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

Nvidia's recent-quarter revenue more than doubled to over $96 billion, while gross margin remained above 70%, reinforcing its AI-chip leadership despite rising competition. 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 comments reduce concerns that Amazon's in-house silicon will materially displace Nvidia, although competition from Amazon, AMD and Intel remains a long-term consideration.

Analysis

The relevant signal is not that AWS will retain Nvidia capacity, but that hyperscalers are converging on a dual-sourcing architecture: proprietary silicon for bounded, high-volume inference workloads and Nvidia for frontier training, rapidly changing models, and customers requiring CUDA portability. That protects NVDA's premium software-and-networking attach rate over the next 1-3 quarters, while limiting the addressable share available to custom ASICs rather than eliminating them. AWS's custom-chip revenue run rate is therefore better read as incremental AI infrastructure demand than a clean displacement metric, unless AWS begins shifting its highest-value training instances away from Nvidia.

The competitive pressure is likely to emerge through mix and pricing rather than an abrupt unit-share loss. AMD can win capacity-constrained or price-sensitive deployments, but its upside depends on software maturity and cloud availability; AMZN benefits from vertically integrated cloud gross margin when customers accept Trainium/Inferentia. NVDA's key exposure is that hyperscaler capex can continue rising while its revenue growth decelerates if internal silicon absorbs inference growth—a multiple risk over 6-18 months even if absolute revenue remains strong.

Consensus may be too binary on custom silicon. A larger total-compute market can support both NVDA and ASICs, but the valuation distinction matters: NVDA needs sustained premium-system demand and gross-margin resilience, whereas AMZN monetizes chips primarily through AWS utilization and retention. The near-term article-driven signal is weak; Jassy's commentary is qualitative and should not alter positioning without corroboration from instance availability, AWS capex, and disclosed Trainium adoption.

Key Decisions for Investors

• Maintain, do not chase, an NVDA core long into the next earnings cycle; add only on a 10-15% drawdown or if management reaffirms data-center growth with stable gross margin. Thesis fails if hyperscaler capex remains elevated but NVDA data-center growth materially decelerates for two consecutive quarters or gross margin guides below expectations.
• Express the architecture split as a 3-6 month pair: long NVDA / short AMD in equal dollar exposure if AMD's cloud-instantiated supply and software adoption do not show measurable acceleration. Risk/reward is roughly 2:1 if NVDA sustains premium mix; stop if AMD announces broad hyperscaler training deployments with credible utilization evidence.
• Use AMZN as the cleaner custom-silicon beneficiary only as part of an AWS-margin thesis, not a direct NVDA hedge. Watch AWS operating-margin progression and growth in proprietary-chip workloads over the next two earnings reports; absent evidence of margin expansion or accelerated cloud growth, there is no standalone chip-driven trade.
• Set an alert for hyperscaler capex guidance revisions: upward capex with flat NVDA procurement commentary is the early warning that ASIC substitution is affecting mix. That setup would favor reducing NVDA exposure and revisiting long AMZN or a broader cloud-infrastructure basket rather than shorting AI compute demand.