On Semiconductor ON

Thesis

On Semiconductor (Nasdaq: ON) designs and manufactures power-management and analog/sensing semiconductors organized into three segments: Power Solutions Group (silicon carbide and silicon power devices for EVs, industrial drives, renewable energy), Analog Mixed-Signal Group, and Intelligent Sensing Group (CMOS image sensors leading the automotive ADAS market). The compounding case is straightforward: an entrenched #1/#2 share position in automotive silicon carbide and image sensors, manufacturing scale (Hudson and East Fishkill 200mm/300mm fabs plus the Czech SiC ramp), and a 10-year ROIC average of only 8.3% that masks a recent ROIIC of 57.5% — meaning the marginal dollar reinvested into SiC and ADAS has earned spectacular returns even though the legacy book drags the long-run average. FCF conversion of 1.43x signals real cash generation behind GAAP earnings (TTM owner earnings $1.95B). Net debt sits at 0.18x EBITDA — effectively a cash-rich balance sheet — and share count is essentially flat over a decade (+0.33%). The entire opportunity is the gap between price ($103.03) and intrinsic value: base IV $231.34 (P/IV 0.45), low-band $118.74. Reverse-DCF implies the market is pricing only 3.8% perpetual growth into a company whose 5-year incremental returns argue for double-digit reinvestment runway. Composite score 83/100 (top quartile). The thesis is simple: buy a structurally-advantaged power-and-sensing duopolist near low-band IV, own through a cyclical inventory correction, and let mid-teens FCF growth plus multiple normalization compound.

Moat

On Semiconductor's competitive position rests primarily on three of the five classical moat types — cost advantages from scale, switching costs in design-in cycles, and intangibles in process IP — with weak brand and no network effects.

Cost advantages (PRIMARY). Damodaran observes that 'in businesses where scale can be used to reduce costs, economies of scale can give bigger firms advantages over smaller firms' [5]. ON owns its own fabs (the GlobalFoundries East Fishkill 300mm fab acquired in 2022, plus 200mm fabs in Hudson NY, Gresham OR, Roznov, and the new Bucheon SiC mega-site in Korea). Power semiconductors and SiC are capital-heavy with steep learning curves: capacity costs $4-8B for a leading-edge SiC fab and yield improvements take years. Three to five players (Infineon, STMicro, Wolfspeed, Onsemi, Rohm) capture nearly all the SiC value pool. Stress test: even a $10B/5-year well-funded entrant would struggle — Wolfspeed's pure-play attempt on the same playbook ended in Chapter 11 in 2025 because process yields and customer qualification cycles defeated capital alone. ON's vertically integrated 150mm-to-200mm SiC transition, paired with substrate self-supply, is a genuine structural cost edge.

Switching costs (STRONG). Once a power module or image sensor is designed into an automotive platform, the OEM commits to a 5-7 year program with re-qualification costs measured in tens of millions and 18-month requalification timelines. Damodaran's framing of Microsoft's switching-cost moat applies analogously: 'a user who has Microsoft Office installed... has to run multiple gauntlets' [5]. An automotive Tier-1 buying ON's image sensor or SiC traction inverter module faces AEC-Q100 requalification, software re-validation, and warranty re-baselining. This is why ON enjoys the kind of multi-year revenue visibility — 'long-term supply agreements' (LTSAs) — that pure-play commodity semis lack.

Intangibles (NARROW). ON holds thousands of patents in power MOSFET design, SiC trench architecture, and back-side-illuminated CMOS image sensors. Damodaran cautions that 'the companies that will see the greatest increases in value are not necessarily the companies that spend the most on R&D, but those who have the most productive R&D' [1]. ON's R&D-to-revenue runs ~12-13% — middle of the pack — but the productivity is real: it is the leader in automotive image sensors with ~50% share. Patents alone do not constitute a wide moat in semis (the cycle of imitation is fast), but combined with manufacturing know-how they raise the bar for new entrants.

Brand (NONE). Power semis are spec-driven; OEMs care about datasheet performance, supply continuity, and price — not the ON brand.

