These 15 Words From Amazon’s Andy Jassy May Eliminate Nvidia’s Biggest Risk

Nvidia's recent-quarter revenue doubled 100% to more than $96 billion, while gross margin remained above 70%, reinforcing its leadership in AI GPUs despite competitive concerns. Amazon's custom-chip annual revenue run rate has exceeded $25 billion, but AWS CEO Andy Jassy said customers will use Nvidia chips "for as long as we can foresee" and that AWS will remain the best place to run them. The article argues that rising global AI-compute demand can support both Nvidia and competing chip providers rather than materially displacing Nvidia's market position.

Analysis

Amazon’s continued Nvidia deployment matters less as a read-through on NVDA share than as confirmation that hyperscalers are adopting a tiered-compute architecture: proprietary ASICs for predictable, cost-sensitive inference and Nvidia systems for frontier training, rapidly changing models, and workloads dependent on CUDA/software portability. That segmentation can expand total accelerator spend while lowering the risk that custom silicon produces a sudden demand cliff for NVDA. The second-order loser is Intel: it lacks both Nvidia’s software lock-in and the captive cloud distribution channel supporting AWS Trainium/Inferentia or Google TPU.

The market should distinguish workload growth from revenue substitution. AWS’s internal chips may reduce Nvidia’s share of low-end inference over 6-18 months, but they also lower customer AI cost, increasing application deployment and ultimately demand for premium training and high-performance inference capacity. The nearer-term valuation risk for NVDA is not Amazon displacement; it is any evidence that hyperscaler capex growth decelerates before utilization and enterprise monetization absorb installed capacity, which would compress the premium multiple even if revenue remains strong.

Consensus may be too complacent on AMD’s opportunity: AWS validation of a multi-silicon vendor strategy improves the commercial logic for customers to qualify MI-series systems as a negotiating and supply-diversification tool. AMD need not take dominant share to drive material earnings revisions from a low accelerator revenue base. Conversely, claims of durable Nvidia demand should be treated as capacity-planning commentary rather than a binding procurement commitment; monitor AWS accelerator instance pricing, lead times, and capex guidance for independent confirmation.

Key Decisions for Investors

• Maintain NVDA as a core long over a 6-12 month horizon, but do not add solely on this commentary; add on a 10-15% drawdown or after independently confirmed hyperscaler capex guidance. Thesis fails if aggregate AMZN/MSFT/GOOGL/META capex guidance falls for two consecutive quarters or NVDA datacenter gross margin guides below 70%.
• Initiate a 3-6 month long AMD / short INTC pair, sized dollar-neutral. AMD benefits from accelerator vendor qualification and supply-diversification spending, while INTC remains exposed to weaker competitive positioning in both AI accelerators and foundry execution; reassess if AMD AI revenue guidance fails to rise at the next earnings report.
• Use a barbell rather than an outright AMZN short: long AMZN alongside NVDA or AMD. AWS custom silicon can improve cloud gross margin and support lower AI-service pricing, while its continued third-party GPU availability protects AWS from losing high-end workloads to competing clouds.
• Set an alert around cloud AI pricing and utilization over the next 1-3 months: falling GPU-instance prices without offsetting utilization would signal emerging capacity oversupply and warrant trimming semiconductor beta, especially NVDA and AMD.