This Week's Wolf Pick - ON Semiconductor; ON Gets 11x More Silicon Per Nvidia Rack

ON Semiconductor is being positioned as a key beneficiary of NVIDIA’s 800V AI data center transition, with the article arguing each 800V rack could contain roughly 11x more ON silicon than today’s 54V racks. Q4 2025 revenue was $1.53B, EPS was $0.64 versus consensus by $0.02, and free cash flow hit a record $1.4B, while AI data center revenue exceeded $250M and doubled year over year. The stock still trades at about 20x forward earnings versus 42-52x for peers, with a $6B buyback authorization adding support.

Analysis

The market is still treating ON like a cyclical SiC recovery story, but the more important setup is an architectural repositioning from EV optionality to AI infrastructure indispensability. If 800V racks become the default for high-density AI clusters, ON’s content per deployed rack scales nonlinearly because the value is in the conversion chain and protection layers, not just the headline wattage. That creates a second-order winner effect: every accelerated AI capex plan by hyperscalers and neo-clouds should lift ON’s mix and gross margin faster than the street expects, while commodity-minded power semi peers risk being left behind.

The key contrarian point is that consensus is likely underestimating the timing gap between design wins and revenue recognition, so the next 1-2 quarters may still look deceptively ordinary even if the order book is inflecting. That makes the stock vulnerable to headline skepticism until management proves that AI revenue can re-rate the whole model, but the lead-time extension is exactly the kind of channel signal that usually resolves 2-4 quarters before the P&L does. If utilization climbs into the low-80s, margin expansion could come in a step function rather than gradually, meaning the earnings power gap versus the current multiple can close quickly.

The competitive knock-on is more interesting than the direct trade: NVIDIA’s 800V transition likely pulls forward spending from OEMs, integrators, and power subsystem vendors, but it can also compress legacy 54V ecosystems and create stranded design content for slower-moving suppliers. CRWV and NBIS benefit indirectly through capacity differentiation if they can market 800V readiness, while MPWR and POWI are at risk of being forced to defend their premium multiples if ON proves it can capture AI growth from a lower base. The main failure mode is schedule slippage: if Rubin Ultra deployment or customer qualification drifts by 2-3 quarters, the stock can de-rate before the earnings bridge arrives.

Key Decisions for Investors

• Long ON into Q1 2026 earnings with a 3-6 month horizon; reward/risk is favorable if management confirms AI data center revenue acceleration and utilization recovery, but trim if commentary suggests 800V revenue is still only design-stage.
• Buy ON vs MPWR as a pair trade for 6-12 months; ON has the better operating leverage to the 800V cycle and a much lower multiple, while MPWR likely remains priced for perfection.
• Consider a small starter long in NVDA for a 12-24 month horizon as a toll collector on the architecture shift, but hedge with short-term index or semiconductor weakness because the incremental benefit is second-order and may not show up in near-term estimates.
• Avoid chasing CRWV/NBIS purely on this theme; use them only as expression vehicles if they announce 800V-linked capacity or power-dense buildouts, since the current upside is more narrative than quantifiable.
• If ON rallies hard into earnings, use calls rather than stock for new exposure; the catalyst path is asymmetric, but headline volatility around timing risk remains high.