Amazon shares have been on fire lately. Now we know why

Amazon shares jumped nearly 14% last week and nearly 20% since March 27, with Evercore ISI arguing the move is being driven by Anthropic-related AI demand rather than the U.S.-Iran ceasefire. Analyst Mark Mahaney said possible Trainium usage for Anthropic’s Mythos model is a proof point for Amazon’s chip product, and he reiterated outperform with a $285 target, about 20% above Friday’s close. The article is constructive for Amazon, but the catalyst is still inferential and mostly driven by analyst interpretation.

Analysis

The market is starting to re-rate Amazon from a consumer/infrastructure hybrid into a credible AI hardware monetization story. The important second-order effect is not just incremental chip revenue; it is the signaling value that Trainium is good enough for frontier-model training, which widens the addressable market for AWS beyond cost-sensitive workloads and improves the odds that AI capex stays in-house instead of leaking to NVIDIA ecosystems. If that perception sticks, Amazon can command a higher multiple on AWS margin durability even before chip economics become material in reported numbers.

The cleaner trade is that Anthropic’s use case creates a competitive wedge versus other cloud platforms: once a leading model developer is seen as comfortable on Trainium, AWS gains a procurement advantage in the next wave of AI budget allocations. That matters because the AI spend cycle is shifting from headline model training toward repeat training runs, inference, and specialized workloads, where price/performance and supply assurance matter more than raw peak performance. In that regime, Amazon benefits from being the cheaper, vertically integrated option while smaller cloud rivals look increasingly like resellers of third-party GPU capacity.

Consensus may be underestimating how much of this move is about positioning rather than near-term earnings. The stock can keep rerating for months if investors decide Amazon is one of the few mega-caps with both AI optionality and defensible non-AI growth, but the setup is vulnerable if Anthropic later clarifies that Mythos did not rely materially on Trainium or if hyperscaler AI capex guidance compresses margins faster than the market expects. The most important risk is that the current enthusiasm is front-running proof of scale that may not show up in 2025 financials, creating a gap between narrative and realized earnings power.

Separately, Alphabet is the best comp for the market’s behavioral shift: investors are rewarding companies that convert regulatory or product uncertainty into an AI platform premium. Amazon could experience a similar multi-month multiple expansion if AWS reclaims the narrative as an AI winner rather than a low-margin cloud utility. The near-term catalyst path is simple: more customer examples, clearer Trainium adoption, and any commentary that frames AWS as a structurally lower-cost path to frontier AI training.

Key Decisions for Investors

• Maintain/add to AMZN on pullbacks over the next 2-6 weeks; the stock can grind higher on narrative alone if Trainium adoption is repeatedly reinforced. Risk/reward favors staying long while the market is repricing AWS as an AI platform rather than a commodity cloud provider.
• Buy AMZN Jan-2027 call spreads to express a 6-12 month multiple expansion view with defined risk. This is preferable to stock if you want exposure to the rerating without taking full downside from any post-event fade.
• Pair long AMZN / short a basket of GPU-dependent cloud beneficiaries over 1-3 months if you want to isolate the Trainium adoption theme. The thesis is that vertically integrated AI infrastructure should compress perceived moats for pure infrastructure providers.
• Trim or hedge if AMZN fails to hold gains after the next AWS/Anthropic clarification window. A failed follow-through would suggest the move is sentiment-driven rather than fundamentals-driven, which is a good setup for a 5-8% retracement.
• Use GOOGL as a relative-value hedge only if you expect the market to rotate toward AI winners broadly; otherwise avoid overcomplicating the pair. The cleaner expression is AMZN long versus an AI-infrastructure laggard, not a generic megacap basket.