Network effects (NONE). No two-sided dynamics in semiconductor sales.

Erosion risk. The honest concern: Chinese power-semi entrants (Silan, BYD's in-house SiC, Hua Hong) are subsidized and can absorb losses. Damodaran reminds us 'in competitive sectors, though, the presence of these excess returns will attract new entrants and imitation will push excess returns down' [3]. Inventory cycles in automotive (2024-2025 destocking) compress pricing power, and the 'walking dead' problem Buffett describes for insurance [Failures-1] has a semis analog: subsidized competitors who keep capacity flooding the market past rational economics. Wolfspeed's bankruptcy actually helps ON medium-term by removing supply, but it shows how brutal SiC economics can be.

See's Candy taught Buffett that 'long-term competitive advantage in a stable industry is what we seek' [4]. Power semis are NOT a stable industry — they are deeply cyclical and capital-intensive. The moat protects margin during the upcycle but is tested during downcycles like 2025.

Moat verdict: NARROW.

Management & Capital Allocation

Hassane El-Khoury (CEO since Dec 2020, ex-Cypress) and CFO Thad Trent (formerly CFO at Cypress, where he and El-Khoury worked together pre-Infineon acquisition) inherited a sprawling, low-margin commodity semis business and have executed a deliberate transformation: exit low-margin standard products, concentrate on automotive/industrial 'intelligent power and sensing,' and vertically integrate SiC.

Reinvest (A). Capex peaked above $2B in 2023 to fund the East Fishkill 300mm conversion and the Bucheon SiC fab — counter-cyclical at the top of the cycle, which raised eyebrows but is exactly the textbook Buffett move ('be greedy when others are fearful') if SiC content per EV scales as forecast. The 5-year ROIIC of 57.5% is the proof that recent reinvestment has been spectacularly productive even though 10-year ROIC averages only 8.3%. The scorer's note that base CAGR was clamped from 23.1% to 14.0% reflects appropriate analytic caution — but the underlying engine is real.

Acquire (B). The 2022 East Fishkill fab purchase from GlobalFoundries (~$430M plus a multi-year wafer-supply commitment) was a brilliant deal — a scarce 300mm asset acquired below replacement cost. Earlier acquisitions (Fairchild 2016 for $2.4B, Aptina 2014, AMI 2008) were more mixed: Fairchild has been digested and turned into the Power Solutions backbone, but the integration drag depressed ROIC for years. Management has not done a major dilutive acquisition since 2016 and has shown discipline in walking away (e.g., Magnachip bid blocked by CFIUS concerns).

Debt (A). Net debt/EBITDA of 0.18x is essentially zero leverage for an industrial business. ON refinanced in 2023 with a new revolver and 0% convertible notes that raised cash on attractive terms. Interest coverage is so high the scorer reports null (numerator dominates).

Buybacks (B+). Management commits 50% of FCF to repurchases. They bought back stock through 2022-2024 at average prices in the $60-90 range — below current $103 and well below base IV of $231. P/IV at point of repurchase has been roughly 0.3-0.4, which is what you want from a buyback program. Share count has risen only 0.33% over a decade (extraordinary discipline given SBC pressures from a workforce of ~30,000). They are not, however, leaning hard into the current weakness as aggressively as a Buffett-style operator would.

Dividends (N/A). No dividend, which is correct: at 57% ROIIC, every retained dollar is more valuable in the business than in shareholder pockets — for now.

Communication. El-Khoury's earnings calls are unusually candid for the industry — explicit about destocking timelines, willing to say 'we will not chase share at the bottom,' and provide real LTSA disclosure. The 2025 inventory correction was flagged early and frankly. The 10-K segment disclosure (Power Solutions, Analog Mixed-Signal, Intelligent Sensing) is clean, with one customer above the 10% concentration threshold (the customary disclosure in the filings) — a real risk to acknowledge.

Concerns. SBC is meaningful (~$200M/year). The post-Wolfspeed competitive opportunity could tempt over-aggressive capacity additions. CEO compensation skews toward stock with multi-year performance triggers — aligned, but if they hit any of the short-term targets the optionality cost is real.

