Why I've Begun Accumulating ON Semiconductor

ON Semiconductor’s Q2 2026 AI data center revenue more than doubled Y/Y, and management expects the segment to more than double in full-year 2026. Free cash flow rose to $425.4M (+300.94% Y/Y) with non-GAAP gross margin at 39.3%, and Q3 guidance calls for 40.0%–42.0% while utilization improved to 83% (from 77%). The company is accelerating shareholder returns via buybacks—$332M repurchased in Q2 under a new $6B authorization—and repurchased ~105% of free cash flow YTD, though the stock’s ~262x trailing P/E highlights valuation risk if AI demand or execution slips.

Analysis

ON looks less like a classic cyclical analog name and more like a constrained-capacity beneficiary of AI infrastructure spend. The key market mechanism is not just higher unit volume; it is mix shift into power delivery content where attach rates can scale faster than rack deployments, so earnings can inflect even if top-line growth looks ordinary. That makes ON a cleaner second-derivative way to express AI capex than NVDA, which remains the primary demand engine but is already much more fully owned.

Relative winners are hyperscaler capex beneficiaries and power-chain peers, but the spread is uneven. AMZN’s data-center build supports the thesis on the demand side, while NVDA benefits indirectly from better power density and thermal constraints being solved; however, ON captures more of the incremental content per rack. TXN is the main relative loser on multiple perception: if investors conclude power semis are entering a durable AI runway, broad analog may be viewed as a lower-growth bucket rather than a quality compounder. WOLF is the high-beta alternate expression, but its balance-sheet and execution risk make it a financing-sensitive substitute rather than a superior one.

The important risk is timing: power-semiconductor orders usually lag hyperscaler spend by quarters, so any pause in AI capex would likely show up in ON before it shows up in cloud revenue. The contrarian miss is that the market may be extrapolating design wins into straight-line share capture; once supply tightness eases, pricing power can normalize faster than volume growth. The thesis is falsified if AI-related order strength rolls over for two consecutive quarters, utilization stops improving, or management walks back margin expansion into the next earnings cycle.

Key Decisions for Investors

• Long ON vs. TXN for 1-3 months: express the view that AI power content creates faster earnings revision momentum than broad analog; target relative multiple expansion, stop if TXN re-accelerates guidance while ON margin commentary stalls.
• Add ON on post-earnings pullbacks, not into strength: use any 5-8% drawdown after a print or guidance reset as entry, because the better risk/reward is buying when the market questions sequencing rather than paying up for peak narrative.
• Pair long ON / short WOLF for 3-6 months: ON has the cleaner balance sheet and customer validation, while WOLF remains the more fragile way to play SiC demand; cover the short if WOLF secures non-dilutive financing or ON loses margin discipline.
• Watch AMZN capex and NVDA rack-level commentary as confirmation signals over the next 1-2 quarters; if hyperscaler power density spending slows, trim ON aggressively because the thesis is more capex-sensitive than consensus assumes.