This Company Could Become the Nvidia of AI Inference

ON Semiconductor is positioned to benefit from the shift in AI spending toward inference, which the article says could overtake data center infrastructure spending in a few years. Management now expects data center revenue to double year over year in 2026, implying roughly $500 million in revenue versus about $6.47 billion in total company revenue expected for 2026. The article is broadly bullish on ON's long-term growth potential, though it is primarily an analytical commentary rather than a fresh corporate event.

Analysis

The market is still valuing AI spend as if the main prize is the initial buildout, but the more durable economics sit in the operating layer. That is a meaningful shift for power semis: inference creates recurring load, higher thermal intensity, and a longer replacement/upgrade cycle, which tends to support both content growth and pricing discipline. In other words, the revenue pool becomes less lumpy and more infrastructure-like, even though the spend is technically “software-driven.”

ON looks like an early beneficiary because it has exposure to multiple inference vectors at once: hyperscaler datacenters, edge workloads, and electrified/industrial systems that need power management regardless of where the compute lives. The second-order effect is that the AI supply chain broadens beyond GPUs and optics into power delivery, efficiency, and cooling-adjacent components, which should compress the relative valuation gap between “AI pure-plays” and enablement names if the narrative persists for 4-8 quarters. A key tell will be whether data-center revenue starts to represent a larger mix faster than consensus models assume; if so, the market will likely rerate ON on sustainable growth rather than cyclical recovery.

The main risk is timing mismatch: inference spending can be real but still slow to show up in reported revenue because hyperscalers phase deployments and qualify suppliers conservatively. Over the next 1-2 quarters, the stock is vulnerable to any capex digestion phase, margin pressure from mix, or signs that AI efficiency gains reduce incremental hardware intensity. The contrarian concern is that the market may already be extrapolating a multi-year runway from a small base; if inference architectures become materially more power-efficient, the total TAM could grow, but the attach rate for semis may not scale as fast as investors expect.

Key Decisions for Investors

• Long ON into any post-earnings pullback over the next 2-6 weeks; use weakness tied to broader semicap consolidation as entry, targeting a 12-18 month rerating if data-center revenue continues compounding above 30% with limited margin dilution.
• Pair trade: long ON / short a basket of slower-moving industrial semiconductor names over 3-6 months; the thesis is that AI inference provides a new growth leg while non-AI industrial exposure remains in a slower macro recovery.
• Buy ON call spreads 9-12 months out to capture the multi-quarter inference re-rate without paying full upside premium; attractive if implied volatility stays elevated but below expected revenue acceleration.
• Keep a hedge via short-term NVDA puts or a reduced NVDA long if the trade is to own inference enablers over direct compute, since the market may rotate from GPU beta into power and thermal bottlenecks.
• Watch for a catalyst at next guidance update: if management raises data-center growth commentary again, add to ON on the first confirmation, because the inflection is likely to be recognized in stages rather than all at once.