Better AI Infrastructure Stock: Nvidia vs. AMD

AI data center infrastructure spending is projected to rise from ~$750B this year to >$1T next year, supporting continued GPU/CPU demand. AMD is seeing momentum with data center revenue up to $6.7B (more than doubled) and management guiding server revenue up 70% this year and data center revenue “more than double” in 2027, but the stock trades at a very high trailing P/E of 121 vs a 33 tech-sector average. Nvidia is portrayed as the better-value pick after Q2 total sales more than doubled to $96.2B (vs $92.1B consensus), adjusted EPS rose 120% to $2.22/share, and Q3 revenue guidance of ~$108B implies +89% y/y, with an estimated 86% market share and a lower ~29 P/E than AMD.

Analysis

This reads less like a new fundamental update than a validation of the current AI capex regime, which means the market is likely to keep paying up for the most exposed platform winner rather than the lowest-quality “cheap” exposure. The key mechanism is not unit growth alone but mix: the company that captures the full rack, software, and ecosystem attach rate should hold pricing power longer, while the second player is still proving it can convert headline TAM into durable share.

Over the next 1-3 months, the main risk is digestion: if hyperscaler capex cadence normalizes after current buildouts, multiple compression can hit the higher-beta name first even if revenue continues rising. AMD’s CPU optionality is real, but it is a 2026-2030 story unless agentic workloads clearly force a larger CPU attach rate sooner; until then, the market is paying forward execution that has to be proven in guidance. A useful tell is whether NVDA maintains outsized guide beats while AMD’s data-center growth starts to decelerate on a tougher compare.

Contrarian angle: consensus may be underestimating how long the AI spend cycle can run, but overestimating AMD’s ability to close the strategic gap without a material software/distribution advantage. If the spend pool expands as expected, the bigger second-order winners may be adjacent infrastructure names tied to power, networking, and memory rather than the “fast follower” chip narrative. The thesis breaks if NVDA’s next guide implies a step-down in growth, or if AMD’s margins and operating leverage fail to improve despite higher server revenue.

Key Decisions for Investors

• Relative-value: long NVDA / short AMD for 1-3 months; the spread favors the name with stronger pricing power and lower multiple, with upside if AI capex remains sticky and downside if growth is simply repriced.
• If entering outright long exposure, prefer NVDA on pullbacks rather than chasing AMD strength; use AMD only as a momentum trade if next earnings confirms accelerating data-center margins and CPU attach.
• Set an alert on 2026 hyperscaler capex commentary: if large cloud buyers signal digestion, trim AI beta quickly—AMD should de-rate faster than NVDA on any guide slowdown.
• Sector proxy trade: long SMH over broad tech for 3-6 months; semis retain direct AI leverage, but avoid paying for names where valuation depends on 2027-2030 TAM.
• Watch-fail level: if NVDA guide or gross margin misses by enough to imply share/price pressure, the entire AI infrastructure complex likely needs a de-risking; that would be the point to rotate into quality balance-sheet beneficiaries only.