Did Amazon Just Say Checkmate to Nvidia?

Amazon says Trainium2 offers 30% better price performance than GPUs and is nearly sold out, with Trainium3 and Trainium4 also largely accounted for. The article argues this creates healthy competition rather than a threat to Nvidia, noting Nvidia revenue growth is still accelerating with quarterly YoY growth of 73% in Q4 and analyst estimates of 79% in Q1 and 85% in Q2. Overall, the piece is constructive on both Amazon's custom AI chips and Nvidia's long-term positioning.

Analysis

This is less a winner-take-all chip war than a capacity-arbitrage setup: hyperscalers will route workloads to whichever silicon is available first, not just whichever benchmarks best. That means Amazon’s custom silicon strength is an AWS monetization lever, but it also functions as a demand sponge for AI inference/training that would otherwise pressure Nvidia pricing; the near-term effect is not share loss so much as faster total market expansion. The second-order winner here is the broader AI supply chain—advanced packaging, HBM memory, networking, and power infrastructure—because mixed-chip deployment raises total system complexity and keeps capex elevated across vendors.

The key risk for AMZN is that custom chips improve utilization but can also intensify customer lock-in optics, inviting larger enterprise buyers to multi-cloud more aggressively over the next 12-24 months. For NVDA, the real bear case is not a sudden demand cliff but margin/attach-rate compression if hyperscalers increasingly reserve Nvidia for portability-sensitive workloads and shift lower-margin volume to in-house silicon. The market is likely underestimating how long this “both/and” phase can persist; in a shortage environment, competition expands the pie before it meaningfully redistributes it.

Contrarian read: the headline should not be interpreted as a direct negative for Nvidia. When custom chips are sold out, it signals constrained supply, which usually preserves Nvidia’s pricing power because customers still need a fallback path. The more interesting overhang is Intel: as a legacy CPU supplier, it loses not only share but relevance in the AI stack narrative, and that can spill into supplier negotiation leverage across cloud and server procurement.

Near term, this is a multi-quarter story, not a one-day catalyst. The cleanest expression is to stay long both AMZN and NVDA but expect AMZN to be the better relative performer if AWS keeps converting silicon efficiency into cloud margin while NVDA remains the higher-beta absolute momentum name. The trade only breaks if AI capex pauses, or if a large cloud customer publicly standardizes on one architecture and triggers a broader procurement reset.

Key Decisions for Investors

• Long AMZN vs NVDA pair trade over the next 3-6 months: favor AWS monetization and custom-chip margin expansion while keeping exposure to the AI capex cycle; target modest outperformance for AMZN if utilization remains tight.
• Maintain/add to NVDA on 6-12 month horizon rather than fade the stock: capacity scarcity across the ecosystem still supports revenue growth; the risk/reward favors holding the leader until evidence of AI spend deceleration appears.
• Avoid or underweight INTC as a relative loser in the AI infrastructure stack for the next 12-18 months; it is the clearest share-and-relevance casualty if custom cloud silicon continues to scale.
• Look for a basket long in AI infrastructure enablers (HBM, packaging, networking, power) as a derivative winner from multi-chip deployment; this is the cleanest second-order way to express rising total compute demand.
• If NVDA rallies sharply on headline-driven optimism, consider selling covered calls 1-2 quarters out to monetize elevated implied volatility while retaining core exposure to the secular demand trend.