Amazon Just Delivered Mind-Boggling News to Shareholders

The important second-order read-through is not simply that AMZN is gaining AI share, but that AWS is turning from a generic cloud utility into a vertically integrated AI infrastructure stack. That matters because the economic moat shifts from pure scale to bundled demand capture: GPU capacity for training, lower-cost in-house silicon for inference, and CPU compute for agentic workloads. If that bundle is sticky, AWS can defend share even if a customer splits workloads across vendors, because switching costs rise at the orchestration layer rather than just the hardware layer.

The clearest loser is not NVDA in absolute terms, but marginal pricing power at the edge of the AI supply chain. Every incremental workload that migrates from third-party GPU rental to AWS-owned silicon compresses the value pool available to AI infrastructure aggregators like CRWV, while also extending the useful life of legacy CPU-heavy architectures in ways that reduce the urgency of a full GPU-only buildout. MSFT is the real strategic comparator: if Azure cannot match a similar breadth of chip optionality, it risks higher unit costs in inference and a slower margin expansion path in agentic AI deployments.

The market may be underestimating the timing mismatch between training and inference demand. Training is lumpy and capex-intensive; agents create persistent, lower-intensity but much larger unit volumes, which favors providers with both price flexibility and owned silicon economics. That is a longer-duration bullish setup for AMZN over the next 6-18 months, but it also creates a near-term risk: if AI monetization pauses or enterprise demand shifts slower than expected, the capex efficiency story can be postponed and sentiment could fade quickly.

Contrarian take: the consensus is likely too focused on Nvidia-vs-AWS and not enough on AWS-vs-everyone-else’s cost curve. If AWS can keep falling through to better inference economics, it can widen the gap between cloud market share and profit share, which should support AMZN even if topline growth merely stays elevated. The risk is execution: any evidence that Trainium adoption is slower than management suggests would weaken the thesis, because the market is implicitly paying for a credible path to structurally lower capex intensity.