The 'Next Nvidia' Trade? Why Investors Are Suddenly Watching Advanced Micro Devices, Arm Holdings, and Marvell Technology

Arm Holdings, Marvell Technology, and AMD are positioned to benefit from AI inference demand, with projected earnings growth of 22%/41%/30% for Arm, 34%/43%/37% for Marvell, and 76%/76%/39% for AMD across the next three forecast periods versus Nvidia’s 87%/41%/23%. AMD cited a server CPU market growing 35% annually through 2030 to more than $120 billion, while Arm sees revenue rising from $4.9 billion to $25 billion over five years. The article argues these smaller chip names have outperformed Nvidia this year as investors rotate toward AI infrastructure beneficiaries beyond GPUs.

Analysis

The market is rotating from AI training scarcity toward AI deployment efficiency. That transition changes the profit pool: less of the incremental spend should accrue to the incumbent GPU platform and more to the silicon and architecture layer that improves watts-per-inference, customizes workloads, and reduces hyperscaler dependency. The second-order winner set is therefore broader than the article implies: foundry/capex ecosystems that support custom silicon, networking vendors tied to data-center buildouts, and IP licensors with content embedded across multiple accelerators.

Among the three highlighted names, AMD has the cleanest near-term operating leverage because server CPU share gains can compound without requiring a full platform change at customers. Arm has the most durable royalty/architecture exposure, but the market is already paying for a long-duration option on its ecosystem capture, so upside is more sensitive to proof of monetization than to narrative alone. Marvell is the most cyclical of the trio; its multiple can expand fast on design-win headlines, but that also makes it the most vulnerable to order pushouts if hyperscalers slow custom chip spending.

The consensus may be underestimating how quickly AI capex can reallocate rather than simply expand. If training demand normalizes or GPU lead times shorten, investors could de-rate the "next Nvidia" trade and re-rate only the names with clear shipping revenue and earnings visibility over the next 2-4 quarters. Conversely, if inference adoption keeps compounding, the market could reward lower-power architectures for years because cost-per-token becomes the dominant KPI for cloud margins.

The key risk is that this is a crowded relative-value trade rather than a pure fundamental one. These stocks have already rerated on the same thesis, so any miss in guidance, slower custom silicon ramps, or evidence that GPUs remain the default inference standard could trigger fast multiple compression even if fundamentals stay solid. The setup is constructive, but the trade needs discipline on entry and hedging because the narrative can unwind before the earnings estimates do.

Key Decisions for Investors

• Long AMD into the next earnings cycle; use pullbacks to accumulate, targeting a 6-12 month horizon. Best risk/reward in the group because share gains can drive both revenue and margin leverage, but trim if server CPU growth commentary narrows below expectations.
• Long MRVL vs short NVDA as a relative-value pair for 3-6 months. Thesis: custom AI silicon and inference monetization should outperform a mature platform name on incremental revenue surprise, but keep the short small because NVDA remains the quality hedge if AI capex re-accelerates.
• Buy ARM on dips only, with a 12-18 month horizon. This is a royalty-compounding story, so patience matters; upside is attractive if ecosystem penetration broadens, but valuation leaves little room for a miss in monetization cadence.
• Use call spreads rather than outright longs on MRVL and AMD over the next 1-2 quarters. The implied upside is already partially in the shares, so defined-risk structures preserve participation while limiting drawdown if the market rotates back to GPUs.
• Avoid chasing NVDA on this headline set; instead treat it as the hedge leg for any long basket in AMD/MRVL/ARM. If hyperscaler demand re-accelerates, NVDA can offset factor-risk in the basket; if not, the relative trade still works.