program: management bought back 6% of shares outstanding at half the current price. No dividend (rare for a power-semi this size; preserved for buybacks). This is the opposite pattern of TXN's buyback timing. Capital allocation grade: A−.

Quarterly EPS trajectory (the operating leverage tell):

Q	EPS Actual	EPS Estimate	Surprise
Q2 2025	$0.53	$0.53	0%
Q3 2025	$0.63	$0.59	+6.6%
Q4 2025	$0.64	$0.62	+2.6%
Q1 2026	$0.64	$0.62	+4.1%

EPS has stabilized at $0.63–0.64 and beats consensus by 2.5–6.6% in each of the last three quarters. Operating leverage is starting to work — small revenue gains are dropping disproportionately to EPS as fixed costs get absorbed. Forward EPS consensus is $4.27 for FY2026. At $118 that's a 28× forward P/E, which would compress to ~22× on consensus FY2027 EPS in the high $5s if the leverage continues.

Balance sheet is solid. $2.15B cash, $3.0B debt, $7.67B equity, current ratio 4.9, debt/equity 44%. No solvency concerns. PEG ratio 0.34 looks attractive, but that's misleading — PEG uses 5-year EPS growth which is anchored on a trough year and overstates true growth durability.

Management — clean enough

CEO Hassane El-Khoury has run onsemi since December 2020. He came from Cypress Semiconductor (sold to Infineon) and is the architect of the strategic pivot toward Power + Sensing + EV — he's the right operator for this thesis. CFO Thad Trent joined in 2021 from Cypress as well. Both are aligned by background and have ridden the full cycle together. CEO comp $1.24M base, CFO $1.19M — normal levels for a $46B-mcap semi; the meaningful comp is stock-based.

Insider activity in 2026:

Insider	Role	Shares	$ Sold	Avg price
Thad Trent	CFO	60,000	$5.79M	$93–100
Thad Trent	CFO	30,000	$2.40M	$80
Thad Trent	CFO	60,000	$4.27M	$71–73
Hassane El-Khoury	CEO	20,000	$1.46M	$73
Simon Keeton	Officer	48,860	$3.51M	$71–72
Sudhir Gopalswamy	Officer	6,114	$0.36M	$58

~$18M total in C-suite + officer sales in Feb–Apr 2026 as the stock ran from $58 to $100. CFO Trent is the biggest seller at $12.5M. These look largely like 10b5-1 plan sales aligned to vesting cycles (Feb 5 and Feb 20 are typical Q4 RSU vest dates for tech companies) — the CEO selling only $1.5M out of a multi-million-dollar holding is consistent with "diversification, not signal." The pattern is materially less concerning than NVTS's (where directors with no operating role sold the bulk), but it's worth flagging that the CFO is the largest seller and the rate stepped up as the stock rose from $70 to $100.

Total insider ownership: 0.33% (extremely low — onsemi has been public a long time and the founder-equity history is exhausted). Institutional ownership: ~90% adjusted (yfinance shows 112.9% which is a known data artifact; real number is the bulk of the float). Short interest: 10.2% of float, 2.88 days to cover — not in the squeeze-risk zone.

Management grade: B+. Clean governance, no shell-entity flags, no related-party transactions in the most recent DEF 14A worth flagging, capital allocation track record is A−, the only yellow flag is the CFO's aggressive sale rate into the rally.

Competition

Direct SiC power-semi peers: Wolfspeed (recently out of bankruptcy, SiC-only, no GaN), Infineon (broader power-semi but lighter SiC), STMicro (large SiC presence, Tesla design wins), Renesas, ROHM. The China side: BYD Semiconductor, BHi Microelectronics (rising). In GaN onsemi competes with Navitas, Infineon, Texas Instruments, Power Integrations, Innoscience.

Moat-quality read: onsemi's SiC vertical integration is a structural moat that took ~$3B and five years to build. Wolfspeed has it on the boule side but lacks the device-fab scale. Infineon is closing the gap via the Cree wafer acquisition (2021) but is still partially externally sourced. The Tier 1 automakers prefer onsemi for multi-year programs specifically because of the vertical position. The moat is real and durable through this cycle. The GaN business is a different question — there onsemi has no moat yet, just an option.