Capital allocator: B+.

Industry Structure

Porter's Five Forces applied to the power-and-sensing semiconductor segment ON occupies:

Threat of new entrants — MODERATE-HIGH. Capital intensity is the primary barrier (a leading SiC fab is $4-8B), and process know-how takes years. But government subsidies on three continents (US CHIPS Act, EU Chips Act, Made in China 2025) are lowering the effective cost of entry. Chinese competitors are aggressively entering automotive SiC backed by state capital. Wolfspeed's 2025 bankruptcy demonstrates entry is hard, but state-backed entry is qualitatively different from private capital entry.

Bargaining power of buyers — HIGH. Automotive OEMs and Tier-1s (Bosch, Continental, Magna, Aptiv) are large and concentrated. ON discloses one customer above 10% of revenue. OEMs use multi-sourcing as a deliberate procurement strategy — they explicitly avoid single-source dependence on any silicon supplier. Long-term supply agreements (LTSAs) signed in 2022-2023 actually mitigate this in the near term but face renegotiation pressure as inventory correction lingers in 2025.

Bargaining power of suppliers — LOW-MODERATE. ON is increasingly vertically integrated in SiC (substrates, epi, devices, modules). Wafer suppliers (where used) are concentrated (TSMC for some specialty processes), but ON owns most of its fab base. Specialty chemicals and lithography tools (ASML, Applied Materials) have monopoly-like power, but ON doesn't operate at the leading edge so its dependence is on mature toolsets.

Threat of substitutes — MODERATE. Silicon carbide is itself a substitute for silicon IGBTs in EV traction inverters. Gallium nitride (GaN) is the next candidate to substitute SiC in some applications. CMOS image sensors face limited substitution risk in automotive (LiDAR is complementary, not substitute, for ADAS). The deeper substitute threat is BEV adoption itself: if BEV growth slows, SiC TAM shrinks.

Industry rivalry — HIGH. Infineon, STMicro, Onsemi, Rohm, Wolfspeed (in restructuring), Renesas, Texas Instruments, and Chinese entrants are all chasing the same automotive SiC and ADAS opportunity. Pricing has cracked in 2024-2025 as inventory normalized. This is the most important force right now and it is biting.

Value pool location and trajectory. Damodaran observes that 'over time, there is a tendency, albeit slow, for the returns at companies to converge on industry averages' [3]. The value pool is migrating from horsepower-on-a-die generic logic toward application-specific power and sensing — favorable for ON. But within power semis, value is being competed away faster than secular content-per-vehicle growth is creating it. EV content per vehicle ($700-1500 of power semis vs. $200 in ICE) supports a ~10-12% TAM CAGR; supply rationalization (Wolfspeed exit, capex cuts at competitors) supports gross-margin recovery; but Chinese entrants are pulling pricing the other direction.

The combination produces a 'good-but-not-great' industry — better than DRAM and worse than analog/mixed-signal pure plays like ADI/TXN. ON's 8.3% 10-year ROIC reflects the cyclicality and capital intensity; the 57.5% 5-year ROIIC reflects the favorable transition the segment is in right now. The two numbers will probably converge to something in between (15-20%) at steady state.

Industry Verdict: Average — better than commodity DRAM, worse than the analog elite, with structural tailwinds partially offset by Chinese supply.

Inversion (Bear Case)

I am now the short-seller. I do not care about the bull thesis. Here is why ON is a value trap.

1. The single event that kills this. A two-year deferral of EV mass adoption. Tesla's growth has flattened, Ford has cut F-150 Lightning production, GM has pushed back EV targets, European OEMs are walking back 2030 ICE bans. If global BEV penetration stalls at 18-20% rather than continuing toward 40-50% by 2030, the entire SiC TAM thesis — every analyst spreadsheet that gets ON to a $231 IV — is wrong by half. SiC content per ICE vehicle is trivial; the entire excess return story rides on the BEV mix shift continuing at the previously-projected slope. Hybrid powertrains use less SiC than pure BEVs. ICE-extension extends ON's silicon power-MOSFET legacy book at sub-segment-average margins.

