Amazon.com

The central thesis is AWS. Cloud and compute demand should continue compounding, and AWS is one of the largest and highest-quality infrastructure businesses in the world. It sells capacity on a consumption basis to a customer base that spans startups through governments, which makes its revenue both recurring in practice and expandable as workloads grow.

The reason this matters more now than it did three years ago is that AI workloads are additive to existing cloud migration rather than a substitute for it. Training consumes capacity in large contracted blocks; inference consumes it continuously once a model is deployed. Both land on the same infrastructure Amazon already operates at scale.

Beyond AWS, Amazon holds a dominant e-commerce position and logistics scale that few competitors can replicate. Robotics and longer-term automation are optionality. Neither is required for the thesis to work.

Consolidated margins are the honest place to look for whether this is working. Retail is structurally lower-margin than cloud, so blended operating margin understates AWS while the capital spending for AI capacity runs through the same statements. Comparing Amazon's margin profile to a pure-play software business compares two different objects.

I initiated the position on a pullback rather than earlier. That framing carries an obvious risk: buying a large-cap on weakness only works if the weakness was about price rather than about the business. The combination that would falsify it is AWS growth decelerating while capital spending stays elevated, which would mean the company is building into demand that is not arriving. Sizing has to account for what else the book already owns. AWS, Google Cloud and Meta's data-center build are all funded out of the same AI capital-spending cycle, so the correlation between these holdings is higher than their different end markets suggest.