The 5-year lock-up test: if you couldn't sell for five years, would you happily own onsemi? Mixed answer. Yes on the SiC franchise (real moat, secular EV demand, durable cash flow). Less so on the cycle timing — you'd be buying at +191% off the low. Quality verdict: durable. Valuation question is separate.

Why now — but with timing nuance

The cycle position is right. Auto inventory destocking is done, EV sell-through is reaccelerating, industrial is bottoming. Onsemi participates in all three. SiC content per vehicle is a long-tail secular tailwind (a single EV traction inverter uses 24–36 SiC MOSFETs at $5–10 each = $150–350 SiC content vs. $30–80 silicon IGBT content). The forced 800V DC datacenter transition is an emerging optionality leg.

The reason "now" is partially priced is that the stock anticipated this. The +191% rally from $40 to $118 happened in the eight months between the August 2025 low and now. The first +YoY-revenue print didn't happen until Q1 2026 (reported May 2026), and the stock had already tripled before then. The remaining upside compresses to "execution on already-priced expectations."

Risks, structured

Risk	Likelihood	Mitigants	Mgmt de-risk plan	Closable?
EV demand re-disappoints in H2 2026	Medium	LTSA backlog covers revenue floor	Diversification across 5+ Tier 1 OEMs	Partially
Gross margin recovery slower than consensus	Medium-High	Buyback supports EPS through any GM miss	Mix shift toward higher-margin AI/industrial	Yes, over 2-3 yrs
SiC ASP compression as Chinese supply ramps (BYD, BHi)	Medium	Vertical model offsets ASP pressure with internal margin	Geographic mix shift away from China-only	Partially
Vertical GaN claim fails to ship	Low impact	Small revenue contribution today	TBD — no public roadmap	This is optionality, not core
Wolfspeed post-bankruptcy price warfare	Low-Medium	Onsemi has cost advantage	LTSA pricing locks in volume	Yes, on Wolfspeed financial constraints
Multiple compression as cycle matures	High	Buyback yield supports floor	None (this is the cycle)	This IS the bear scenario
AI datacenter ramp delayed beyond 2027	Medium	Not load-bearing for current valuation	Continued investment in vertical GaN	Yes, on named customer disclosures

Bear case: Q2 2026 prints in-line revenue +6–7% YoY, gross margin only inches up to 34%, consensus FY2027 EPS estimates come down 5%. Stock retraces to $90–100 (50d MA $77, recent breakout zone $95). Downside target: $92 (−22%).

Base case: Cycle plays out roughly as consensus expects. FY2026 revenue +8% to $6.5B, FY2027 +12% to $7.3B, gross margin recovers to 38% by FY2027. EPS $4.30 FY2026 → $5.60 FY2027. Stock holds 24–26× forward = $135–145 by mid-2027 (+14% to +23%).

Bull case: Cycle is stronger than expected. AI datacenter power ramp adds 3–5pp to FY2027 revenue growth. Gross margin recovery accelerates with LTSA renegotiations expiring. FY2027 EPS $6.20. 28× forward = $175 (+48%).

Probability-weighted view: Bear 30%, Base 50%, Bull 20%. Expected return ~+12% over 12–18 months at $118 entry. Acceptable, not exciting. The same arithmetic at $95 entry gets you to ~+35% expected return, which is the price point worth waiting for.

What needs to be true for $118 to make sense

Working backward from a 22× P/E on FY2027 earnings (the multiple a recovered analog/power semi typically trades):

• FY2027 revenue: $7.0B+ (consensus $7.3B — achievable)
• FY2027 gross margin: 38%+ (consensus 38% — achievable but not slam-dunk)
• FY2027 operating margin: 22%+ (FY2024 was 27%; recovery to 22% is reasonable)
• FY2027 EPS: $5.50–6.00
• 22× $5.75 = $127 — slightly above current price

So $118 is approximately fair for the base case. To get a real margin of safety you need either FY2027 EPS to surprise to the upside (>$6.00, requires stronger SiC pricing recovery) or multiple expansion (25×+, requires the bull case AI datacenter optionality to firm up). The current entry assumes everything goes right with no buffer. That's the case for waiting.

