Making Cents Sense

For years, Microsoft trained investors to think of cloud computing as one of capitalism’s more elegant machines. Build the platform, fill the data centers, sell recurring access, and watch software economics spread across an enormous customer base. It was not effortless, but it looked wonderfully scalable. Each additional workload could ride on infrastructure Microsoft had already built, while subscription revenue arrived with the dependable rhythm of a direct debit nobody in accounting wanted to cancel. Then artificial intelligence showed up carrying a shopping list. The list included graphics processors, central processing units, custom silicon, networking equipment, cooling systems, power contracts, land, concrete, backup generators, fiber, and data centers large enough to make an airport terminal feel intimate. Suddenly, the cloud business that investors loved for its software-like margins began consuming capital with the appetite of a heavy industrial project. I do not view this a...

I used to think AI was software. Clean. Abstract. Floating somewhere in the cloud like a polite hallucination. Models, prompts, APIs—everything felt weightless, like intelligence had finally escaped gravity and taken up residence in a server rack labeled “innovation.” Then I started pulling the thread. And like most things in tech, the deeper I went, the less magical it looked—and the more brutally physical it became. Because AI doesn’t start in the cloud. It starts in heat. In sand. In factories that look less like Silicon Valley and more like something between a chemistry lab and a nuclear facility. It starts in places where mistakes aren’t bugs—they’re scrap. And once you see that, you can’t unsee it. So let me walk you through what I’ve come to realize—the actual map of value creation in AI hardware. Not the polished version. Not the investor deck. The real one. The one that explains who actually makes money, who pretends to, and who quietly prints cash while everyone else a...