infra dossier

Amazon

Amazon is underappreciated if AI demand rewards boring infrastructure, logistics discipline, and enterprise distribution.

Confidence

64%

AWS, Trainium, retail logistics, and robotics give Amazon several ways to capture value if intelligence becomes a workload and a physical-world optimizer.

Hypotheses

Trainium does not need to beat NVIDIA broadly; it needs to lower Amazon internal and AWS strategic dependence.

AI in fulfillment may matter more financially than consumer-facing Alexa narratives.

Open questions

Can AWS regain narrative leadership against Azure and Google Cloud in AI workloads?

Will Trainium become credible outside Amazon-first customers?

What would change our mind

Persistent Trainium under-adoption despite price advantage.

AWS losing high-value AI platform customers in a way that damages ecosystem gravity.

Recent evidence

Hyperscalers continue moving from rented frontier capacity toward owned silicon and owned data centers.

Robotics and warehouse automation remain one of Amazon's most defensible physical AI lanes.

Linked themes

hyperscaler capexcustom siliconphysical AI