2. Why the moat is narrower than bulls think. Bulls quote 'first to 200mm SiC' and 'East Fishkill 300mm scale.' Reality: Infineon is bigger, better-capitalized, and has tighter relationships with European OEMs. STMicro has Tesla locked in. Rohm has Toyota. ON's flagship Bucheon SiC fab cost ~$2B and ramps into a market where prices have fallen 30%+ in 18 months. The Wolfspeed bankruptcy bulls cite as supply rationalization is just as easily read as 'this entire category cannot earn its cost of capital at current pricing.' Damodaran is explicit that 'in competitive sectors, though, the presence of these excess returns will attract new entrants and imitation will push excess returns down' [3]. Chinese SiC suppliers (Silan, San'an, BYD captive) do not need to earn cost of capital — they are subsidized. ON does. The 57.5% ROIIC is a backward-looking figure during the upcycle peak. Forward ROIIC on the Bucheon ramp may be negative for 24 months.

3. Why management is worse than it appears. El-Khoury's deliberate exit from non-core lines is fine in theory but cost ON revenue at the worst possible time. He committed counter-cyclical capex right at the cycle peak (2023) — a courageous-looking move that may turn out to be top-ticking the SiC capacity build. CFO Thad Trent's compensation, like the CEO's, is partly geared to share-price targets, which gives management an incentive to maintain buybacks even at uneconomic prices. Stock-based compensation of ~$200M/year is a real cost the GAAP earnings ignore — that's roughly 100bps of margin not captured in the headline ROIC. The 'one customer above 10%' disclosure is hiding revenue concentration risk that bulls wave away.

4. What bulls are extrapolating that won't hold. Bulls extrapolate (a) automotive SiC at 30%+ CAGR, (b) ON gross margin recovering to 47-48%, (c) ROIIC remaining above 30%, (d) buybacks at accretive prices. Each link is questionable. SiC TAM grows at 15-20% if BEV slopes are right and lower if not. Gross margin in automotive semis is structurally pressured by Chinese entrants and the fact that LTSAs are now being renegotiated DOWN, not up. ROIIC will revert to industry mean (10-15%) as Bucheon ramp absorbs incremental capital at peak depreciation. The buyback math only works if you trust the IV calculation — but the scorer itself flagged maintenance-capex uncertainty so wide it widened the IV range.

5. Valuation trap (multiple compression / regime change). TTM P/E of 28.4 is HIGHER than the trailing 10-year average of 48.9 only because the denominator collapsed; on normalized earnings, current P/E is closer to 18-22x. EV/FCF of 37 is rich for a cyclical hardware business that has just completed its largest capex cycle ever. The reverse-DCF implies 3.8% growth — bulls call this 'priced for stagnation' but a reasonable bear could say it is priced for a normal commodity-semi outcome (3-4% real growth, cyclical margin around 15% operating, ROIC reverting to 10%). In that world, IV is closer to $80-100, not $231. The base IV of $231 requires both the secular SiC ramp AND multiple expansion AND continued ROIIC at 30%+. Two of three is not enough.

The bull thesis is internally consistent but front-loaded with optimism — every assumption needs to land. The bear thesis just needs ONE of (a) BEV deceleration, (b) China supply pressure, (c) capex overrun on Bucheon to compress IV by 40%.

If I am right, the stock could be worth $65-75 within 2-3 years.

Lollapalooza Bias Check

Biases active in me, the analyst, right now:

Anchoring (HIGH). I am anchored on the scorecard's $231.34 base IV. The very act of starting from that number primes me to view $103 as 'cheap' rather than asking whether the IV inputs themselves are correct. The scorer's own note that base CAGR was clamped from 23.1% to 14.0% should actually anchor me toward MORE skepticism, not less — that clamp is a flag that the unconstrained model thought ON would grow at semiconductor-bubble rates, which is itself a recency-bias artifact.

