Not a retailer. It's an attention tollbooth disguised as a cardboard box. Andy Jassy inherited this architecture from Bezos in 2021 and has spent three years doing something his predecessor never prioritized: making it efficient. The result? If you're trying to understand Amazon in 2025, forget the delivery vans. Follow the margins. Forget the revenue number for a second. It's converting the act of selling things into four separate, higher-margin revenue streams that most people don't even notice. Start with the trick that makes the whole thing work: negative working capital. Customers pay Amazon immediately. That gap — multiplied across hundreds of billions in transactions — creates a permanent float of free cash that funds expansion without borrowing. The problem is, it's the same trick insurance companies use, except Amazon does it with toothpaste and phone chargers. The marketplace is where the model gets clever. It's a tax on a tax. AWS is the profit engine that makes everything else possible. Thirty-seven percent margins. Most companies just don't bother. Advertising is the segment that changed the financial narrative. They're buying. The ad appears at the moment of purchase intent, inside a commerce environment where conversion is directly measurable. Brands can't ignore it. They comparison-shop less. They try more Amazon services. The rest — Whole Foods, Amazon Fresh, Kindle, Echo, Fire TV, One Medical, Amazon Pharmacy — these are either traffic generators, data collectors, or long-horizon bets on massive markets. Devices are sold at or near cost to drive service engagement. None of these segments need to be independently profitable because the financial architecture doesn't require it. Retail generates cash through working capital dynamics. AWS and advertising generate profit. Everything else is funded by the spread between the two. When a mid-size retailer decides where to sell online, the decision comes down to one factor: where are the buyers already standing? Amazon has 200 million Prime members with credit cards on file and one-click purchasing enabled. That's not a marketplace. That's a captive audience with pre-authorized wallets. Walmart, Shopify, and every other e-commerce platform compete for the remaining attention. Walmart is the rival that keeps Andy Jassy awake. Americans visit Walmart stores 150 million times per week. Each visit is a chance to attach an online order, sign up for Walmart+, or scan a QR code that pulls them into digital commerce. Walmart's 4,700 US stores function as fulfillment nodes that enable same-day delivery without the warehouse construction costs Amazon bears. The pitch is consolidation: you already pay us for Office, Teams, security, and identity management. Adding Azure means one vendor, one bill, one support contract. For a CIO under budget pressure, that's compelling regardless of whether AWS has more services. If enterprises standardize on GPT-4 for internal AI and GPT-4 runs best on Azure, the workload follows the model. Shopify represents the anti-Amazon thesis: merchants who want to own their customer relationship rather than rent it from a marketplace. 200 million behaviorally locked-in Prime members. Jassy spent 2023 cutting: 27,000 corporate roles eliminated, dozens of facilities closed or delayed, the fulfillment network reorganized from a national spaghetti map into eight regional hubs. By FY2024, the results were undeniable. It goes after the exact mechanism that converts marketplace traffic into Amazon's highest-margin revenue. The FTC alleges that Amazon punishes sellers who offer lower prices elsewhere by burying them in search results and stripping Prime eligibility. Structural remedies could force separation of marketplace from retail, restrict how seller data flows between divisions, or limit the bundling of fulfillment with search ranking. Any of those outcomes would hit billions in annual profit. That's not a crisis. It's a slow squeeze. The labor situation is the one that keeps me up at night if I'm an Amazon board member. And unlike AWS margins, you can't engineer your way out of it with better algorithms. It's density. Amazon's per-unit delivery cost drops with every additional package in a given zip code. But the logistics network is the obvious part. That's not a rational calculation — it's a psychological one. Most CTOs look at that equation and decide to stay. Breaking into that loop requires simultaneously offering better selection AND better prices AND faster delivery AND a large enough audience to attract sellers. Nobody has done it. When someone searches on Amazon, they're holding a credit card. Purchase intent at the moment of buying decision is structurally different from informational intent, and it's why Amazon's ad conversion rates justify the premium brands pay. Andy Jassy's Amazon is not Jeff Bezos's Amazon. That's the point. It's the regionalization of the US fulfillment network into eight geographic zones where orders are fulfilled locally instead of shipped cross-country. Boring. Defining. The big bet is AI infrastructure. Custom Trainium2 chips for training. Inferentia2 for inference. Amazon Bedrock as the managed service layer where enterprises access foundation models from Anthropic, Meta, Mistral, and Amazon's own Nova family. Amazon Q as the enterprise AI assistant. It doesn't need to be the flashiest AI platform. It needs to be the most convenient one for existing customers. Amazon has to sell it cold. The advertising trajectory is more certain. Prime Video ads reach 200 million households. Grocery surfaces through Whole Foods and Fresh create physical-world ad inventory. The DSP extends Amazon's purchase-intent data across the open web. Healthcare is the decade bet. But healthcare moves at regulatory speed, not Amazon speed. Three years from now, this is still a work-in-progress. The FTC lawsuit is the wild card nobody can model. Structural remedies that separate marketplace from retail would break the flywheel economics that fund everything else. My judgment: Amazon settles with behavioral concessions that cost money but preserve architecture. Nobody remembers this, but Amazon almost got named Cadabra. As in abracadabra. Jeff Bezos's lawyer talked him out of it because it sounded too much like 'cadaver' over the phone. Bezos was at D. E. Shaw in Manhattan, one of the most secretive and profitable quantitative trading firms on Wall Street, pulling in the kind of compensation that makes people stay forever. Not 23 percent. Twenty-three hundred. He made a list of twenty product categories that could work online and picked books for coldly rational reasons. Three million titles in print. No physical store could stock more than 150,000. An online catalog could offer everything. The product was cheap to ship, impossible to damage, and attracted exactly the kind of educated early-adopter who was already comfortable with the internet in 1994. Here's what I find fascinating about the founding decision: Bezos didn't quit his job because he was passionate about books. He quit because he ran a mental exercise he called the 'regret minimization framework.' At eighty years old, would he regret not trying this? Obviously yes. Would he regret trying and failing? The asymmetry of regret made the decision trivial. His boss David Shaw took him on a walk through Central Park, told him it was a great idea for someone who didn't already have a great job, and wished him well. Bezos and MacKenzie Scott packed a car and drove from New York to Seattle. He chose Seattle for two reasons that had nothing to do with tech culture: a major book distributor (Ingram) had a warehouse in nearby Roseburg, Oregon, and Washington state's small population meant fewer customers would owe sales tax. Within the first week, they'd sold books to customers in all fifty states and forty-five countries. They hit that number in the first year. But the near-death moment came later. The dot-com crash of 2000-2001 cratered the stock from over $100 to under $6. The IPO had happened earlier, May 15, 1997, at $18 per share.