ON Semiconductor, the Phoenix-based supplier of power management and image sensing chips, could be an overlooked beneficiary of the AI infrastructure buildout, according to a prediction published by 24/7 Wall St.

The financial news outlet framed the company as a "sleeper AI infrastructure winner" — a supplier whose connection to AI capex is less visible than that of GPU makers, but whose revenue could still track the same spending cycle. The prediction rests on a straightforward thesis: AI data centers consume enormous amounts of power, and every watt that flows into a rack must be converted, regulated, and protected by power semiconductors — a category where ON Semiconductor has spent two decades building a portfolio.

The company's silicon carbide (SiC) business, marketed under the EliteSiC brand, has been its most prominent growth story in recent years. ON built the position largely through its 2024 acquisition of United Silicon Carbide, deepened by a long-term wafer supply agreement with Quebec-based Wafer supplier Norstel — now its own subsidiary after ON bought the Swedish SiC crystal and epitaxial wafer producer outright. Vertical integration from boule growth to finished modules has let ON secure supply for automotive customers racing toward electrification.

The automotive anchor matters for any AI-infrastructure argument. Electric vehicles demand SiC MOSFETs for traction inverters because the wide-bandgap material switches more efficiently than silicon IGBTs, extending range. That same efficiency argument applies inside AI data centers, where power delivery losses translate directly into cooling costs and stranded capacity. Hyperscalers and server OEMs are already qualifying SiC and gallium nitride devices for power supply units and 800-volt DC distribution architectures as rack densities climb past what conventional silicon can handle efficiently.

ON's other core franchise — sensing — is further from the AI story. Its CMOS image sensors ship in volume into automotive advanced driver assistance systems and industrial machine vision, segments driven by different demand cycles than GPU clusters. The prediction from 24/7 Wall St. therefore hinges on the power side of the portfolio rather than the sensing side.

The "sleeper" framing carries a caveat investors should weigh. 24/7 Wall St. is known for aggregation and stock-screening commentary rather than deep semiconductor channel research, and the article is labeled a prediction, not reporting on confirmed design wins. No specific revenue figures, capacity numbers, or named hyperscale customers appear in the piece. What is well established, separately, is that AI data center construction is forcing an architectural rethink of power delivery — from grid to GPU — and that this shift is pulling wide-bandgap semiconductors into server racks faster than many supply-chain models assumed.

ON Semiconductor is not starting from zero in this market. Its power discretes, MOSFETs, and integrated power stages already ship into server and communications infrastructure customers, and it has publicly discussed data center power as a target application for its SiC roadmap. Competitors including Infineon, STMicroelectronics, and Navitas Semiconductor are chasing the same sockets, and Infineon has announced dedicated AI power capacity expansions of its own.

For ON, the competitive question is whether it can convert an automotive-weighted SiC cost structure into data-center volumes at prices hyperscalers will accept while its core EV customers digest inventory. 24/7 Wall St.'s prediction implies the market is underpricing that option. If AI rack power densities keep climbing, ON's power portfolio gives the claim a plausible mechanism — but confirmation will come only from disclosed data center revenue or design-win announcements in future quarters.