This Company Could Become the Nvidia of AI Inference

ON Semiconductor is positioned to benefit from AI inference spending, which the article argues will eventually surpass data center infrastructure spending. The company’s data center revenue rose 30% in Q1 and management now expects that segment to double year over year in 2026, implying roughly $500 million of revenue versus about $6.47 billion in total 2026 revenue consensus. The piece is broadly constructive on ON’s long-term growth, especially from power technology for hyperscaler and edge inference applications.

Analysis

The market is still underappreciating that inference is not just a larger TAM for the same AI buildout; it changes the mix of spend from lumpy capex to durable operating intensity. That matters because the economics of inference favor vendors that sit in the power-and-thermal stack rather than the headline accelerator layer, which creates a longer runway for ON’s content per deployment to expand even if GPU unit growth normalizes. The second-order effect is that each incremental inference node should require more supporting power management than a pure training footprint, making ON’s exposure more resilient than a typical “AI hardware” narrative.

The bigger implication is competitive: as AI workloads move closer to the edge, ON’s addressable market broadens from hyperscale to industrial, automotive, healthcare, and embedded systems, but those markets should not be valued as separate stories anymore. They become a portfolio of inference endpoints sharing the same power-silicon architecture, which could raise investor willingness to pay if management proves repeatable design wins. The risk is that consensus may be extrapolating early revenue inflection too aggressively; data center growth can decelerate fast if customers pause orders after initial buildout or if hyperscalers optimize power architecture around fewer suppliers.

Near term, the stock likely trades on revision momentum, not the full AI inference thesis. The key catalyst is whether management’s 2026 growth commentary turns into upward estimate revisions over the next 2-3 quarters; if not, the market may re-rate ON back toward a cyclical semiconductor multiple. The contrarian angle is that the best way to play the theme may be through suppliers with cleaner leverage to power density and thermal constraints, not just the most obvious AI names.

Key Decisions for Investors

• Add ON on 2-3 month pullbacks, with a 12-month horizon; thesis is multiple expansion from estimate revisions as data center revenue scales faster than headline sales, but size the position assuming episodic semiconductor volatility.
• Pair trade: long ON / short a basket of higher-multiple AI hardware names over 6-12 months; ON has more underappreciated operating leverage to inference power demand, while consensus is already crowding the accelerator trade.
• Buy ON call spreads for 6-9 months rather than outright equity if entering after a strong move; the cleanest payoff is from upward guidance revisions, but upside is likely capped by cyclicality and customer concentration risk.
• Use any sign of order pushout or muted quarterly data-center commentary as a de-risk trigger; if growth fails to accelerate over the next two earnings prints, the inference thesis becomes a story stock rather than a fundamentals trade.
• For lower-beta exposure, consider a basket long ON / NVDA as a way to isolate power-silicon content versus compute content; if inference spending shifts as expected, ON should outgrow the average AI semis complex on a revenue mix basis.