← Feed Deep Dive Matrix Subscribe

Prediction: ON Semiconductor Could Be a Sleeper AI Infrastructure Winner - AOL.com

www.aol.com 2026-09-28 AOL.com
Entities
Tags
ON SemiconductorAI Data CenterSilicon CarbidePower Semiconductor800V DC PowerValuation AnalysisNVIDIAAWSAutomotive ElectronicsAnalog ChipsFree Cash FlowCapacity UtilizationDesign WinRestructuring
News Summary
ON Semiconductor's positioning as an AI infrastructure play represents a compelling valuation arbitrage opportunity within the power semiconductor space. The core thesis hinges on a structural shift: ... Read original →
Industry Analysis
The market is mispricing ON Semi as a cyclical automotive tail, blind to the architectural reset 800V DC delivers to power semiconductor BOMs. Technical cascade: Shifting data center busbars from 415V AC to 800V DC doesn't merely add SiC content—it multiplies it. Per-rack SiC value jumps from roughly $15K to $115K. This mirrors the 2017 800V EV platform's displacement of IGBTs. SiC MOSFETs are becoming the power vasculature of AI compute. Once NVIDIA's MGX reference design locks in, hyperscaler power-IC selection windows compress sharply. ON's co-development with AWS is less a design win than a standards-authority play. Competitive dynamics: TI (28x) and ADI (23x) command premiums for full-stack analog capability, but ON's vertical SiC integration rivals Infineon's playbook. The real structural threat isn't TI—it's AWS building in-house power silicon, echoing Graviton's disruption of the ARM ecosystem. A 13-to-0 analyst buy/sell split means no sell-side hedge exists; any AI capex deceleration hits valuation with zero buffer. The $329M restructuring and Synaptics integration are execution-level reefs. 12-24 month call: 2026 Q2 is the 800V production-validation gate. If AI revenue crosses 25% of mix, re-rating from 17x toward 25x is near-deterministic. But this is execution-priced, not narrative-priced. There is no margin of safety.
Read Original Article →
Related
This page displays AI-generated summaries and metadata for research purposes. Original content belongs to the respective publishers.