Making Cents Sense

For years, I treated Amazon’s retail business as the world’s most elaborate customer-acquisition program for AWS. That was the accepted story: Amazon retail created scale, loyalty, Prime memberships, and an ocean of consumer data, while Amazon Web Services produced the margins Wall Street actually loved. Investors tolerated the thin economics of shipping millions of low-priced objects to millions of impatient people because AWS made the consolidated income statement look civilized. Then something inconvenient happened to that tidy narrative. Amazon’s retail operation started making serious money. Not “nice little improvement” money. Not “the holiday quarter went well” money. In the second quarter of 2026, Amazon’s North America segment generated $9.1 billion in operating income, up from $7.5 billion a year earlier. The International segment added another $1.7 billion, compared with $1.5 billion in the prior-year period. Together, those two commerce-heavy segments produced $10.8 billion...

I came for the packages, stayed for AWS, and nearly missed the machine quietly monetizing purchase intent across the entire company When I think about Amazon, advertising is not the first thing that enters my mind. Boxes enter my mind. Delivery vans enter my mind. A smiling arrow enters my mind, followed by the uneasy realization that I once ordered batteries at 11:42 p.m. and received them before I had emotionally committed to needing batteries. If I am thinking as an investor, AWS usually barges into the conversation next. That makes sense. Cloud computing turned Amazon from a famously low-margin retailer into something far more powerful, and AWS has earned its reputation as the company’s profit engine. It is difficult to overlook a division that reported $42.2 billion in second-quarter 2026 sales and $16.6 billion in operating income. But while everyone watches the cloud, Amazon has been building an advertising business large enough to qualify as an empire and quiet enough to be mis...

The market loves the engine. I’m looking at the electrical system keeping it alive. When investors talk about artificial intelligence, the conversation usually begins and ends with GPUs. We discuss accelerator road maps, training clusters, model sizes, memory bandwidth, and whichever chief executive has most recently held up a large piece of silicon under theatrical lighting. I understand the fascination. The GPU performs the glamorous work. It is the muscle behind training and inference, the component that gets photographed, named, back-ordered, and discussed as though it possesses a minor royal title. But a GPU cannot eat enthusiasm. It needs electricity—an enormous amount of it—delivered at the correct voltage, at precisely the right moment, with minimal waste, limited heat, and enough reliability that a multimillion-dollar AI rack does not become an extremely sophisticated space heater. Between the utility grid and the processor sits a long chain of power conversion, control, prote...