Ownership & sentiment

• Institutional ownership: ~90% (large index + active core)
• Insider ownership: 0.33% (low; long-public company history)
• Short interest: 10.2% of float, 2.88 days to cover
• 29 analysts covering, consensus BUY, mean PT $103.45 (range $68–125)
• Stock at $118 = 14% above mean PT

Coverage is deep; this is not an unknown-name idea. The mean PT below spot suggests sell-side has not raised numbers fast enough — there will be PT bumps in the coming weeks as Q1 prints are absorbed. Wait for the PT revisions before paying $118+.

Position sizing and entry discipline

WATCH at $118. Better entries:

• $100–105 — initiate 3% portfolio starter. Aligns with breakout retest zone and analyst PT.
• $90–95 — initiate 5% portfolio core position. Aligns with prior consolidation zone, sub-mean-PT.
• $75–85 — initiate 7%+ portfolio conviction position. Aligns with 50d MA and pre-rally base.

Triggers that change the buy zone:

• Q2 2026 print (early August): if revenue +10%+ YoY AND gross margin recovers to 35%+, the entry window may close at $105 — initiate at $105 instead of waiting for $95.
• Named 800V DC datacenter customer announcement: re-rate the bull case, raise targets by 10–15%.
• China SiC ASP collapse (BYD pricing data, channel checks): bear case fires, $75–85 zone becomes likely.

Invalidation: EliteSiC loses a named Tier 1 automotive program OR FY2027 consensus EPS falls below $5.00. Either signals the moat narrative is breaking.

Comp-set ranking — final read for the screen

For the SiC/GaN high-voltage thesis, ranked by risk-adjusted attractiveness at current prices:

1. ON @ $118 — WATCH (waiting for $90–100). Real business, real moat, fair-to-slightly-rich price.
2. NVTS @ $22 — PASS (waiting for $14). Real thesis, real product, but valuation requires faith in unproven ramp.
3. TXN @ $308 — WRONG VEHICLE. SiC/GaN <5% of revenue. Own for analog cycle, not this thesis.
4. IFX @ $78 — DIVERSIFIED PROXY. Broader exposure, IA-flagged perf laggard, fully-valued.

Pink's action: if she must own one today, none of the four are the right entry. The SiC/GaN exposure she already has via AIXA in the vault is the cleanest current position. The next addition should be ON at $90–100, not now.

Sources

• Live yfinance pull, 2026-05-15 (financials, balance sheet, cash flow, insider transactions, earnings history, recommendations)
• 2026-04-25-power-semis-part-2-sic-vs-gan-wolfspeed — Irrational Analysis Substack, archived
• 2026-04-21-power-semis-800v-dc-and-gan-mini-note — Irrational Analysis Substack, archived
• SEC EDGAR — ON 10-K FY2025, 10-Q Q1 2026, Form 4 insider filings Feb–Apr 2026 (referenced via yfinance; full DEF 14A reading deferred)
• NVTS — comp-set partner deep-dive (NVTS = pure GaN, ON = vertical SiC, complementary not overlapping)
• AIXA — upstream MOCVD/SiC epi equipment exposure (cleaner sector long)
• AEHR — SiC/GaN burn-in test adjacent

Cross-refs

• NVTS — pure-play GaN alternative; different thesis vehicle, different risk profile
• TXN (to be written) — wrong vehicle for this thesis; full page deferred
• IFX (to be written) — diversified power semi proxy; IA called out as performance laggard
• themes/sic-gan-high-voltage-thesis (to be written) — sector thesis page consolidating both deep-dives + IA posts
• AIXA — MOCVD reactor; cleaner sector exposure
• SOITEC — engineered SiC substrate monopoly
• AEHR — SiC/GaN burn-in test
• Wolfspeed — explicitly excluded per Pink's request; competitive context only

SA cross-check

SemiAnalysis mirror returned older 2022 power-semi roundup mentions of onsemi but no recent deep-dive specifically on EliteSiC or the 800V DC architecture. SA's broader AI Datacenter Anatomy series supports the sector thesis. No contradictions to flag. SA covered the SiC supply crunch favorably for vertically-integrated players in 2022–2023 — that view aged well for onsemi specifically.

Briefings

• 2026-05-15 · onsemi (ON Semiconductor) (ON) · vault