The Watson Weekly is sponsored by Avalara — the agentic AI platform automating global tax and compliance for leading eCommerce brands with simple plans, transparent pricing, and complete coverage across Shopify, Stripe, WooCommerce, BigCommerce, and more.

Amazon's second quarter is two businesses sharing one income statement, and advertising is the one paying for everything else.

The reported figures: net sales of $200.6B, up 20%. Income of $27.5B. Online stores at $70.4B, up 15%. Advertising at $19.8B, up 26%. Trailing twelve month free cash flow came in as an outflow of $7.6B. Sitting above all of it is $53.4B in non-operating pretax income, a meaningful share of which is paper value on Amazon's Anthropic position rather than something Amazon sold to a customer.

The number everyone quoted was the AWS backlog at $496B, with commitments running out to 2028. Backlog is a commitment to buy future compute. It converts to revenue if the companies making those commitments are still funded when the bill arrives. Most of them will be. Some of them won't, and Amazon does not break out which is which.

More curious is what Amazon has decided not to do. It reorganized the AI unit and stepped back from building a frontier model of its own. Read Jassy's comments over the past year and the thesis is consistent: models end up the way programming languages did. Many of them, cheap, chosen by preference, and nobody makes money selling the language itself. Amazon would rather own Trainium, Inferentia, and the capacity everyone rents. It is a coherent position. It also happens to be the position that makes Amazon's weak frontier standing irrelevant, which is worth noticing without treating it as disqualifying.

Grocery came up again, and Amazon is now describing itself as the second largest grocer in the United States. That claim depends entirely on counting everything consumable that moves through amazon.com. The company also says fresh items are six of the top twenty best sellers on the site, which ranks by units, and bananas at a dollar apiece move a lot of units. What actually changed is infrastructure. Amazon says 2,300 delivery locations can now handle perishables. That turns groceries into something you add to an order you were already placing, which is real behavior with real frequency behind it, and a different business from the weekly stock-up Kroger and Walmart are built around. Both things can be true at once.

Then there's the ad unit almost nobody covered. Sponsored prompts. On a product page, that carousel of suggested questions you can tap is becoming inventory. Advertisers set budget and targeting, and Amazon's personalization engine decides which prompts surface. Amazon says shoppers who engage with prompts spend 21% more, though those shoppers were already the closest to buying before they tapped anything.

Making prompts biddable converts a design decision into an auction, and it works better on Amazon than the same idea does inside a general chatbot, because people arrive at Amazon already shopping.

The question Amazon hasn't answered is what the auction optimizes for. A prompt that gets tapped and a prompt that leads to a purchase are not the same product, and advertisers are about to fund both.

Global-e moved $2.09B of GMV last quarter, up 44%, and collected $299M of revenue on it, up 39%. That five-point gap put take rate at 14.3% against 14.8% a year ago. Fulfillment revenue grew 42% while the higher-margin service fee line grew 36%, and non-GAAP gross margin followed to 45.3% from 46.5%.

Passport makes the mix worse. Global-e guides the acquisition to $55 to $59M of second-half revenue on $3 to $4M of adjusted EBITDA, against a parent running at 20.9%. It also supplies most of the raised outlook. Full-year revenue guidance moved up about $80M at the midpoint and roughly $57M of that is Passport.

Adjusted EBITDA of $62.4M grew 62% on 39% revenue growth, and since the measure strips out Shopify commercial agreement amortization in both periods, the leverage is clean. GAAP operating profit is a different matter. It went from $10.5M to $44.7M, and $12.9M of that swing is the Shopify asset finishing its amortization schedule. Amir Schlachet credited AI efficiency gains and did not quantify them.

First-half free cash flow was $271K.