HScreen

lineage status · rendered 2026-09-09 02:16 UTC STALE inputs (497h)

ON · ON Semiconductor Corporation 23d old

conviction2
coverage tierT2-PARTIAL
horizon12 months

Return distribution probability-weighted scenarios, never collapsed to a point

-32%28%
+12%45%
+45%27%
EV +8.59%
why these probabilities: BULL (+45%, to ~$120): utilisation reaches the low 90s and the full ~700bp margin bridge lands while AI data-centre revenue more than doubles off a base already above $500m and 800V content per rack begins showing in ASPs. Earnings power moves toward a mid-cycle base multiples higher than FY25's $0.29, and the market pays a power-for-AI multiple rather than an auto-analog multiple. +45% only takes the stock to ~$120, still below the 52-week high of $134.92 — this is a re-rating to a level the market has already paid inside the last year, which is why I am comfortable above the bull base rate. BASE (+12%): utilisation grinds into the mid-to-high 80s, margin recovers most of the way but not all; AI data-centre grows strongly but off a small base, auto stays soft to flat as guided (ON/ON_2025Q1.json:39). Earnings inflect visibly, the multiple compresses on the higher earnings base rather than expanding, and the stock delivers roughly the earnings revision. Modest, because much of this is in the price at $82.66. BEAR (-32%, to ~$56): the falsifier fires — auto/industrial demand deteriorates rather than recovers, utilisation stalls or reverses back toward 70%, and the 700bp of headwind stays a headwind. Simultaneously the AI data-centre leg disappoints against the "more than double" guide, either because the 800V retrofit cadence is slower than the architecture change implies (manag
not built — options-implied distribution on the same axis, with the KL number and where the disagreement sits
needs: desk-options implied bands (v1.4 Part 4a) — activates at 60% coverage; currently 33%

Visibility how far ahead this name can be seen

not built — visibility score
needs: visibility_score.py

Trust how much weight this name's own record can carry (credibility Z)

not built — credibility
needs: credibility.py

Technology supply chain graph-backed context · read-only

not built — supply-chain attributes
needs: ticker not mapped in supply_chain.yaml

Sizing survival-first size (half-Kelly) vs the enforced ladder

kelly not computable — no data

Payout durability does the distribution survive a downturn?

PASS
DPS trajectoryinsufficient_history
interest cover now10.54x
interest cover at FORWARD rates4.61x

Falsifiers what would prove this wrong

Fab utilisation continues rising toward the low 90s, delivering the ~700bp gross-margin recovery management has quantified
AI data-centre revenue more than doubles off a base already above the prior $500m target
800V DC rack architecture translates into higher ON content per rack rather than being competed away
Auto and industrial demand stabilise rather than deteriorate further, so the AI leg adds to recovery instead of offsetting decline
The hyperscaler capex pool underwriting the AI power demand holds at the ~$1.18-1.475trn trajectory two independent brokers size it at
Capital returns continue at or near 100% of free cash flow without balance-sheet strain

Conflicts surfaced, never netted out (T4)


The revision tape is unavailable, not neutral. The estimates block returns n_up 0, n_down 0, breadth 0.0, median_move null and an empty outliers list on both EPS and revenue across a 365-day window, on 192 available esti		high

My valuation score of 4 (full) sits against management's own revealed valuation view: they repurchased stock last quarter at 160% of free cash flow and explicitly described the price as a dislocation (ON/ON_2026Q1.json:6		high

My p_bull of 27% is 19.3pp below the node_pooled bull anchor of ~46.3% and 13.4pp below the historical ~40.4%. I am deliberately far below both, on the argument that the historical distribution's qualifying 45% advances 		medium

Internal tension in my own scoring: growth 7 on a credible, independently-sized AI data-centre driver, against valuation 4 on the view that the recovery is largely priced. These pull in opposite directions on the expecte		medium

Management gives a bullish characterisation (AI data centre to more than double, above the prior $500m target, ON/ON_2026Q2.json:29) alongside an explicit deflation of the same story (800V racks are a content step functi		medium

The case against independent adversary, different model family (P6) weakened

Version history from the research record

archived versions4
archivedhashbytes
2026-08-17T14:24:556e138e1aa96137,459
2026-08-17T14:27:28b478657150b637,520
2026-08-17T15:40:0025f2c29c050f39,630
2026-08-19T08:32:0109ebb530bca639,587

What changed in the latest version

-ticker: "ON" # MUST stay quoted: unquoted ON is YAML 1.1 boolean true
+ticker: 'ON'
-
- value: >
- ON Semiconductor: power and sensing semiconductors across three groups (Power Solutions,
- Advanced Solutions, Intelligent Sensing). Historically an auto/industrial analog-power
- house levered to SiC traction inverters and image sensors; the 2025-26 development is a
- pivot in the demand mix toward AI data-centre power delivery — the company describes
- ramping content "from the wall all the way to the XPU", high-voltage through low-voltage,
- with 800V DC rack architectures as the step-function opportunity. FY2025: revenue
- $5.995bn, operating income $748m, net income $121m, EPS $0.29 — a trough year with
- ~700bp of gross-margin headwind from underutilisation (utilisation ran 68-74% through
- 2025, low 80s by 2026Q2). Distribution is ~58% of revenue, roughly half fulfilment and
- half demand creation. No dividend. Capital intensity guided to mid-single-digit percent
…diff truncated
Rendered from canonical artifacts only (T1) — no terminal database exists. Every figure is read from the system's own files at render time; where an artifact is missing the panel names the dependency rather than showing an empty box. No execution path exists anywhere in this interface (T5), not even a link.