Confirmation bias (HIGH). I came into this analysis aware that ON has a strong balance sheet, a high composite score (83/100), and a P/IV of 0.45. Once you see those numbers, you start hunting for reasons the bull case is right. I had to consciously reach for the inversion to surface the BEV deferral risk, the China supply pressure, and the LTSA renegotiation threat — none of those came up naturally during the bull-side reasoning.

Authority (MEDIUM). Hassane El-Khoury is widely respected in the analyst community; Cypress was a successful run; Trent has been a steady CFO. I am inclined to grade management at A-/B+ partly because they 'feel' competent. Buffett warns: 'if a business requires a superstar to produce great results, the business itself cannot be deemed great' [Buffett 2007, 4]. ON in 2026 still depends materially on management execution of the SiC ramp — the moat does not run the business by itself.

Recency (MEDIUM). The 2024-2025 destocking is recent and vivid. There is a real risk I am over-weighting it (and so is the market — that is the bull case) but also a risk I am UNDER-weighting it (assuming it ends in 2026 because I'm tired of waiting for a thesis that hasn't worked). The honest answer is that the inventory correction may take longer than the 6-12 months bulls expect, in which case the entry should be cheaper.

Commitment / consistency (LOW-MEDIUM). Once I write a buy thesis, I will be loath to flip later when fresh data arrives. The protection is to specify in advance the disconfirming evidence: if 2026 Q1 LTSA disclosures show pricing renegotiation worse than 10%, or if BEV global penetration shows a fourth consecutive quarter of deceleration, the thesis is wrong.

Social proof (MEDIUM). Buffett's Berkshire owned a small TXN-style analog position in the 2010s and the fundamentals investor crowd considers semis suddenly 'investable' for buy-and-hold. I should not assume that consensus 'compounder' framing is correct just because it is now common.

The lollapalooza danger here is that anchoring on a high IV plus confirmation bias plus social proof of 'semis-as-compounders' creates a coherent-feeling story that lets me underweight the bear. I have built in conservatism by setting target buy at low-band IV with discount, but I want to be honest that I am leaning bullish.

10-Year Outlook

Same fundamental business model in 2036? Probably yes. Power conversion and sensing are unavoidable functions in any electrified, automated economy. Even if BEV adoption slows, hybridization, industrial automation, data-center power, and renewables all consume power semis. ON will still be a power-and-sensing house in 2036.

Customer base larger? Likely. Automotive content per vehicle is a structural tailwind that survives even a slow BEV transition (hybrids consume more power semis than pure ICE). Industrial energy storage and grid power conversion are nascent. The data-center power story (48V architectures, GaN/SiC PSUs) is still early. Wider customer base than today seems likely.

Profit per customer higher? Uncertain. Pricing in semis trends down over time at fixed performance points (Moore's Law analog for power: cost-per-amp falls 5-7%/year). The bull case is content-per-vehicle and content-per-system rises faster than per-component price erodes. The bear case is Chinese supply forces price erosion to outpace content growth. Honest answer: 50/50.

Moat wider? Slightly. Manufacturing scale advantages tend to widen with each capex cycle if executed well — ON's Bucheon and East Fishkill ramps should be more productive in 2030 than 2026. But the LTSA-driven switching cost moat is at risk of erosion as automotive OEMs build dual-source discipline. Net: moat probably the same width, maybe a half-step wider.

Single biggest threat? China. Fully subsidized Chinese SiC and silicon power competitors who do not need to earn cost of capital can sustain pricing pressure for a decade. The US/EU subsidy response (CHIPS Act, EU Chips Act) helps, but Chinese domestic market access for ON is at risk in a continued decoupling scenario. ON discloses meaningful China revenue exposure.

Secondary threat: capex overrun on the Bucheon SiC mega-fab and East Fishkill 300mm transition. ON has roughly doubled fixed-asset intensity in 5 years; if the depreciation tail collides with a slow recovery, ROIC suffers structurally.

A Buffett-style 10-year forecast on ON has medium confidence. The business model is durable, but the cyclical and competitive dynamics are not predictable enough to underwrite a 10-year hold without active price-discipline. This is NOT a See's Candy. It is a high-quality cyclical with a current price advantage.

CONFIDENCE: medium